Are you wondering how Virtual SAN is doing? The recent earnings announcement revealed that… Virtual SAN is doing GREAT! Over 3500 customers so far (21 months after the release!) and 200% Year over Year growth. I loved how Pat Gelsinger described Virtual SAN: “VMware’s simple enterprise grade native storage for vSphere”. It doesn’t get more accurate and to the point than that, and that is how people should look at it. vSphere native storage, it just works. Just a couple of things I wanted to grab from the earnings call (transcript here) that I think stood out with regards to VSAN:
and I – having been three years at EMC as a storage company, part of it is it just takes a while to get a storage product mature, right, and that – we have crossed the two-year cycle on VSAN now. The 6.2 release, as I would say, checks all the boxes with regard to key features, capabilities and so on, and we are, I’ll say right on schedule, right, we’re seeing the inflection point on that business, and the 6.2 release really hit the mark in the marketplace very well.
I’d say we’re clearly now seen as number one in a hyper-converged infrastructure space, and that software category we think is going to continue to really emerge as a powerful trend in the industry.
I think Zane mentioned a large financial services company. We had a large EMEA retailer, a large consumer goods manufacturer, a large equipment engines company, and each one of these is really demonstrating the power of the technology.
We also had good transactional bookings as well, so it wasn’t just in big deals but also transactional performance was good. So the channel participation is increasing here.
So we really left Q1 feeling really good about this area, and I’m quite bullish about its growth potential through the year and 2017 and beyond.
I think I don’t need to add anything other than… Go VSAN!
ericcsinger says
Just my .02, but vSAN still has a way to go before I’ll ever consider replacing my SAN. There are just too many limitations with vSAN that I’m not cool with.
1. With shared storage, I can scale it independently of my compute, both from an IOP and capacity perspective. vSAN’s biggest issue here IMO, is the lack of support for external DAS (JBOD), or at least the lack of HCL. Specifically, what I would recommend is keeping the storage outside of the host. This way you can upgrade you host and re-attatch the JBOD. The problem is, most vendors like Dell, don’t support transferring JBODS from one generation to another. Which isn’t vSAN’s fault, but it’s a limitation IMO that makes vSAN expensive. That is, having to buy new storage every time I buy a new compute node. If we COULD get good support of external storage, you could theoretically scale your capacity at the very least independent of your compute.
2. Having to pony up extra for vSAN on top of an already expensive ESXi license is like salt in a wound. Frankly this should just be part of ESXi. Maybe having something like all flash is only available in EP, and you can do hybrid with standard edition. Microsoft does this with storage spaces. You can debate all day long about vSAN being better than storage spaces, but none the less, its included with Windows at no extra charge. Given how much Hyper-V is chomping at ESXi, I would think vSAN (and NSX) could have been compelling reasons to continue investing into ESXi, now you’re just making it more expensive. Meaning, when people say “oh there’s nothing big in ESXi” you could have said “we just added vSAN and NSX at no extra cost top that Microsoft.”
3. No offence, but Vmware’s production support has totally tanked over the years. I used to brag about how good Vmware’s support was and now I loathe having to call them. Its one thing to have a single host issue, and needing to deal with bad support, it’s a whole other issue when one issue might mean affecting not only your compute, but your storage as well.
4. Lack of shared storage support for physical clustered systems is still an issue for shops like mine. While its only a few systems, it still means I now need two different storage solutions.
5. I wish anti-affinity rules actually meant that VM’s *and* there disk copies were kept completely separate in a cluster. TMK this isn’t the case. Maybe I’m wrong here though.
6. Lack of reference design for maximizing performance and having benchmarked solutions is a big issue IMO. You guys claim to support multiple disk groups and up x drives, yet I haven’t found a single example of the maximums running in the wild, nor have I found anything greater than a single disk group per host being benchmarked. There was one example with two disk groups released by you guys, but IIRC it was a little light on the details. Would just be nice to see something like “here’s a 6 node cluster with 1 disk group, here’s the same with 2 disk groups, here’s the same with 3, etc.” just to see if how linear the performance improvement is. If I have 24 disk, do I want to make 3 disk groups, 4 disk groups or 5 disk groups? I could experiment, but it would be nice to see some examples from you guys.
7. I would LOVE to see the ability to mix hybrid and all flash in the same cluster. One type for my SQL server and the other for my file servers.
That said, its very clear you guys are pouring a ton of resources into the product, and that is VERY cool. I just wish the same level of innovation was still occurring in the core ESXi server, which sadly IMO isn’t happening. There’s still so many things Vmware could do inside the hypervisor that would continue to set it apart from the likes of MS, KVM and others. Everyone says the hypervisor is a commodity, and I
Duncan Epping says
Thanks for providing your thoughts around why you feel VSAN is not a fit for what you are doing in your environment. Which is fine. For customers like yourself we have developed VVols and the SPBM Framework which allows you to manage you systems in a similar smart way. Just some things I want to comment on:
1. You can have nodes in your cluster that do not contribute storage. You can also go up to 35 capacity devices per server, which is typically more than anyone needs. (35*4 or 6TB is a lot of capacity)
2. I understand your comment, but I am assuming that you didn’t get your current storage system for free? Building a solution like this has a high cost, and we are a business in the end.
3. I will take this feedback back to the folks who own this, I cannot comment on it other than that as I am too far away removed from it.
4. Fair point, see my initial comment. (we are fixing this though at some point in the near future)
5. Great feature request, I will file it for you
6. There are a whole bunch of papers on performance and apps etc. What explicitly would you like to see?
7. Noted, and feature request has been filed a while back. Being considered and being worked on.
Thanks for your open and honest feedback, appreciated.
All of this though, doesn’t take away the fact that we have the largest number of customers in the HCI space and have grown immense, and I am extremely proud of that.
Thanks,
Duncan
Eric Singer says
Hi,
I think my last paragraph got cut off a bit, which is basically thanking / praising you for your blog. So just to get that ahead of any replies, really appreciate your blog over the many years that I’ve been following it.
1. To point one, my point is that HCI has one flaw, which is being able to smartly scale storage / compute independently. HCI is a GREAT fit for web / vdi because those workloads tend to scale in a linear manner. Generic enterprise however does not. Having hosts without storage to me sounds like a hack, and more than likely, I’d be deploying more hosts because I would lack the storage not the compute. Again, that’s why I think your Achilles heel at the moment is external DAS. If you could clinch that + get some vendor agreement to support JBODs / SAS cards in newer servers, you’d enable vSAN to be a lot more flexible and more importantly enable me to get more milage out of my storage.
2. Sure, but I’m paying a lot of money for a hypervisor that to be blunt has lacked serious innovation, slower feature releases, etc. I’ve been working with ESX since 3.x and I remember how feature packed each release was and how often new features were rolled out. What I’m saying, is make vSAN a feature of ESXi not an add on. You think you’ll lose money, I think you’ll retain revenue / increase new customers. Plus going to point that I think go cut off, if you guys switched to per core licensing (MS is doing this anyway) you’d probably get MORE revenue for larger shops (like mine) all the while making your product more attainable for the little shops (new customers).
6. Think of it from my view, the SysAdmin. I go to Nimble (which I own) and say I want to know how fast scenario X is and scenario Y is, and for the most part they can give me a report for each scenario and each configuration of their SAN. What I would love to see (as one of many examples) is show me a 5 node cluster. Show one config with a single disk group one with 2 disk groups and one with 4 disk groups. Show me single VM max IOPS @8k 100% random read / write and then 70/30. Now show me throughput for that single VM. Maybe demonstrate the difference in disk striping too, or this is erasure coding vs. mirroring. Then repeat said tests but for multiple concurrent VM’s. I realize there are probably 100’s of different tests that could be run, but every test I see, its always single disk group, its always multiple VM’s, rarely all flash, and mostly third party (Anandtech / StorageReview as an example). I’m not saying the tests are bad, it would just be nice to have a central place to go, and basically try to get an idea of how vSAN would perform as its scaled not only out, but up / deep. And also see the performance view from a single VM as much as we see multiple VM benchmarks. If my SQL server for example needs 1.5GBps of throughput for a huge BI report, it doesn’t do me any good that vSAN’s aggregate performance is 4GBps, but a single VM can only do 400MBps (as an example).
Anyway, I love how fast vSAN evolving. I’m not saying I’ll never run vSAN, just not right now.
Georgi Ikonomov (@GIkonomov) says
“if you guys switched to per core licensing (MS is doing this anyway) you’d probably get MORE revenue for larger shops (like mine) all the while making your product more attainable for the little shops (new customers)”
“vRAM entitlement” should be enough said on that point!
ericcsinger says
Not even close. vRAM was confusing, complex to calculate and it was non-standard with anything else in the industry I blogged about the the real cons of per socket. Its a scheme that made sense when there was one to a few cores per socket. Now it makes zero sense.
1. It’s just keeping pace with everyone else. Per socket isn’t a salable model. Its the whole reason Oracle and now even Microsoft are switching over.
2. It only costs more money for folks who run the max cores per server (folks like me) available. HOWEVER, I also have a SQL cluster that’s per core and guess what, I’m locked into running a limited number of cores per host in that cluster because of SQL, which means VMware is costing me more than it needs too. See below for the prime example of what I have to balance
2a. Related to my SQL cluster, I would much rather run 4 sockets / 8 cores per @3GHz (which is what I do now) then 2 sockets / 16 cores per @ 2.3GHz. Unfortunately VMware now costs me more than it would if it was just per core.
4. Its the most flexible, cost efficient model for both the vendor and the customer. The vendor gets compensated correctly as servers get bigger / badder and the customer has the option of purchasing servers that are sized based on what they actually need, not at maximizing cost. Think about a small business customer. Single socket, quad core. Making them pony up for a single socket of VMware is like asking them to subsidize your now 22 core socket. That’s part of the reason they’re switching to hyper-v among others.
5. VMware is raising their prices anyway. I’d contend that never would have needed to happen with per core.
Don’t get me wrong, I love it when I buy a new server and don’t have to worry about whether its going to effect my VMware sockets, but at the same time, I have a mixed environment where there are hosts that I can’t just will nilly pick a server with the most cores. Both from a what I’m entitled to run AND because I do care about performance. Single threaded operations do much better with higher clock rates. Again, I get peoples hesitation, but the trend is already started and its picking up steam. That’s ok though, because it makes sense, unlike a lot of other schemes.
Duncan Epping says
Guys, thanks for commenting, but this is not a discussion board for random topics. If you have more thoughts, please use the VSAN VMTN Forum for it and not my blog.
CK says
We have VSAN running in our VDI environment. So far, not very impressed with it. Wish it was easier to manage, install, etc…probably will replace with Nutanix.
Kevin says
We have some Dell 730xd servers coming in and we are looking forward to configuring a new AF vSAN. Thank you for the blog,