** Disclaimer: I am a VMware employee **
** I am not affiliated with Dell, just picked them as their website is straight forward **
About a year ago I wrote an article about scaling up. I have been receiving multiple requests to update this article as with vRAM many seem to be under the impression that the world has changed but did it really? Yes I know I am about to burn myself but then again I am Dutch and we are known for our bluntness so let me be that Dutch guy again. Now before this turns into a “burn the witch who dares to speak about vRAM” thread let me be clear, this article is not about vRAM per se. Of course I will touch upon it and explain why I don’t think there is a problem in the scenario I am describing, but that is not what this article is about.
In my previous article I discussed the benefits of both scaling up and scaling out. Now as I stated, I had that discussion with customers when hosts were moving towards 32GB per host and now we are moving towards 32GB dimms instead easily cramming 256GB in a host. The world is changing and so is your datacenter, with or without vRAM (there is that word again). Once again I am not going to discuss vRAM by itself as I am not an analyst or responsible for pricing and packaging within VMware but what I do want to discuss is if vRAM has an impact on Scale-out vs Scale-up discussion as some are under the impression it does.
Lets assume the following:
- To virtualize: 300 servers
- Average Mem configured: 3GB
- Average vCPU configured: 1.3
That would be a total of 900GB and 390 vCPUs. Now from a CPU perspective the recommended best practice that VMware PSO has had for the last years has been 5-8 vCPUs per core and we’ll come back to why this is important in second. Lets assume we will use 2U servers for now with different configurations. (When you do the math fiddle around with the RAM/Core/Server ratio, 96 vs 192 vs 256 could make a nice difference!)
- Config 1:
- Dell r710
- 2 x 4 Core – Intel
- 96GB of memory
- $ 5500 per host
- Config 2:
- Dell R810
- 2 x 8 Core – Intel
- 192GB of memory
- $13,000 per host
If we do the quick math that means if we look at it from a memory perspective and assume a roughly 20% TPS benefit (that is very conservative however) and round it up we need 10 x R710s or 5 x R810s. I noticed multiple people making statements about not recommending over-committing on memory because of vRAM, that doesn’t make any sense to me as memory techniques like TPS only lower the overall costs. As mentioned it is recommended to have 5-8 vCPUs per core… Let’s go for 6 vCPUs per core. That means from a vCPU perspective we will need 9 x R710s or 5 x R810s. Now we will take the worst case scenario into account and will go with the larger number for either RAM or CPU. So that results in:
- 10 x Dell R710 = 55k
- 5 x Dell R810 = 65k
Before anyone asks, I also looked at AMD 12-Core systems with 256GB and they come in around 16.5k with 256GB and you would need roughly 4 hosts to accomplish the same looking at the cost of those boxes and comparing it with Intel I would expect a broader adoption of AMD to be honest, but lets focus on the Intel comparison for now. So that is only a 10k difference when looking at hardware but the costs of managing it is lower for the R810s (fewer hosts) and not even talking about I/O ports, cooling and power. (Trying to keep things simple, but when adding these costs the difference will be even bigger.)
So what about that vRAM thingie? What about that huh! Well as I said this is not about vRAM but will it matter when buying large hosts? Well it might, but only when you buy more capacity than you need in this example and want to license all of it before hand… In this case, does it matter? 300 VMs x 3GB vRAM is 900GB vRAM (18.75 licenses for enterprise+), the type of host will not change this or will it? Well actually it will. If you look at the R710 you will need 20 (10 x 2) socket licenses assuming and lets assume Enterprise Plus is used. With the Dell R810 we will need 10 licenses from a socket perspective but 19 from a vRAM perspective using Enterprise Plus.
Lets place it in perspective:
- Scale out
- 20 Enterprise+ licenses required
- 10 Hosts required
- Estimated costs for hosts + licenses 105k
- Scale up
- 19 Enterprise+ licenses required
- 5 Hosts required
- Estimated costs for hosts + licenses 112.5k
Looking at the total costs of acquisition for Scale-Up only in terms of hardware and vSphere licenses in this scenario is indeed slightly more (7.5k) so should you go big?
As mentioned in my other posts there are a couple of things to keep in mind when making this decision and I cannot make it for you unfortunately but there are of course things to factor in. Many of these also have a substantial cost associated with it and I can guarantee that the costs associated with it will more than make up for that 7.5k!
- Cost of Guest Operating System and Applications (licensed per socket in some cases)
- Cost of I/O ports (storage + network)
- Cost of KVM / Rackspace
- Cost of Power / Cooling
- Cost of operating per host (think firmware etc)
- Cost of support (Hardware + Software)
- Total number of VMs
- Total number of vCPUs
- Total number of vRAM
- vCPUs per core ratio
- Redundancy, taking N+1 into account
- Impact of failure
- Impact on DRS (less hosts is less balancing options)
- Impact on TPS (less hosts means more memory sharing means less physical RAM needed)
Now once again I cannot make the call for you, it will depend on what you feel is most important. If you are concerned about placing all eggs in one basket you should probably go for scale out, but if your primary concern is cost and trust your hardware platform, scale up would be the way to go. I guess one thing to consider before you make your decision, how often does a server fail due to a hardware defect vs a human error? Would less servers also imply less chances of human error? But would it also imply a larger impact of a human error?
For those looking for more exact details I would recommend reading this excellent post by Bob Plankers! Bob and I exchanged a lot of DMs and emails on this topic over the last couple of days and I want to thank him for validating my logic and for the tremendous amount of effort he has put in to his article and spreadsheet!. I also want to thank Massimo Re Ferre for proof reading. This article by Aaron Delp is also worth reading, Aaron released it just after I finished this article. Talking about useful articles I would also like to refer to Massimo’s article which was published in 2009 but still very relevant! Scale-Up vs Scale-Out is a hot topic I guess.
Now looking at you guys to chip, and please keep the fight nice and clean, no clinching, spitting or cursing…
Duncan, this post is about scale-up vs scale-out but it will turn quickly (I anticipate) into a vRAM debate. That’s ok.
For the sake of the discussion I’d point out that the vRAM discussion has nothing to do with scale-out vs scale up. The vRAM licensing discussion is all about the Socket to Cores/RAM ratio. Whether you distribute those Sockets/Cores/RAM on many small servers or fewer bigger servers doesn’t really matter.
My 2 years old article you linked (which I wrote when I WASN’T working for VMware) shows clearly the trend and could be summarized with the following two trends in the industry:
1/2 x Sockets, 1 x Cores, 1 x RAM, 1 x VMs
or
1 x Socket, 2 x Cores, 2 x RAM, 2 x VMs
There is a misconception (IMO) that we are raising the price. Effectively we are trying to keep it equal (although I can understand there is a always a desire to pay less – isn’t that a human behavior?).
Either way, if we want to keep the price-per-workload equal going forward we needed to tie it to another subsystem rather than the socket.
Note I used price-per-workload and not price-per-VM on purpose as there is a need to create a mechanism to charge more for a 8 vCPU 256GB VM than for a 1 vCPU 4GB VM.
Massimo.
P.S: as Duncan, I am not talking for VMware here… just my own thoughts of the whole thing.
I have to say that, until you pointed it out, I’d been thinking that all the TPS and overcommitment technology was wasted now with vSphere 5. Not so — you can still cut down on your physical RAM by relying on it, and that saves money, too.
Correct me if I’m wrong, but the main comment we’ve had on Twitter was just about “too many eggs in one basket.” I think if people can overcome that fear, and think critically about how often that really is a problem, more will be won over to the “scale up” side of things.
Thanks for the tag-team, Duncan! It’s been fun, and it’s great to have your perspective!
As Massimo stated, “this post is about scale-up vs scale-out but it will turn quickly (I anticipate) into a vRAM debate.” Unfortunately I’m going to be the first one to post against vRAM.
I don’t think the problem is so much that VMware is changing their licensing policy as it is that they are “forcing” existing customers to purchase additional licenses for existing implementations.
If I’m building a brand new environment from scratch and VMware is my chosen platform, then I have no choice but to accept their new licensing model. In this case however, I have the opportunity to weigh scale-up vs scale-out and design my environment to suite my needs.
On the other hand, if I’m an existing customer with a $1M plus investment and I already have a scaled-up environment, I’m going to see a hugh increase my licensing cost in order to maintain support.
As I’ve said on the VMware forums, anyone who has previously invested in dual socket servers packed full of RAM (128GB+) are in a very bad position when they decide to upgrade to vSphere 5.
Jason
@Jason,
>As I’ve said on the VMware forums, anyone who has previously invested >in dual socket servers packed full of RAM (128GB+) are in a very bad >position when they decide to upgrade to vSphere 5.
My (personal) theory is that those customers have saved money on vSphere licenses in doing so (if what’s in my old post Duncan linked makes sense). The intent is to try to keep the “cost per workload” constant rather than increasing or decreasing. Of course there will be corner cases (how big?) this is not the case.
However if you feel very penalized because of the way you have layed out your current environment reach out your TAM/salesrep. I believe this is an EPIC change (but whose idea is to leave the bottom line unchanged). If that’s not the case reach out to them.
Massimo.
While I understand that you wanted to address the UP vs. OUT debate, I’m gonna soapbox …
(note: I’m coming from a large enterprise, 4000+ VM context)
1.) In vSphere 4, I needed 2 host licenses for my 2 socket, 192 GB host. In vSphere 5, I need 2 licenses. VMware just doubled my licensing cost per host.
2.) In vSphere 4, due to the flexibility of the licensing, I was game to virtualization *large* workloads. Now, more than ever, I am going to have to think that through, and may opt not to virtualize certain larger-memory consumers. Effectively, I’m being encouraged to *not* virtualize everywhere. When faced w/ the choice of reusing older physical servers w/ a little add’l memory, or buying new VMware licenses, I’m going to have a harder case to make for the latter.
3.) In vSphere 4, while we worked w/ customers to right-size VMs, at the end of the day we picked-and-chose our political and technical battles and were willing to allocate memory more freely and leverage ‘real’ memory consumption and vSphere’s memory saving techniques to help keep the balance. Now there’s a new cost to do that. You’ve complicated the political landscape and made virtualization less easy. Talk to most enterprises and you’ll see they lack a true chargeback and consumption based financial model … there’s little real incentive for my customers to lower their memory usage — it’s a battle that the VM admin often can’t control … so now I’ll have to fight. This runs opposite to everything I’ve been promoting over the past 5 years in talking about how flexible virtualization is.
At the end of the day, the vRam pricing threatens to affect a real mindset and cultural shift. The 3 128 GB VMs I was ask to create last week now have ~30k in VMware licensing impact, whereas before I could largely swallow that as doable hardware and resource management challenge.
@Geekinabox,
>1) In vSphere 4, I needed 2 host licenses for my 2 socket, 192 GB host. >In vSphere 5, I need 2 licenses. VMware just doubled my licensing cost >per host.
Is it fair to say that 2 socket server server with 96 GB of memory can support half the VMs you can support with a 192GB host? And if that’s the case aren’t we just keeping the “price-per-workload” equal?
>The 3 128 GB VMs I was ask to create last week now have ~30k in >VMware licensing impact, whereas before I could largely swallow that as >doable hardware and resource management challenge.
That’s the exact thing I was mentioning in my last paragraph (see first comment of this post). Imagine there was no virtualization and you had to give your end-users physical boxes. Wouldn’t you charge them differently if they ask a 1CPU-2GB desktop or a 8-CPU 256GB HE server? I completely agree that you may not have the right culture in place today but isn’t it sound to start thinking about it? On the other hand what other options do your users have? Sure they can try to by-pass you … and pay a heck of money to the Public Cloud provider for a 128GB VM.
>>Is it fair to say that 2 socket server server with 96 GB of memory can support half the VMs you can support with a 192GB host? And if that’s the case aren’t we just keeping the “price-per-workload” equal?
Unless I am missing the point of the original poster, I think the point is many VMware customers have already purchased the bigger box in vSphere4 and are counting upon that scalability going forward. Being an Enterprise (not +) customer, it is made worse in my example. If the OP or others customers like myself were evaluating the additional upgrading hardware then your point may be valid. We have however implemented the 144 or 192GB hosts and are now being asked by VMware to pay them again for the same hosts if we choose to update. Now like Microsoft Vista and XP, we may choose to wait it out and see if something is released to warrant this additional outlay.
So perhaps for the new customer purchasing 9 socket licenses for CPUs you don’t own in your environment is just the price of VMware, but that doesn’t help the existing admins with large memory hosts with few sockets trying to have the conversation about needing more licenses to license processors they never purchased.
@Kent,
fair enough. I totally understand the issue. I think that the problem is that if you were not taking advantage of the memory density we are talking about…you would have to pay for more sockets/servers. So what happened (in a way) is that you saved a certain percentage of vSphere licenses (because of the higher density) and we are now trying to bring that on track charging differently (on track as “before”, not “more”).
Now, while I think this is a fair think to do (your mileage may vary), for people that were caught in the middle of this transition it appears as a “vTAX”. Not because they are paying more but because they are not able to save on licenses what they have budgeted.
Does this make any sense to you?
Massimo.
@Massimo
Are you getting some kind of commission on this or do you have stock in VMware? This is a bad path VMware has chosen and they just lost a customer. We currently have over 100 hosts and I promise if they follow through with this MS BS I will go with XenServer. This is straight up greed. We do not do the cloud thing that everybody seems to be a fan of these days. We provide servers for our company, which means we try to save money. We went with the scale up scenario. It made the best business sense for us to save money on the licenses. Now VMware wants to hold us up in a back alley and take all our money.
It’s odd that you should talk about Scale Up and spec out the R710 the way you did. We use R710’s to scale up, and they can support (at some point) 16GBx18DIMM and currently support 6 cores which is how we would order them.
So in our office we would buy the R710 with 2×6 Core and 144GB RAM (sweet spot for RAM).
So now 300 servers at 1.3 vCPU each is 390 vCPU. At 6:1 that’s 65 cores worth, or 5.4 hosts. You obviously have to round that up to 6, and I’m going to assume the 6:1 vCPU:cores already accounts for N+1 sizing.
300 * 3GB vRAM is 6.25 hosts worth at 144GB. You could over commit but that’s probably not best practice, so based on vRAM we’ll need 7 hosts at least. However with higher density VM counts on a host, the overcommit starts working better as the likelihood of pages being shared is higher.
This means 14 sockets of Enterprise Plus vs 20, or $42K vs $60K. That 18K savings can certainly go towards the extra density cost of the higher spec’ed R710’s. Let’s assume about $8500 for the R710 spec’ed this way, as that’s my experience so 7*8500=$60K So total is about $60K for hardware and $42K for VMware for $102K
Except one problem. I’m only licenced for 672GB of vRAM. (14x 48GB). I *can* run this today on vSphere 4 with 14 licences. To do it with vSphere 5, I need 5 more Enterprise Plus licences, which bumps my cost by 5*3K=$15K.
That, sucks. It’s an extra 14% VMware Tax is what it is.
@Avram,
not sure I follow you (perhaps I am wrong, sorry in a hurry atm) but would that be 14% more compared to how you would buy today 7 x R710 hosts? But today, based on Duncan’s math, you would need to buy 10 x R710 hosts. Isnt’ the cost of 7 x hosts +14% circa the cost of 10 x hosts? If it is we are keeping the “cost per workload equal”.
Massimo.
I spec’ed the R710 out this way as 2 x 4core with 96G is probably one of the most ordered configurations and has been for the last months as far as I am aware.
I agree with massimo that you will need to do an end-to-end comparison to understand what the cost per workload is of this configuration. So one one hand you might be paying more for vSphere licenses on the other hand you are saving costs around operating the environment, guest OS licenses, I/O, cooling etc/
Hardly had user order quad cores for esx host now, most of it at least come with 6 cores per CPU
Duncan’s math is all good and fine, and astutely shows that scale up costs pretty close to scale out – but under vSphere 5 only, not comparing v4 to v5.
This is really the root of the issue that people are fundamentally p*ssed off about, how a scale-up setup in v5 costs a lot more than it did in v4. Hence, customers who scaled up are now having to pay out the nose to upgrade to v5. In Duncan’s 5x R810 example, it requires 19 vSphere licenses, whereas previously only requiring 10. The Delta of 9 is the core issue here. Under v4, the R710 example would cost the same, but the R810 example would cost roughly $25K less! VMware essentially took away any significant up-front cost savings of scale-up.
I don’t really think it matters *how* VMware licenses its products (vRAM, pRAM, sockets, cores, threads, color of server, whatever). What matters is that bigger customers who are paying full SnS are having to pay up the nose to upgrade many of their scaled-up datacenters. If they grandfathered existing customers (or provided heavy discounts), much of this debate wouldn’t even be happening. People would simply be griping about how much *more* expensive the cost of entry into the VMware universe is.
Luckily, I’m not personally affected by the v5 licensing in my environment (we’re a fairly small SMB (40 VMs, with vRAM to spare), but I still certainly disagree with VMware’s direction on this issue, particularly as it pertains to existing, scaled-up customers.
@Scott,
forget about all Duncan’s math.
The misunderstanding here is that scale out vs scale up was always defined as (for example) 4 small 2 way servers (8 sockets) VS 2 bigger 4 way servers (8 sockets). In that situation the VMware “vTAX” remains the same and everybody is happy.
I am puzzled how people misses to understand the fundamental issue here which is: you are defining scale out vs scale up as 2 NONDENSE 2-way servers (4 sockets) VS 2 DENSE 2 way servers (4 sockets).
How can we think it’s fair to charge, for the same amount of money, (4 sockets) a solution that can support twice as many VMs?
People focus on the fact that we WILL be charging MORE for the dense server. I focus more on the fact that we ARE charging LESS for it (and that is the distortion of the current model IMO).
Massimo.
Massimo, your reasoning and probably VMware’s as well is that vRAM licensing keeps price per workload constant and that is fair. But is it really? Workloads are growing, new editions of Exchange server consume more RAM, SAP, Oracle, etc. grows with time. In all physical world it was not a problem, because you could buy new more powerful hardware for the price of old thanks to Moore’s Law. Commoditized x86 hardware where big vendors, and chip producers Intel and AMD are trying to innovate without having a monopoly keeps prices down. So you just swapped your Exchange 2003 boxes with new ones with Exchange 2010 at the same cost. That is not the case with hypervisors. Having no real competition VMware is able to innovate while driving the price up. It is perfectly understandable in the capitalistic market but do not expect any sympathy for it.
We understand your frustration but when you look at your comparison and think about it something is not entirely correct. When Intel or AMD release new chips these are always more expensive then their current offerings. When Microsoft introduced Exchange 2003 they also introduced a change in licensing. Again we understand your situation and all I can say at this point is address your concerns with data to your VMware representative.
“When Intel or AMD release new chips these are always more expensive then their current offerings.”
I don’t agree with that. Yes, the current chip models have been discounted from their initial pricing, but from what I’ve seen is that Intel and AMD try to come in around the same price as the previous chip’s *initial selling price*, but with more features and performance.
But to follow your and Massimo’s arguments to their logical conclusion, VMware should also start charging more for faster CPU’s. Maybe some multiplier based on the GHz of the processor. They could come up with a clever name, like, oh I don’t know, maybe “Power Units”
One of the biggest problems I have is that VMware has tried to tell us that this was done to benefit the customer. As one poster in another forum suggested, let the customer choose between a V4 and a V5 model and let the customer decide what is best. What do you think most of them would choose?
@DW,
charging for CPU (cores / GHZ, sockets etc) doesn’t make any sense today since it’s not the “limiting factor” and the subsystem around which you are sizing.
>As one poster in another forum suggested, let the customer choose >between a V4 and a V5 model and let the customer decide what is best. >What do you think most of them would choose?
Yesterday (V4) you were paying VMware 10 for a 4socket-96GB server to support 30 VMs. Today (V4) you are paying VMware 5 for a 2socket-96GB server to support 30 VMs.
Do you really need to ask that question? It’s like asking you do you want to pay that car its regular price or half the price? You’d be dumb if you were to say “full price”. Yet that’s the value of the car and the car vendor should sell it for its price (I am talking about a markatable price, not stealing money). If you think we are stealing money from your pockets you have a right to do so but it has nothing to do with per socket or vRAM based pricing.
That’s at least how I see it.
Massimo.
@Tomas,
this is a good discussion to have. You are making good points. While I don’t agree across the board it’s a legitimate/debatable topic.
**
Workloads are growing, new editions of Exchange server consume more RAM, SAP, Oracle, etc. grows with time. In all physical world it was not a problem, because you could buy new more powerful hardware for the price of old thanks to Moore’s Law. Commoditized x86 hardware where big vendors, and chip producers Intel and AMD are trying to innovate without having a monopoly keeps prices down. So you just swapped your Exchange 2003 boxes with new ones with Exchange 2010 at the same cost. That is not the case with hypervisors.
***
I have to disagree with that however. HW resources per Servers are outpacing SW requirements per App by (at least) an order of magnitude higher and the reason behind using virtualization. I used to install Exchange on a 2CPU (2 cores) server with 4GB of memory. Sure you can install Exchange on a 4 socket (24 cores) server with 512GB of memory today (and use it at 0.3% of its capabilities). If it wasn’t for virtualization these HW/CPU vendors would be selling 1 CPU servers (with 4 cores enabled out of 12 available) with one DIMM slot to install an 8/16GB DIMM.
Also, I remember when I started with ESX 1.1 we used to deploy VMs that had around 1 or 2 GB of memory. I see Duncan is using 3GB of memory per VM (average). No one is arguing with that (average). Sure you may have a 128GB VM to deploy, but how many 2GB “regular” VMs are deploying for each of the 128GB VM?
Having this said the 1CPU-48GB (Ent+) number is not set in stone. That’s a metric that can be adapted as these SW requirements grow. We would have been fool to allow for a 1CPU-4GB entitlement, but we would have been as fool to allow for a 1CPU-128GB entitlement. Someone (not me) thought that 48GB per socket is the right ratio to not penalize the vast majority of the customers. There are of course leading edge customers out there that have already started to lower their cost-per-workload (using higher density servers) and have a right to have the feeling of being treated badly.
>Having no real competition VMware is able to innovate while driving >the price up. It is perfectly understandable in the capitalistic market >but do not expect any sympathy for it.
I found this quite funny:
http://www.thebiggertruth.com/2011/07/summer-love-and-software-licensing/
Massimo.
@MASSIMO
I see your point, about having 2+2 vs 4, in that it’s the same processor (and license) count. But that’s also assuming that you’re getting the same processors across the board, which often isn’t the case. Part of the strategy with scale-up is buying fewer, higher-core faster clock processors. You can buy Xeon’s today that range from 2 to 8 cores, varying wildly in features and clock speeds (and opterons up to what? 12 cores now?). The fundamental problem – two dual processor systems do not equal one quad system.
VMware obviously realizes this, and has clearly chosen to roll out a licensing scheme that allows them to profit from this. Moore’s law hurts per processor VM licensing, simply because processor growth won’t track virtual growth very well.
@Scott,
I think your analysis is sound. It’s in fact what I was arguing. However I am not sure I agree with the conclusion when you say: “VMware obviously realizes this, and has clearly chosen to roll out a licensing scheme that allows them to profit from this”
Shouldn’t this be more like “…a licensing scheme that allows them to keep the “cost for workload equal”
?
Massimo.
Hosts 16
Sockets 2
Cores 6
RAM 128
Enterprise vSphere 4 = 32 Licenses = $92K
same config above upgraded to vSphere 5 = 64 Licenses = 184K or Enterprise Plus = 150K
@vRAM$$$$
I think you have been caught in the middle of the transition. Of course your very actual situation makes you a good candidate for feeling very angry. Talk to you sales rep about this.
Having this said, you are pointing to the tree. Try to look at the forest.
Yesterday you were probably doing:
Hosts 16
Sockets 4
Cores 4
RAM 64GB
or
Hosts 32
Sockets 2
Cores 2
RAM 64GB
Try to do the licensing math on that.
And what if tomorrow you were doing?
Hosts 8
Sockets 2
Cores 12
RAM 256
Try the math again.
This is the “forest” problem we are trying to address with the vRAM licensing. I have written about this trend 2 years ago (see link in Duncan’s post) when I was working for a HW vendor (not VMware).
Massimo.
On top of that, when scaling up keep in mind all the associated costs that will go down like “opex”, “i/o ports”, “cooling & power” (slightly), rack space, Guest OS licenses, there is a lot to it.
When’s your “yesterday”? Looks like 3-4 years ago and not last year to me. Quite weird logic similar to VMwares arguments about 5:1 ratios etc.
Presuming that a lot of customers of VMware are under-utilizing the license they purchased by doing low ratios (they might be tiny SMBs with only 4-5 VMs on 2 hosts…) it doesn’t give VMware any more right to suddently change the licensing for those who were actually utilizing the licensing already in place.
I’m talking about 4.1 Advanced or Enterprise Plus which allowed 12 cores per CPU. Why do you keep calculating with 4 cores per CPU when the existing 4.1 licensing clearly was made for 3x that amount I wonder?
So… how much memory would you feel is normal to allocate per core with an average consolidation ratio? Let’s go really low on that number to be kind: 8GB per core. That’s 8×12 96GB GB per CPU with old licensing so with old 4.1 Advanced/Enterprise Plus licensing I’d need 2 licenses for a 24-core server with 192GB RAM.
Now I need 4 enterprise plus or 6 enterprise licenses to run the same server. If I had bought 2x vSphere 4.1 Advanced licenses for each of the servers mentioned above it would have cost me 4490 per server.
Upgrading that server to vSphere 5 Enterprise would cost me another 4 licenses assuming that I’m doing a mild oversubscription of RAM etc.
I don’t question that some might have bought servers several years ago with the specs you mention. What I question is the change of licensing based on underutilized old usage numbers.
Would be interesting if we were to see Bugatti announce that they will limit their Veyrons (including all existing ones already sold to customers next time they are in for scheduled service) to 40MpH because that’s the average speed their customers are driving them at (city traffic etc taken into account)? To get more speed their customers would need to purchase license packs for each 10MpH priced at $50k…
My business case:
Hosts 32 (4 clusters of 8 hosts)
Sockets 4
Cores 8
RAM 512
no memory overcommitment
2.000 VMs
vSphere 4 = 128 Licenses Enterprise+ (32 * 4)
same config above upgraded to vSphere 5 = 342 Licenses Enterprise+ (32 * 512 / 48)
==> increase of 267% on license cost!
How can i convince my management to migrate to vSphere 5?
What about all benefits of VMware memory allocation (balloning, TPS, memory compression), they are all gone due to vRAM entitlements…
What will be next, vCPU entitlements?
Enterprises that are happy to fix their cost per workload at the levels of a couple of years ago will do the calculations above – as you say, there isn’t much in it. The rest will scale up as fast as they can, to postpone the upgrade to lower RAM allocations. That buys time to look for economically viable solutions for memory-intensive applications like SQL and Exchange.
I imagine someone at VMware will be keeping a close eye on hardware sales over the next month.
@Michael,
there is more than an eye looking at this, that I can say. We want this to work out, trust me.
Massimo.
Can I thank Massimo for responding to these comments faster than I can read? Thanks M.
I’ll be on vacation next Monday. So you’ll have time to read them 😉
I am not sure why this licensing thing caught my attention. I was supposed to be a techie once…
Massimo.
>> Scale-Up vs Scale-Out is a hot topic I guess.
Scale-Up or Scale-Out does NOT matter if you want to run VMs with a lot of memory on vSphere 5.
For 1 VM with (32 vCPU) and 1TB RAM you will need to buy 21 (twenty one) vSphere5 Enterprise Plus licenses !!
If you scale-out to 21 VMs with 48GB, you still need to buy 21 licenses…
This VM can be hosted on 1x vSphere DL580 G7 4CPU 8Cores X7550 1TB RAM -> 4CPUs = 4 Enterprise Plus licenses
With “vRAM entitlements”: -> 1TB : 48GB per license = 21 licenses -> Price: 21 * €3.078 = €64.638 = €64k for 1 VM
from 4 to 21 = increase of 525% on license cost!!
What do you suggest in this case?
I would recommend VMware to set the vRAM entitlement for Enterprise+ licenses to UNLIMITED.
If your VM really requires 32vCPUs and 1TB of RAM than there is nothing we can suggest. However that wasn’t possible in vSphere 4.1 and as such not a fair comparison in my opinion although I do fully understand your frustration.
I will personally feedback all comments on this article to our licensing team. I can of course cannot make any promises but I can address your concerns.
@DanPan
Although I appreciate that changes always introduce pathos, I’ll take a less conciliant position than Duncan on this. And let me “challenge” you on this (big word I am using – more like… let’s try to think together about this). Ok I’ll be the bad guy.. the one everyone will hate.
>I would recommend VMware to set the vRAM entitlement for Enterprise+ >licenses to UNLIMITED.
If we did that here it is what would happen next year:
2S (24 cores) – I believe AMD has pre-announced something like this –
1TB of memory
While you can’t really create the 32 vCPU/1TB VM on that beast.. you’d get close. You can however create a heck of a number of “regular” VMs. Fair to say that, at a MINIMUM, you can support as many VMs as you could create on a current 4S (24 cores) with 512GB of memory. Probably more given you are NOT bound by cores today.
So you’d pay today 4 x Ent CPU licenses and tomorrow (to support more VMs) 2 x Enterprise CPU licenses.
Question: Do you consider this “fair”?
Not only that. Let’s say that that brand new cool server will cost (more or less) something around (north of) 20/30K$
You are paying circa 6K$ to VMware and north of 20/30K$ for the HW.
Question: Do you consider that representative of the VALUE delivered for your IT? Honest.
People are using the 1TB monster VM to make big noise about this licensing but the reality is that that 1TB VM will either:
1- come from a UNIX partition (where it would have costed “slightly” more than 65K$)
2- be on a physical inflexible deployment that couldn’t take advantage of all the virtualization advantages (not going to list theme here)
3- no matter what that 32vCPU/1TB VM could be running a SW whose license cost would dwarf the 65K$ we are asking for it.
Of course you can run a 1TB VM today for 3K$ (well ok ok you can’t with V4 ok …) but that doesn’t mean it is a fair charge for the value delivered and for everything I said above.
Massimo.
The 1TB debate is hilarious, just because of the irony. Here in reality, if you’re anywhere close to needing the likes of a 1TB VM, then almost guaranteed the project is so large the extra $125K in VMware licensing is peanuts. If you’re buying a Ferarri, I’d wager a guess that you’re probably not loosing any sleep over a $2000 delivery charge.
@Scott
+1
Just want to start by saying great post and thanks!
@Scott
The Ferrari example is not a business example but more of a personal. If I can afford a Ferrari then I shouldnt scoff at a $2K delivery charge? I agree for you or I personally.
Unfortunately CFOs and CIOs dont look at cost like that. Can you picture yourself going into the CFO or CIOs office and telling them that you used to pay $75K for the VMWare license and now we are going to pay $125K can you see any CFO saying “No big deal , if we are spending a million dollars whats 50K more?”.
Jason, absolutely agree. My Ferarri comment was only pertaining to a 1TB VM. I’m going to wager a guess that if you’re truly needing a 1TB VM, then it’s likely at least a 10 or 20 million dollar project. Let me perhaps put it this way: If you’re building a $50M oracle cluster that’s part of a 250 million dollar data center, $50K extra in VMware licensing is a pretty small line item on the grand scheme of things. In other words – the government/military mentality.
But for day to day IT operations, a 50%-200% increases in VMware licensing, especially when it’s simply an upgrade, is a BIG deal. Especially if you’re already forking over for SnS. Completely agree there.
OK. Let’s do it this way:
—————————–
Host 1 (HP DL580G7)
Sockets 4
Cores 8
RAM 1TB
no memory/CPU overcommitment
4x VM 8vCPU/256GB
vSphere 4 = 4 Licenses Enterprise+ (4 * 1)
same config above upgraded to vSphere 5 = 21 Licenses Enterprise+ (4 * 256 / 48)
==> from 4 to 21 = increase of 525% on license cost!!
—————————–
Host 1 (HP DL580G7)
Sockets 4
Cores 8
RAM 1TB
no memory/CPU overcommitment
8x VM 4vCPU/128GB
vSphere 4 = 4 Licenses Enterprise+ (4 * 1)
same config above upgraded to vSphere 5 = 21 Licenses Enterprise+ (8 * 128 / 48)
==> from 4 to 21 = increase of 525% on license cost!!
—————————–
Host 1 (HP DL580G7)
Sockets 4
Cores 8
RAM 1TB
no memory/CPU overcommitment
16x VM 2vCPU/64GB
vSphere 4 = 4 Licenses Enterprise+ (4 * 1)
same config above upgraded to vSphere 5 = 21 Licenses Enterprise+ (16 * 64 / 48)
==> from 4 to 21 = increase of 525% on license cost!!
—————————–
Host 1 (HP DL580G7)
Sockets 4
Cores 8
RAM 1TB
no memory/CPU overcommitment
32x VM 1vCPU/32GB
vSphere 4 = 4 Licenses Enterprise+ (4 * 1)
same config above upgraded to vSphere 5 = 21 Licenses Enterprise+ (32 * 32 / 48)
==> from 4 to 21 = increase of 525% on license cost!!
—————————–
Host 1 (HP DL580G7)
Sockets 4
Cores 8
RAM 512GB
no memory/CPU overcommitment
4x VM 8vCPU/128GB
vSphere 4 = 4 Licenses Enterprise+ (4 * 1)
same config above upgraded to vSphere 5 = 11 Licenses Enterprise+ (4 * 128 / 48)
==> from 4 to 11 = increase of 275% on license cost!!
—————————–
Host 1 (HP DL580G7)
Sockets 4
Cores 8
RAM 512GB
no memory/CPU overcommitment
32x VM 1vCPU/16GB
vSphere 4 = 4 Licenses Enterprise+ (4 * 1)
same config above upgraded to vSphere 5 = 11 Licenses Enterprise+ (32 * 16 / 48)
==> from 4 to 11 = increase of 275% on license cost!!
—————————–
Host 1 (HP DL580G7)
Sockets 4
Cores 8
RAM 256GB
no memory/CPU overcommitment
32x VM 1vCPU/8GB
vSphere 4 = 4 Licenses Enterprise+ (4 * 1)
same config above upgraded to vSphere 5 = 6 Licenses Enterprise+ (32 * 8 / 48)
==> from 4 to 6 = increase of 150% on license cost!!
—————————–
Host 1 (HP DL580G7)
Sockets 4
Cores 8
RAM 256GB
no memory/CPU overcommitment
16x VM 2vCPU/16GB
vSphere 4 = 4 Licenses Enterprise+ (4 * 1)
same config above upgraded to vSphere 5 = 6 Licenses Enterprise+ (16 * 16 / 48)
==> from 4 to 6 = increase of 150% on license cost!!
—————————–
Host 1 (HP DL580G7)
Sockets 2
Cores 8
RAM 128GB
no memory/CPU overcommitment
8x VM 2vCPU/16GB
vSphere 4 = 2 Licenses Enterprise+ (2 * 1)
same config above upgraded to vSphere 5 = 6 Licenses Enterprise+ (8 * 16 / 48)
==> from 2 to 6 = increase of 150% on license cost!!
—————————–
Host 1 (HP DL580G7)
Sockets 2
Cores 8
RAM 128GB
no memory/CPU overcommitment
16x VM 1vCPU/8GB
vSphere 4 = 2 Licenses Enterprise+ (2 * 1)
same config above upgraded to vSphere 5 = 6 Licenses Enterprise+ (16 * 8 / 48)
==> from 2 to 6 = increase of 150% on license cost!!
—————————–
Please show me how i can migrate these realistic workloads to vSphere 5 with an increase on license cost LESS than 150%.
You can not… Checkmate.
Why: VMware based there vRAM entitlement on a ratio 5:1 and max. 48GB / phycical CPU, which is NOT the reality!
corrections:
—————————–
Host 1 (HP DL580G7)
Sockets 2
Cores 8
RAM 128GB
no memory/CPU overcommitment
8x VM 2vCPU/16GB
vSphere 4 = 2 Licenses Enterprise+ (2 * 1)
same config above upgraded to vSphere 5 = 3 Licenses Enterprise+ (8 * 16 / 48)
==> from 2 to 3 = increase of 150% on license cost!!
—————————–
Host 1 (HP DL580G7)
Sockets 2
Cores 8
RAM 128GB
no memory/CPU overcommitment
16x VM 1vCPU/8GB
vSphere 4 = 2 Licenses Enterprise+ (2 * 1)
same config above upgraded to vSphere 5 = 3 Licenses Enterprise+ (16 * 8 / 48)
==> from 2 to 3 = increase of 150% on license cost!!
—————————–
What all your scenarios have in common is that they are exceptional examples. I’ve seen literally hundreds of environments over the years and 10.000s VMs and no environment came close to any of the scenarios you list.
I can spec out servers for you as well:
Host 1 (R815)
Sockets 4
Cores 48
RAM 256
No overCommit
100 1vCPU / 2GB RAM VMs
vSphere 4: 8 Enterprise Licenses (limited to 6 cores)
vSphere 5: 8 Enterprise Licenses (no core limit)
you can use the math anyway you like as long as you tweak the numbers correctly.
@Duncan,
I invite you to have a look at our environment @ GDF SUEZ.
Hosts 32 (4 clusters of 8 hosts)
HP DL580G7 X7750
Sockets 4
Cores 8
RAM 256
By end of this year the new hosts will be HP DL580G7 4CPU 10Cores E7-4850 512GB
Thanks, but I know sites much larger with more extreme numbers. Fact is that most sites don’t have these types of configurations. Like I said you can tweak the numbers anyway you like and that was not the point of my article.
The point is that you will need to take the total amount of configured vRAM and play with Bob’s calculator to figure out what the most optimal hardware config is in terms of TCA, Management, Cooling etc.
@ Duncan/Massimo,
I quit this discussion because it makes NO sense to continue with people who are not willing to understand or brainwashed by VMware…
A VERY disappointed VMware customer!
[end discussion]
Folks I quit this discussion because we continue to repeat the same things…
>Point is vmware knows most important thing for vms is RAM than >CPU that’s why they changed their licensing. NO?
Hell yes that is the reason and it’s my point!
We can continue for months discussing this…
Massimo.
correction:
—————————–
Host 1 (HP DL580G7)
Sockets 2
Cores 8
RAM 128GB
no memory/CPU overcommitment
8x VM 2vCPU/16GB
vSphere 4 = 2 Licenses Enterprise+ (2 * 1)
same config above upgraded to vSphere 5 = 3 Licenses Enterprise+ (8 * 16 / 48)
==> from 2 to 3 = increase of 150% on license cost!!
—————————–
Host 1 (HP DL580G7)
Sockets 2
Cores 8
RAM 128GB
no memory/CPU overcommitment
16x VM 1vCPU/8GB
vSphere 4 = 2 Licenses Enterprise+ (2 * 1)
same config above upgraded to vSphere 5 = 3 Licenses Enterprise+ (16 * 8 / 48)
==> from 2 to 3 = increase of 150% on license cost!!
—————————–
Uhhh, 2 licenses to 3 is not a 150% increase, it’s a 50% increase. 2 * 1.5 = 3.
2 licenses increased by 100% = 4 licenses.
2 licenses increased by 150% = 5 licenses.
(not saying a 50% increase in licensing costs is small or trivial, though).
The point is, no way you need so much CPU/Core Power. Most common scenarios needs RAM.
2 Sockets with 6 Cores running 2Ghz+ is more than enough per Server and this is where the vSphere 5 shows its crazy extra $$$$…
No need for 4 Socket and 10 Core Servers. What are you running that you need so much CPU Power?
Most people here will tell you that typically you have a 64GB RAM Server with 2 Socket Quad or Six Cores plus most CPU have HT enabled as well. I am running about 35vms per Server with 2 Socket Quad Cores, RAM Usage is at about 90% while CPU usage is around 25%. So if I want to add more ram to make the Server more dense I will buy extra ram lets say 32-64GB additional an my CPU usage should be at around 50-60%.
Point is vmware knows most important thing for vms is RAM than CPU that’s why they changed their licensing. NO?
>> What are you running that you need so much CPU Power?
We went with the scale-up scenario, which WAS the best choice to save money on licenses and infrastructure on vSphere 4.
Scale-up means higher concolidation ratio (80:1 per host) (theoretical 114:1).
Scheduled backup and virus scanning processes in the VM need a lot of host CPU resources, if you can do them only during the weekend or after business hours.
>We went with the scale-up scenario, which WAS the best choice >to save money on licenses and infrastructure on vSphere 4.
So is vsphere 5 more expensive or it doesn’t just allow you to “save money on licenses”? 🙂
It’s a subdle (yet) important difference and my whole point in this thread.
Massimo.
Lets say u going to buy a new car. The sales person of brand X tells that the maximum speed is limited to 25mph. Are u really interested in how that limit is enforced? Probably not, the question that u really wondering about is what does it cost to remove that limit and is total price still a good offer or should i go for another brand?
By far the most customers of the company i work for arent interested in buying hard- and software they want to run their business and to do that they need a ICT infrastructure. They want to know how much it costs to keep their business running and expanding, the main reason why we specify what they getting is for legal reasons. Most of the time they just flip to the right page to check the figure on the bottom row. Exactly this is main reason why SAAS and cloudcomputing is taking such a big flight.
Back to your subdle difference is it important why a customer cant get the same mileage out of vSphere 5 as out of vSphere 4? In my opinion not, because the customer still has to pay it.
I seriously disagree with this. The amount of functionality that vSphere 5 offers with features like VMFS-5, SDRS, VASA, Profile Driven Storage, new HA, enhanced DRS/DPM, enhanced scalability and much more by itself will ensure you will get the same mileage at a bare minimum if not more due to the decrease of operational effort associated with managing a virtual infrastructure.
Yes, those are nice features but our customers dont have enterprise (plus) licenses and therefore have no access to most of them.
@halchris and vRAM$$$$ (is that your real name? Wow!),
We hear what you are saying. Competitors giving away products for free shows their weakness not ours (IMO).
Of course we would be fool if we were to charge you a tax on top of the value we deliver on the other hand we also need to allow our 4000+ (and growing) engineers to pay their mortgage. I don’t run the revenue/costs/profits spreadsheet at VMWare but revenue doesn’t only allow us to deliver higher profits but also to cover costs (to deliver better products).
Is this vRAM schema perfect. Far from it! Is it the best way to position us and IT to enter into a different era? I think so. Not to increase the revenue with a “tax” but to keep it aligned with value delivered.
We know there is competition out there and we would be fool to put a tax on our products to give you an excuse to go there.
Is 48GB of memory per socket right? I dont know. What Ican guarantee is that there are a lot of people looking into this
Scale out and Scale up, I think you may want to seriously look into the application design. Infrastructure scale up or scale out, will not decide the redundancy or performance of most of the application today. The same topic been discussed in VMTN had some very good example, they can even scale out or scale up by the application itself. VMware do not understand enough for the application to force the users to accept scale out are the best way to be approach, unless VMware claim they understand all different application in the IT world.
Well, I agree with many readers here, to make more money is fine, but it need to be fair to the users and tie back to the use case. When no more business case, thing will change. If VMware just increase 25% of the software price to balance of more powerful chip in the market, could had stop the frustration of most users. Anyway, I am not working or own vmware inc, is their choice to price it that way and Good Luck to them.
I want to iterate one thing and one thing only. Your customers are NOT happy with this decision. This should be enough for VMware to think about how to revamp this in the very near future because their competition doesn’t do this.
We as the customer have for a number of years been trying to cut hardware costs and ever escalating hardware storage requirements. The datacenters were getting bigger and bigger and customers have been moving larger and larger workloads to Virtual space because of the stability, portability and recovery.
I love VMware and have for a while been trying to fight the companies urge to move to cheaper products because of the quality. There is no quibble with VMware being able to make additional revenue and there is definite advantages to the new product but the customers are feeling screwed. Oracle began doing this with their licensing, pricing per core and I have seen more and more migration to the competition MS SQL Server.
Just my $0.02
Another example is lets say I don’t need all these fancy features of Enterprise or Enterprise Plus. I will be using my Servers for DEV and Test environments where some downtime is accepted if there are major issues.
I am only limited to 48GB RAM per my current 2 CPU license if I upgrade from vSphere 4 Standard to vSphere 5.
So instead of 2 CPU licenses for vSphere 4 Standard. Now its 5 CPU licenses for Standard. For me to buy if I want to use 128GB of RAM.
The point is we would like for the pricing to be similar to vSphere 4 is possible. Yes vmware needs to make $$$$ but the way it is now is joke.
Currently the only way the pricing is similar is when you compare more than 6 cores.
If with vSphere 4 I have 2 socket 8 core Server with 128GB RAM. I need to have 4 CPU Enterprise License.
With vSphere 5 I still only need 4 CPU Enterprise License because now I am paying for RAM instead of cores. This is the only scenario where this is the case. All other cases you pay extra $$$$$$$..
vmware should raise the vRAM limit to at least
32 for Standard
48 for Enterprise
64 for Enterprise Plus
or something like that.
If the current vSphere 5 pricing stays you will see either people stay with vSphere 4 as long as possible or will slowly move to Hyper-V as it has similar features or will have them soon.
I am okay with keeping this thread open and having people responding about the fact that this licensing scheme is incorrect but I would prefer if you would use your own name or at least one with some level of respect.
Now, I can understand your frustration but stating Hyper-V offers the same functionality you are either misinformed or a competitor. I don’t know if you have seen what we offer today with 4.1 but Hyper-V does not come close when it comes to features / manageability. 5.0 offers a multitude of new features on top of that. I respect your opinion, but please be realistic and show respect to the people responding to this thread as well.
> should raise the vRAM limit to at least
>32 for Standard
>48 for Enterprise
>64 for Enterprise Plus
>or something like that.
>If the current vSphere 5 pricing stays you will see either people >stay with vSphere 4 as long as possible or will slowly move to Hyper-V as it has similar features or will have them soon.
Thanks for the feedback. This makes more sense than a generic complain. I trust you agree that the model makes sense. If not we are all ears to understand hat makes a win-win (VMware and customers) and that would allow us to charge proportionally to the value delivered.
I disagree on hyper-v being at feature parity even at lower SKUs (but that’s just me).
Raising the limit to fit current hardware specifications would help those of us who bought vSphere to run on very ordinary mid-range hardware and now find that the SnS for which we have already paid doesn’t cover the upgrade. VMware pretending that a DL380 filled with 8GB RAM chips is in some way a cutting-edge supercomputer just makes those of us who recommended vSphere look stupid, so a sensible “exchange rate” between vSphere 4.1 and 5.0 would take some of the grim comedy out of the discussions we are all going to have to have with our management.
Ultimately, though, vSphere 5.0 will not be competitive for some common workloads – broadly, large-memory, low-CPU workloads which don’t need the more advanced features of vSphere. That is a deliberate change of business strategy by VMware which is unlikely to alter over the next few weeks, and customers and partners will have to adjust to it.
Point taken Michael. Thanks for the feedback.
Enjoy your holiday!
I think the new VMWare pricing makes scale up versus scale out a moot point for most customers my size. We have 6 hosts in our cluster, 2CPU with 96G of RAM each. So we’re well positioned for a transition to vSphere 5. The cost to upgrade the to 16G DIMMs + VMWare license costs make it just as cost effective to add an additional host, since we’re not constrained by networking, space or cooling. It’s all a wash for us.
That being said, I do agree the pricing should have been set at 64G per CPU. 48G per CPU is targeting today’s sweet spot of 6 x 8G on Nehalem/Westmere CPUs. This completely discounts AMD solution or at the least puts a tax on AMD hardware, which will probably act as a disincentive for anyone buying servers today.
I was just making a point that many people would agree most valuable thing is HA and vMotion. Which Hyper-V has similar feature. All other new features of vSphere 4.1 and 5 are great and nice to have but not a must and are rarely used by regular folks.
Some of the advanced Enterprise features I have used in the last several years would probably be the Storage vMotion and I think it was like 5 times in that span.
I would suggest you do a survey of the most features people use every day and need every day, must have and nice to have etc. It would be a good Poll.
With the new licensing scheme you might be swaying people do more with less or leave to other competitors.
At the end of the day its all about the $$$$. Can you explain to your Boss you will need to double your licensing cost in vSphere 5 because of VMFS5, Storage DRS or Hardware version 8?
The point is stated many times is that we like the green monster and we would like them to reconsider the licensing change.
@vRam, your pint is taken too. Thanks.
I do think however that there is more behind choosing vSphere VS hyper-v than feature comparisons and check boxes. I remember people speculating that as soon as ms had live migration VMware was toast (at least in the SMB). Well we could start an entire new discussion on this but I think we all agreed this didn’t happen.
Massimo.
5 features are not features or an advantage unless you can paid for them. However this is a major change and will cost large amounts more. Some have already bought new hardware. So how do you think someone with new cisco UCS B230 with 2 sockets and 512gb of ram, will be able to use it? this was a blindside move on vmware’s part.
Also the VMTN is where all the discussion is, if vmware wants to help people then go to VMTN and explain. Not behind close doors or in personal blogs. That would help alot of people!
I know there is a lot of discussion Ron on VMTN about vRAM but that was not the topic of this article. My article was a response to Tom Howarth’s question in an article about scaling up vs scaling out. I don’t think I can contribute anything to the VMTN thread to be honest.
I have share this blog via my G+ and added some comments – also below:
Personal Blog from a VMWare employee, raises some interesting points/questions about VMWare licensing for vSphere 5.
First thing to catch my eye was: “…best practice that VMware PSO has had for the last years has been 5-8 vCPUs per core…”. Which doesn’t dovetail well with any real-life scenario I have encountered. This would mean our 2S 6 Core servers could host between 60 and 96 vCPUs, which is just not practical – now admittedly we are running our of RAM before we run out of CPU, but…
Next bit was VMWare’s, or at least this employees insistence that VMWare want to keep the “price per workload” constant – which imho, goes against the trend with Hardware at very least – Price per CPU cycle and GB of RAM keeps dropping, while software requirements keep climbing.
At the end of the day VMWare have to make money, and I do not begrudge them that. This change to licensing is not going to make much of a difference to where I am at the moment – though we are at the upper limits of the licensing model per CPU. IMHO what would be fairer/better would be to either:
Increase the vRAM/CPU entitlement to something more relevant for today: say E+ at 64Gb and introduce a higher level (call it E++) where there is no limit, charging a suitably high price for that.
OR be fully transparent/open and have a base price/CPU {or core…} and charge for RAM on top of that (or forget about the processor completely and just charge for RAM per cluster/site).
What gets me more is VMWare introduce a mechanism to over provision RAM, and a way that guests can, essentially, share RAM, then penalise shops who take advantage of this (e.g. one 2 CPU server with 96GB of RAM, it hosts 2 DCs, a couple of Exchange Servers, and a couple of SQL servers, etc… total allocated RAM 128GB – some of these machines can benefit from RAM sharing due to similarity between their roles – for vSphere 5 you will need 3x E+ licenses..)
Wanted to include a post back from Google+ – just found a way:
http://t.co/ivqDJE8
>> What gets me more is VMWare introduce a mechanism to over provision RAM, and a way that guests can, essentially, share RAM, then penalise shops who take advantage of this
Totally agree.
All benefits of VMware memory allocation (balloning, TPS, memory compression), they are all gone due to vRAM entitlements.
VMware has tried to tell us that the new license model was done to benefit the customer. Which is NOT the case for a lot of “their” customers!
VMware, make “these” customers satisfied, do something until it’s too late…
So overcommitting and using TPS etc doesn’t decrease costs? What about the need for more hardware when you don’t use it? What about management of those extra hosts? What about power, cooling, rackspace? There’s also costs associated to all of them. I understand what you are saying but it just doesn’t sound right to me…
Duncan, it just doesn’t “sound right” to you because you apparently feel the need to defend VMWare. And because you probably would get into deep trouble if you – in this public – would admit that the new licensing model is faulty.
Only last week I upgraded a customer’s six ESX servers (IBM x3650M3, e.g. just simple 2 socket servers with 6C CPUs) from 72 GB each to 120 GB each. They are running Enterprise Plus yet STILL they would not be covered by the vRAM limits in vSphere 5. Now 120 GB isn’t such a shocking high amount of memory … consider the IBM x3690X5 which can go all the way up to 1 TB of pRAM! That customer upgraded the memory because their CPUs still sit idle most of the time but they were hitting the memory limit (with the 72 GB per host) constantly. I think they run quite an _average_ environment.
No I am saying that because I seriously think TPS saves you money regardless of the vRAM licensing scheme.
Keep in mind by the way that I also could have not posted anything regarding this topic. If people feel I am blatantly off than they are free to respond to this topic. I feel I made my point and that scale-up / scale-out is not impacted by vRAM licensing, the same rule set applies you will need to do the math!
>Duncan Epping says:
> No I am saying that because I seriously think
> TPS saves you money regardless of the vRAM
> licensing scheme.
It does, but – frankly – this is no only not the point, it is not relevant at all.
The point is that the vRAM base for licensing is just a bad decision as it contradicts principles of ESX of the past years. And even if we were to accept this, any such change of licensing scheme should be cost neutral at the point in time when it is done. And now PLEASE don’t reply with the broken record of repeating that “most customers” are not affected and we “should do the math”. Just read what has been said above.
Maybe it’s just me but at least with linux guest VMs, I usually see little to no benefit to TPS. Talking maybe double digit megabyte savings per several gigs of ram. I do see pretty high TPS savings on windows VMs by contrast but the number of those that I have is tiny compared to linux.
Does anyone else know of any company that tries to charge “per workload” ? I had not heard of this concept until this discussion I believe.
If you need 1TB of ram in a VM – even if the licensing was the same as vSphere 4 – why would you put it in a VM? I’d much rather have it on real hardware. 1TB of ram is a lot of ram, and 32 vCPU limit in vSphere 5 is not good enough since 1TB ram hosts will likely have at least 40 if not 80 or even 100+ cpu cores.
I’d be willing to bet money that a lot of the vmware customers that have these small machine configurations are probably not very experienced, and likely are getting experience on this round of hardware and when they go to do a hardware refresh will prefer to scale up, rather than scale out(licensing permitting). I’ve seen it time and time again. I’ve gone through it myself even(vmware user/customer for 12 years now).
VMware’s per-VM licensing on some of their products is just as bad, if not worse than vRAM licensing. Like vRAM per-VM licensing causes customers to have fewer, larger VMs (Running more tasks), rather than larger numbers of smaller VMs running more isolated tasks.
At least with those per-VM licensed products you can easily avoid them. Harder to do when the licensing is tied to the hypervisor.
I agree with an earlier poster that most people likely do not need most of those fancy features (myself included). I don’t even need HA myself, vMotion is nice to have though. But due to licensing restrictions I’m forced into higher tiers of software whether I want it or not.
I have been a very vocal proponent of vmware for many, many years, fought tooth and nail to keep others out, this vRAM thing is a very unfortunate development, and it will just make it nearly impossible to keep vmware beyond 4.1 — with that said, having looked at vSphere 5 I see *NOTHING* that gets me excited, even if the licensing was unchanged yet alone the licensing hit many of the more advanced users could face if they opt’d to upgrade. My own strategy is to stay on 4.1 for another year or two and then migrate to a competing platform – most likely KVM. Ironically I have been telling folks this for a few years now, saying/hoping KVM would be stable/mature enough to be a decent hypervisor by ~2012 or so. I just didn’t know vmware would show me the door so quickly on their own.
Short of eliminating this vRAM thing entirely I can think of two things vmware could do to make it better
– charge based on what is USED rather than what is PROVISIONED (you own the hypervisor you already have access to this data) – and exclude things like memory assigned to disk buffers/cache
– license vRAM and CPUs separately, vRAM licensing should be a small fraction(30% or less) of the cost of base CPU+Vram licensing and have no support fees associated with it. Think of it like a vRAM “acceleration kit”, and then you buy just vram licensing on top of the base kit.
I believe I read the free ESXi has been limited to 8GB of vRAM as well, if true that is just incredibly stupid and greedy.
It’s sad that vmware has come to this, I guess the emc stuff finally got to them, reminds me of reading this a while back –
http://virtualization.info/en/news/2009/10/vmware-now-under-massive-reorganization.html
All I want is a rock solid hypervisor, more than anything else show me a vSphere feature list and I’ll knock off 95% of the features as things I don’t care about (nice to have granted, but not deal breakers), give me pricing for the core hypervisor (unlimited cores per socket and unlimited ram per host) and I’ll be happy. vMotion is the one feature I think I’d really want, though I’ve lived without it for years before, it’s not the end of the world.
I’ve liked vmware primarily because of the quality of the low level software, I have never had ESX crash on me, nor GSX, nor Vmware server, I can’t remember vmware workstation ever crashing on me. Maybe vmware did in the early days (before the workstation tag was assigned to it – yes I still have my vmware 1.0.2 CD), though if it did it was rare enough that it didn’t leave any memories in my head.
A few years ago I got into a pretty big fight with my (new) boss over Xen vs ESX 3.5. That, among other things convinced me to leave the company, the boss directed the remainder of my team to try Xen – a month later after failing to get Xen funtional after a month they gave up and went back to ESX. Just one of my many stories…
>It’s sad that vmware has come to this, I guess the emc stuff finally got to >them, reminds me of reading this a while back –
>http://virtualization.info/en/news/2009/10/vmware-now-under-massive-reorganization.html
I am not sure what EMC has to do with all this frankly.
Also I am not sure what linking a 2 years old article that talks about 100 layoffs could add to this discussion (especially when we hired a “few thousand” people in the last “few years”).
All other technical points re vRAM are taken.
Massimo.
the point was that EMC has finally gotten it’s dirty hands into every corner of vmware. It usually takes a few years for it to happen after being acquired. First couple years often the parent operates “hands off” and sees how it goes, then they start to infiltrate. EMC, like many big companies are well known for their overcharging of their stuff (either that or charging in line and massively under delivering on promises).
There were signs of it earlier, such as the introduction of Enterprise Plus licensing back in vsphere 4, as well as per-VM pricing on several products like SRM.
At one point vmware was going to focus more on the “management” layer since the core hypervisor was mature (I seem to recall vmware themselves even saying the hypervisor was going to become commodity or something). I guess they didn’t find enough customers interested in those more advanced features so instead they integrate them into the package and jack up the price to justify the features (which is fine if you need the features).
And here I was hoping/expecting vSphere 5 would be a completely free upgrade for those with up to date support contracts. I recall in ESX 3 the licensing policy was upgrades were free only for the major version you were on (ESX 3 was the first ESX I used, prior to that I was using GSX), then VMware did a wonderful about face and said vSphere 4 would be free to those with support contracts. Now it’s of course the opposite for many (not most I’m sure) customers.
Netflix was projecting upwards of 20% customer churn with their price hikes, it wouldn’t surprise me if VMware had similar on their existing customer base.
Hopefully this conversation continues across the internet in various forums until vmware does something to rectify the situation, though it make take many customers leaving to trigger it. Time will tell..
@Nate,
are you saying that if we raised the price 20% no one would have complained? Well that is good to know.
As I said earlier changing a licensing model is always a pathos. I think there is a lot of it going on here. I also know that there are some customers that are being impacted by it and I have my own opinion on what we should have done or should do in the future. What I know/think is that the principles are sound. There is also a lot of other people that are looking into this as well because, contrary to what you are saying, I don’t think we are doing this to steal money “opportunistically”.
Thanks for the feedback.. rest assured is being read and crunched.
One final note: sorry but you know nothing about EMC / VMware. Oh well you know what you read on the Internet. Buy me a beer (well perhaps 3 or 4) and we can chat more on this off-eyes… I will of course decline to have ever said anything to you. I will also decline to have ever written this too 🙂
Massimo.
>> My own strategy is to stay on 4.1 for another year or two and then migrate to a competing platform.
You will be not the only one…
All companies who are not vRAM compliant today (due to a scale-up strategy) will not be able to have a positive business case to upgrade to vSphere 5.
“All I want is a rock solid hypervisor, more than anything else show me a vSphere feature list and I’ll knock off 95% of the features as things I don’t care about (nice to have granted, but not deal breakers), give me pricing for the core hypervisor (unlimited cores per socket and unlimited ram per host) and I’ll be happy. vMotion is the one feature I think I’d really want, though I’ve lived without it for years before, it’s not the end of the world.”
If you get what you want (Unlimited=free), VMware will be out of business in 3 years, and then MS & RH will own virtualization, and innovation & any quality or “rock solidness” you mention will be out the window…Walmart might as well enter the market at this point and sell a hypervisor…
First off, thank you Duncan for running this website; your blog is always of the highest quality and I look forward to many more interesting and informative posts. I am replying because you said you would pass on comments to your internal contacts at VMWare licensing.
I have a scenario that has not been discussed previously. I recognize that I am currently “the little guy” and that VMWare has very little incentive to keep customers like me happy.
To reiterate what many people have said… VMWare’s new licensing scheme makes it significantly more expensive to upgrade from vSphere 4 to vSphere 5 on my EXISTING hardware.
My existing HW/SW setup:
–2 x Dell R710 with the following relevant hardware specs
—-2 socket x 6 core 3.33 ghz pCPUs
—-128 GB pRAM
–1 x vSphere 4.1 Essentials Kit ($560)
—-This consists of Licenses for a maximum of 3 Hosts, 6 x pCPUs
This simple example is the core of our Lab/Dev/Test environment, and makes extensive use of the “scale up” principle you discuss. I have been planning on upgrading that Essentials Kit license to
–1 x vCenter Server 4.1 Standard + Basic support ($4,894)
–6 x vSphere 4.1 Standard + Basic support (6 x $1,068 = $6,408)
So in vSphere 4.1, VMWare stands to make about $11k off of me. A little profit for them, and I get several nice features I don’t currently have.
UNFORTUNATELY in vSphere 5.0, assuming the pricing remains the same as what is published right now:
–My Licensing would entitle me to use only (6 sockets x 24GB ) =144 GB of vRAM, which means I am unable to use 112 GB of my pRAM (I am losing approximately half of my current capacity)
–Not only can I not use all my pRAM, I am no longer able to take full advantage of memory optimization techniques such as TPS, compression, etc
–I am being penalized for my execution of a extreme “scale up” strategy.
–My dynamic lab environment has many VMs being stood up/down at any moment, but on average there is very low Activity
To fully utilize all my pRAM, I would need to purchase an additional 6 x Standard licenses… costing me $6,408 MORE in v5 than it would in v4.1 and the ONLY new feature I get for my money is VMFS5 and the minor improvements that come with it for a Standard license. (Please correct me if I am wrong, what other NEW features are in v5 Std that are not in v4.1 Std?)
A much more fair licensing scheme would be to charge based on “vRAM Active” and not “vRAM Allocated” as the current model is headed. You have mentioned several times in your ESXTOP articles of the importance of the Active memory metric, the usefulness of TPS, the benefits of overcommitment, etc. There has been similar statements made over past years in VMWare’s marketing materials regarding a 5:1 consolidation/overcommitment ratio. I think it would be much more fair for VMWare’s licensing to follow a model that reflects their own statements.
So in summary:
–An upgrade to v4.1 would cost me about $11k, whereas v5.0 would cost me about $17,500. Now that I see what VMWare thinks about us small customers, they are unlikely to get ANY of that revenue if they don’t change course fast.
–Consider an alternate licensing scheme based on Active vRAM
–Remember that not all customers with 96GB+ boxes have upper SKUs like Enterprise+, some lab environments have dozens of VMs sitting idle (taking advantage of memory overcommitment techniques)
Am I really such an edge case that this didn’t come up in internal discussions?
I will provide your feedback to the VMware licensing team. VMware appreciates every single customer including you.
Gents, you will need to update your calculations with new significantly increased vRAM entitlements to be presented on 3rd of August.
The updated license limits still are insufficient. License entitlements are “significantly increased” ONLY for the more expensive license types. Their new model does not help the lower cost license types significantly (i.e. Essentials).
My example above goes from using 144 GB vRAM in 5.0 to now be 192 GB vRAM in 5.0. I am STILL UNABLE to use 64GB of pRAM and with the exception of VMFS5, I get NO NEW FEATURES OR FUNCTIONALITY for my money.
Why on earth would I upgrade to 5.0?
Had a look at the revised licensing vRAM figures and have to say they have been significantly increased for higher end versions of vsphere(mostly due to the customer feedback). Based on the implementations that i have been part of 90% of the would be unaffected as most of them had enterprise/+ licensing (though there will always be use cases where customers will be looking at increased costs due to upgrade to vsphere 5).
Thanks, I can tell you that the licensing team did appreciate all the feedback in this post as well.
It’s pure fun to read this discussion now that the new CEO expected standing ovations for the removal of the “vTax” 😉