login vsi company logo login vsi company logo 250x40
header-02.jpg

Forum Announcement

The Login VSI forum has been disabled and now only serves as an archive. If you would like to submit a new support ticket, please send an email to support@loginvsi.com.

Login VSI: TEST boot storms of VDI?

  • merrist
  • merrist's Avatar Topic Author
17 Apr 2012 11:20 #191

Like the title said. Does the Login VSI support to test the boot storms of VDI (VMWare View)?
if it supports,what should I do for testing the boot storms?

I can just have a glance at the Login VSI Admin Guide paper, so I think testing the boot storms is possible.

we should use the Login VSI PRO, because we must use the custom workload function, we can use the ourselves' script to start the VDI's desktop simultaneously, right?

who can tell me the concrete step to test the boot storms of VDI?

thanks :) :) :)

Please Log in to join the conversation.

  • jventresco
  • jventresco's Avatar
17 Apr 2012 12:58 #192

LoginVSI is for testing user workload, not bootstorms.

You can test bootstorms easily enough without any tools other than ESXTOP and your storage array tools. Use ESXTOP to measure GAVG (average guest milliseconds per command) for your desktop filesystems to see latency during the bootstorm, and use your array tools to measure latency at the array level. GAVG is a key metric because it is latency measured at the host, whereas the array stats are important but not compared to what the host actually sees (although the two metrics will likely be similar).

If you are seeing > ~50 ms from GAVG you can assume that any desktop on that file system will see degraded performance during that period of the bootstorm. I like to keep file system performance within 15ms at all times, but at 50 ms the desktops should still be usable (this depends on your image of course). At 100ms I'd expect the desktops to be booting so slowly that certain services do not start at bootup, which is a problem.

- Jason

Please Log in to join the conversation.

  • merrist
  • merrist's Avatar Topic Author
18 Apr 2012 08:11 #198

jventresco wrote:

LoginVSI is for testing user workload, not bootstorms.

You can test bootstorms easily enough without any tools other than ESXTOP and your storage array tools. Use ESXTOP to measure GAVG (average guest milliseconds per command) for your desktop filesystems to see latency during the bootstorm, and use your array tools to measure latency at the array level. GAVG is a key metric because it is latency measured at the host, whereas the array stats are important but not compared to what the host actually sees (although the two metrics will likely be similar).

If you are seeing > ~50 ms from GAVG you can assume that any desktop on that file system will see degraded performance during that period of the bootstorm. I like to keep file system performance within 15ms at all times, but at 50 ms the desktops should still be usable (this depends on your image of course). At 100ms I'd expect the desktops to be booting so slowly that certain services do not start at bootup, which is a problem.

- Jason


so,the Login VSI does not surpport bootstorms of VDI.right?

Please Log in to join the conversation.

  • jventresco
  • jventresco's Avatar
18 Apr 2012 12:48 #202

Correct.

LoginVSI only measures the data that is recorded during the LoginVSI test. That data is measured by the LoginVSI platform, and is independent of any ESxi or disk stats that you might examine.

- Jason

Please Log in to join the conversation.

  • merrist
  • merrist's Avatar Topic Author
18 Apr 2012 23:51 #208

vjason wrote:

Correct.

LoginVSI only measures the data that is recorded during the LoginVSI test. That data is measured by the LoginVSI platform, and is independent of any ESxi or disk stats that you might examine.

- Jason


thank you :)

Please Log in to join the conversation.

  • merrist
  • merrist's Avatar Topic Author
20 Apr 2012 08:52 #226

boot storms of VDI

Can the Workload Core surpport it? :huh:





like the picture。

the Workload Core is the way to testing the boot storms for VDI?

【select workload Core + VSITimers】

Attachments:

Please Log in to join the conversation.

  • jventresco
  • jventresco's Avatar
20 Apr 2012 10:31 #227

No, core is just an empty workload that logs in users only.

www.loginvsi.com/documentation/v3/performing-tests/workloads

All of those workloads are for logging in users and performing tests; none are related to bootstorms.

If you want to see how a bootstorm test is measured start reading at page 65 of this document (disclaimer: I wrote this document): www.vmware.com/files/pdf/vmw-emc-infrast...-solutions-guide.pdf

The GAVG value is the most important as that is disk latency measured at the vSphere host level. The disk array stats are important as well, but at the end of the day if the GAVG is acceptable and all desktops booted successfully it really doesn't matter what the array was doing.

Please Log in to join the conversation.

More
23 Apr 2012 05:54 #233

Hi Jason,

You are correct that VSI is designed to test user workloads, there are multiple things to check on a bootstorm fully static like you do: Making sure all machines are online or as we typically do: Boot all machines (optionally just configuring an idle pool) and logging on sessions at the same time. Reason we do this is simple; if you are ever in a disaster recovery scenario you don't only want to know how long it takes to boot all machines but also how long it takes to effectively log them on again and even more importantly: how the performance is while all machines are booting / logging on / working.

Please Log in to join the conversation.

  • merrist
  • merrist's Avatar Topic Author
23 Apr 2012 06:45 #234

vjason wrote:

No, core is just an empty workload that logs in users only.

www.loginvsi.com/documentation/v3/performing-tests/workloads

All of those workloads are for logging in users and performing tests; none are related to bootstorms.

If you want to see how a bootstorm test is measured start reading at page 65 of this document (disclaimer: I wrote this document): www.vmware.com/files/pdf/vmw-emc-infrast...-solutions-guide.pdf

The GAVG value is the most important as that is disk latency measured at the vSphere host level. The disk array stats are important as well, but at the end of the day if the GAVG is acceptable and all desktops booted successfully it really doesn't matter what the array was doing.
hi,Jason

so,I know the Login VSI is not surpport testing the bootstorm.
The bootstorm means that we power on VMs at the same time.
But,I think the bootsorm has an another explain.
The boostorm means that the users log to on VMs for user.
So,Is the Login VSI surpport the bootstorm of this explain?

Please Log in to join the conversation.

More
23 Apr 2012 06:47 #235

Yes!

Please Log in to join the conversation.

  • merrist
  • merrist's Avatar Topic Author
23 Apr 2012 07:01 #236

markp wrote:

Yes!

hi,markp

so,our Login VSI is surpport this bootstorm.and do we use the 【select workload Core + VSITimers】mode to surpport?

Please Log in to join the conversation.

More
23 Apr 2012 07:04 #237

merrist wrote:

markp wrote:
Yes!

hi,markp

so,our Login VSI is surpport this bootstorm.and do we use the 【select workload Core + VSITimers】mode to surpport?

Yes, but I recommend selecting the workload: Medium since this is a more realistic scenario.

Please Log in to join the conversation.

  • merrist
  • merrist's Avatar Topic Author
23 Apr 2012 07:20 #238

markp wrote:

merrist wrote:
markp wrote:
Yes!

hi,markp

so,our Login VSI is surpport this bootstorm.and do we use the 【select workload Core + VSITimers】mode to surpport?

Yes, but I recommend selecting the workload: Medium since this is a more realistic scenario.

The bootstorm just is happened.The reason is many login operation have been executed.
so,I think the bootstorm just need to workload Core + VSITimers:just have a login operation.

why do you recommend selecting the workload: Medium?
because i think the realistic scenario just a login operation.

Please Log in to join the conversation.

  • jeffq
  • jeffq's Avatar
25 Apr 2012 14:17 #260

so i've been interested in testing a boot storm too, not necessarily relying on loginVSI to give me stats, i can collect the details out of esxtop but I want loginVSI to orchestrate the simultaneous reboot of 250 desktops and then perform a login.

i see how i can kick off a custom test that logs into the desktops and issues a reboot but at that point loginVSI test would end, right? is there a way to use loginVSI to initiate a reset of all 250 desktops and then log in?

or am i barking up the wrong tree?

Please Log in to join the conversation.

More
27 Apr 2012 18:41 #265

Jeff,

I suggest scripting the reboot and the start of the VSI test. I do something similar for Project VRC although we do not initiate a test right after the reboot. We intentionally have a delay between the reboot and the start of the test.

Let me know if you need any help or pointers regarding the script.

Please Log in to join the conversation.

  • jeffq
  • jeffq's Avatar
27 Apr 2012 18:46 #268

so what I ended up doing was creating a reboot script and placing that on my loginvsi share, then adding it to the workload customization for Core.

then i kicked off a core test to reboot all of the desktops.

the only problem I had doing this was that the Core workload test wouldn't execute the script until i saved my script into the global workload customization, it wasn't being picked up when it was in core specifically.. I checked the test config files in the vsi share and it showed my script in one of the Core ini files(can't remember which exactly).

either way though I did get the data I wanted to measure so i'm happy, thanks.

Please Log in to join the conversation.

Start Delivering the Best End User Experience Today

Request a Demo

Login VSI, Inc.

3945 Freedom Circle
Suite 670
Santa Clara, CA 95054

Phone: +1 408 899 7418

Login VSI B.V.

De Entree 85
1101 BH Amsterdam
The Netherlands

Phone: +31 20 705 1200