Sizing VDI: Steady-state workload or Monday Morning Login Storm?

For quite some time now we have been sizing VDI workloads by measuring what people are doing during the day on their virtual desktops. Or even worse, we use a synthetic workload generator. This approach WILL work to size the storage during the day, but what about the login storm in the morning? If this spikes the I/O load above the steady stte workload of the day, we should consider to size for the login storm…



Measuring login and steady state

In order to see the I/O’s performed during login and steady-state I once more used the 3D graphs of vscsiStats. This time I measured a Windows 7 Linked Clone desktop (using VMware View 5) during logon and during steady state in order to compare them. If the login spike is very high, and you have an environment where a lot of people login at the approximately the same time, it may be smarter to size for this login storm instead of the steady state workload. In that case it would be good to know how many IOPS a booting VDI image takes, and what its read/write ratio is…


Measuring the base disk

First I measured the base disk during login and steady state:

VDI base disk reads during Login (VM1) and Steady-State (VM2) of a 1,5GB Windows 7 Linked clone VDI desktop

As we can see here the graph of VM1 takes a lot more reads than during steady-state; more than double the amount in fact. For writes we see this:

VDI base disk writes during Login (VM1) and Steady-State (VM2) of a 1,5GB Windows 7 Linked clone VDI desktop

Here a huge write burst during login compared to the steady state workload. In the next section, I look at the disposable disk…


Measuring the disposable disk

The disposable disk is basically the disk where the Windows swap file resides. During boot and steady state there are hardly ANY reads going on here. Writes is a different matter:

VDI disposable disk writes during Login (VM1) and Steady-State (VM2) of a 1,5GB Windows 7 Linked clone VDI desktop

Here we see not that much difference. During steady state we see a burst of writes to the disposable disk that is very much on par with the writes going on during login.


Measuring the user disk

Not really all too important for this blogpost, but since I ran the measurement I’ll include the userdisk usage for both reads and writes here as well:

VDI user disk reads during Login (VM1) and Steady-State (VM2) of a Windows 7 Linked clone VDI desktop

VDI user disk writes during Login (VM1) and Steady-State (VM2) of a Windows 7 Linked clone VDI desktop

As you can see, there isn’t too much going on on the user disk. But it may give you an idea of how to size for the personal disks (since they’re often offloaded to other arrays or NASses).


Conclusion

As you can see from the graphs above, the login storm consumes far more IOPS than the steady-state workload. In environments where big login storms are to be expected, it may be wise to size for these storms rather than measuring on steady-state. What’s interesting, is that the number of reads during login appear to be about the same of even more than the number of writes. This would mean that we need to calculate (or better: measure during a pilot phase) what the impact is of both, and then use the worst-case one of those: Logins appear to generate more IOPS, but mostly reads where steady-state computing generates less IOPS, but mostly (around 80%) writes.

Comments are closed.

Soon to come
  • Coming soon

    • Determining Linked Clone overhead
    • Designing the Future part1: Server-Storage fusion
    • Whiteboxing part 4: Networking your homelab
    • Deduplication: Great or greatly overrated?
    • Roads and routes
    • Stretching a VMware cluster and "sidedness"
    • Stretching VMware clusters - what noone tells you
    • VMware vSAN: What is it?
    • VMware snapshots explained
    • Whiteboxing part 3b: Using Nexenta for your homelab
    • widget_image
    • sidebars_widgets
  • Blogroll
    Links
    Archives