Yellow Bricks

by Duncan Epping

Virtual SAN Datastore Calculator

· ·

Over a week ago I wrote an article around how to size your Virtual SAN Datastore. I included an equation to help everyone who is going through the same exercise. I figured I should be able to make life even easier by creating a simple Virtual SAN Datastore calculator based on this equation.

<EDIT> VMware has launched a great calculator. My tool served as a temporary solution as there was no calculator available, now that there is I highly recommend using the official tool: http://virtualsansizing.vmware.com/ </EDIT>

Reader Interactions

Comments

  4. How do we calculate the IOPS vSAN can provide (combine with SSD/flash disk) with the assumption made above? It should be better than just combing the SAS IOPS total performance, right?

    Does it provide improve performance if we size up the SSD/flash disk versus SAS?

    • Sure it will be a lot better. However it is almost impossible for me to incorporate SSD into the equation as there are so many variable. There are many different SSD types and the performance will be determined by the type of IO / IO size etc. So what performance would look like from the VMs perspective is really difficult to predict. I can add theoretical numbers of course, but not sure how valuable that would be.

    • Every RAID or RAIN system has an overhead. In the case of VSAN for each write a number of writes will occur on other hosts depending on the selected failures to tolerate policy.

      So if you have 1 VM and 1 failure to tolerate and do 1 write. The 2 hosts will need to write the IO to SSD.

    • Basically it means that you will be able to rebuild all of your components of the VM without losing performance when a host failed

  7. Hi Duncan — great tool. Will you be updating this tool to include the 10% Flash:HDD ratio of compute VMs, the tool seems to use the size of the full datastore.

  8. Why can’t VMware publish easy to understand documentation like this? Have you read the sizing and design guide. It’s horrible. Thank for picking up the slack Duncan.

  10. Awesome tool Duncan, thanks for putting this online! I have some questions/comments:

    1) How/why does the average RAM used per VM affect the VSAN capacity requirements? (Your article about calculating VSAN datastore size does not appear to mention this.)

    2) I really appreciate your insight into clusters that have space to rebuild themselves in the event of a complete host failure. If I understand this correctly it means that without this extra space the cluster will be running without any redundancy until a failed host is repaired/replaced, and that presumably a second host failure would bring the cluster down just like a RAID5 array with no hot spares. Is that correct?

    3) Unless I am missing something the “Total cluster disk capacity required” is reporting the value without rebuild space but the “Total number of disks needed per host for VSAN Cluster” is using the value with rebuild space. For example I have a configuration with 6 hosts. It calculates 9.35TB disk capacity per host without rebuild and 14.02TB with rebuild. “Total cluster disk capacity” is reporting 56.07TB (6×9.35) With 300GB HDDs it shows 47 are required per host, which is 84.6TB total. (47*0.3*6) This closely matches the total capacity needed with rebuild of 84.12TB. (6*14.02) Assuming I’m correct I’d suggest adding an “Include Rebuild Capacity” check box and then changing both values accordingly so it is clear which values are being used. Also consider putting a “Physical disk capacity” output box next to the “Select a disk type” and maybe calculate the difference between the selected HDD total capacity and the required capacity so the user can see how much potential wastage there is in their selected HDD capacity.

    4) For performance does it matter how the Flash capacity is split over HDD groups? E.G. Is a single group with 6 x 2TB HDD & 1 x 400GB SSD going to perform the same as 2 groups each with 3 x 2TB HDD & 1 x 200GB SSD? The VMWare documentation I have found so far recommends lots of small groups for performance, but does not make this clear if this requires a separate host per group or multiple groups per host.

    5) Is it possible to have the tool work out the possible SSD and HDD group combinations possible using enterprise-size SSDs (100, 200, 400, 800) given the current restrictions of a maximum of 7 HDD to each SSD? It is especially important once the number of disk per host exceeds an even multiple of 7, as that means extra SSDs will be needed regardless of the total Flash capacity needed per host. (E.G. If I need 10xHDD per host and 300GB of Flash I will need 2 x 200GB SSDs and not 1 x 300GB as the drives will have to be split into 2 x 5HDD groups.)

    6) Talking of cost, is it possible to add price input boxes for the HDD and Flash selections so the user can add in the current market prices? Once this is in place it would be possible to automatically calculate both the total cost of the solution and also averages like $$$ per IOP, per TB and per VM.

    Thanks again,
    Dan.

    • 1) Because the swap file needs to be accounted for to be precise. I added that to the calculator to avoid people over estimating available capacity
      2) Yes
      3) When I have time I will look at expanding the physical disk section with your suggestion
      4) Sure this can make a difference… it will mean more IOPS per GB
      5) Not sure, will need to look at that. I threw this together quickly to help people, but VMware is releasing a tool as well soon which may do what you look for
      6) Will have a look if I can do that…

      PS: time is limited with holidays coming up and VMworld and finalizing a book on the subject of VSAN so not sure how much I can do.

  11. I used 300 VMs, 4Gig vRAM, 50GB image, FTT=1 and 4 servers and got 32,400 with your Jan. formula and this calculator gives 34,800 — what am I missing?

    • Yes I found a bug in the calculator. It looked liked the “swap file” was not being taken in to account for whatever reason. Note that i calculate MAX needed storage and not “consumed”, so you should be good unless you use 100% of all possible storage.

  12. Hi Duncan, What is the Projected percentage VMDK capacity used? as changing this figure makes no difference to the results?