[THIN] Re: Video Performance

  • From: Chris Grecsek <chris@xxxxxxxxxxxxxxxxxxxx>
  • To: "'thin@xxxxxxxxxxxxx'" <thin@xxxxxxxxxxxxx>
  • Date: Tue, 22 Jul 2008 08:53:52 -0700

Our first "test" was the subjective multi-media experience test and we were 
surprised to see that the multi-media experience was "the same" on the VM CSG 
and on the physical CSG box. At that point we decided to get a bit more 
scientific/gather concrete evidence to confirm that result via the performance 
counters.

The test was late in the day so there were maybe 70 users on the VM CSG and 
yes, we only had the single user on the physical CSG. They physical CSG was a 
modern dual proc, dual core box w/4gig or ram so it wasn't a gimped box in any 
way. Actually we used the most powerful spare box we could find just to try to 
make the physical test box look as good as it could.

-----------------------------
chris grecsek | centered networks | t:415-294-7776 | f:415-294-7772 | 
chris@xxxxxxxxxxxxxxxxxxxx<mailto:chris@xxxxxxxxxxxxxxxxxxxx>
________________________________
From: thin-bounce@xxxxxxxxxxxxx [mailto:thin-bounce@xxxxxxxxxxxxx] On Behalf Of 
Steve Greenberg
Sent: Monday, July 21, 2008 11:35 PM
To: thin@xxxxxxxxxxxxx
Subject: [THIN] Re: Video Performance

That is good feedback, thanks for coming back and sharing the results. I 
wonder, on the physical box, aside from counters, did you subjectively 
experience better multi-media performance? You said that the test box only had 
one session on it, I have to think it performed better than the VM with 400 
users?!?




Steve Greenberg

[cid:image003.jpg@01C8EBD8.77582DB0]Thin Client Computing

34522 N. Scottsdale Rd D8453

Scottsdale, AZ 85266

(602) 432-8649

www.thinclient.net
steveg@xxxxxxxxxxxxxx

________________________________
From: thin-bounce@xxxxxxxxxxxxx [mailto:thin-bounce@xxxxxxxxxxxxx] On Behalf Of 
Chris Grecsek
Sent: Monday, July 21, 2008 10:38 PM
To: 'thin@xxxxxxxxxxxxx'
Subject: [THIN] Re: Video Performance

I thought I'd share the results of our testing in comparing video performance 
between a VM WI/CSG and a physical WI/CSG. We measured a number of counters 
with respect to session latency, bandwidth usage, etc. and it turns out that 
our performance/metrics was just about identical on the VM as on the physical 
machine. The VM CSG still had a number of active sessions and the physical only 
had a single, test session running on it.

We were able to isolate the performance issue to the CSG - we confirmed that 
the WI is not part of the degradation.

So, while this was good in that we now have a definitive answer in that no 
matter how much available bandwidth we have to help improve video quality, it 
won't get any better as long as we're connecting through a CSG. I guess it's 
time to get an eval of the Netscaler products.

Thanks!
Chris

-----------------------------
chris grecsek | centered networks | t:415-294-7776 | f:415-294-7772 | 
chris@xxxxxxxxxxxxxxxxxxxx<mailto:chris@xxxxxxxxxxxxxxxxxxxx>
________________________________
From: thin-bounce@xxxxxxxxxxxxx [mailto:thin-bounce@xxxxxxxxxxxxx] On Behalf Of 
Steve Greenberg
Sent: Friday, July 18, 2008 4:26 PM
To: thin@xxxxxxxxxxxxx
Subject: [THIN] Re: Video Performance

NIC is I/O and in a virtualized environment it has significant overhead in that 
a software layer is added to virtualize the NIC in a way that slows it down 
quite a bit. I would suggest, in general, that with 400 concurrent connections 
you have exceed what a VMWare based CSG can effectively handle, probably by 
quite a bit actually. The quick and easy proof is to spin up a CSG on a raw box 
and check the different in response times. The overhead in this case is not 
going to show in monitored utilization but more in the user experience of 
latency on the rich media stuff you are referring to. It is not a matter of 
utilization of VMWare resources as much as the fact that the virtualization I/O 
translation layer is simply too inefficient. Stay tuned- later this year Intel 
promises to have hardware NICS that are virtual machine aware...

The Netscaler is the product I am referring to. If you buy Netscaler Enterprise 
it includes a feature called Access Gateway. However, you can also buy the same 
product with only the gateway feature and it is called Access Gateway Enteprise 
Edition. It is very confusing, but in the end they are the same box with 
different software licenses loaded!!




Steve Greenberg

[cid:image003.jpg@01C8EBD8.77582DB0]Thin Client Computing

34522 N. Scottsdale Rd D8453

Scottsdale, AZ 85266

(602) 432-8649

www.thinclient.net
steveg@xxxxxxxxxxxxxx

________________________________
From: thin-bounce@xxxxxxxxxxxxx [mailto:thin-bounce@xxxxxxxxxxxxx] On Behalf Of 
Chris Grecsek
Sent: Friday, July 18, 2008 1:50 PM
To: 'thin@xxxxxxxxxxxxx'
Subject: [THIN] Re: Video Performance

Excellent feedback - thanks Steve! When you say VMware is bad with I/O do you 
mean disk or NIC or both? The perfmon counters we're running on the VMs show 
very little utilization of disk or NIC so I guess we'll just see how it goes.

We do have the WI and CSG on two separate VMs. We have about 400 concurrent 
connections on this WI/CSG set.

You also mentioned the Netscaler Access Gateway as a possible upgrade path from 
the WI/CSG. I'm doing some research but I must say, Citrix's product naming 
bonanza makes me not 100% sure I'm looking at the correct product. I found 
information on the Access Gateway Enterprise Edition (AGEE) but is that a 
different product than the Netscaler Access Gateway? Was there a specific 
feature of the product that you think would be making the difference or is it 
just the fact that the device is designed to handle/manage that specific type 
of traffic/load?


-----------------------------
chris grecsek | centered networks | t:415-294-7776 | f:415-294-7772 | 
chris@xxxxxxxxxxxxxxxxxxxx<mailto:chris@xxxxxxxxxxxxxxxxxxxx>
________________________________
From: thin-bounce@xxxxxxxxxxxxx [mailto:thin-bounce@xxxxxxxxxxxxx] On Behalf Of 
Steve Greenberg
Sent: Friday, July 18, 2008 1:07 PM
To: thin@xxxxxxxxxxxxx
Subject: [THIN] Re: Video Performance

Now that we know it is running on VMWare that could be the bottleneck right 
there- CSG is I/O intensive because of the nature of what it is doing as a 
proxy, this is the one area that hypervisor virtualization is still very weak 
at....




Steve Greenberg

[cid:image003.jpg@01C8EBD8.77582DB0]Thin Client Computing

34522 N. Scottsdale Rd D8453

Scottsdale, AZ 85266

(602) 432-8649

www.thinclient.net
steveg@xxxxxxxxxxxxxx

________________________________
From: thin-bounce@xxxxxxxxxxxxx [mailto:thin-bounce@xxxxxxxxxxxxx] On Behalf Of 
Chris Grecsek
Sent: Friday, July 18, 2008 11:21 AM
To: 'thin@xxxxxxxxxxxxx'
Subject: [THIN] Re: Video Performance

Hi Joe,

Thanks for the speedy response!

We're not using any LBs - there are no network devices other than a standard 
PIX.

The WI and CSGs are both VMs on ESX 3.5. Based on your question I think we'll 
setup a WI/CSG on a standalone server "just to see" even though we're not 
seeing high utilization.

At this point we have not isolated the WI or CSG. We're going to setup an 
additional site on the WI that will not use the CSG so that should achieve the 
isolation.

We've done quite a bit of testing with the speedscreen/multimedia stuff - I 
think we've tried every conceivable combination in the policies and farm 
settings . I believe the policies should apply the same whether we come in via 
the CSG/WI or the PN.
            Image acceleration using lossy compression: enabled; compression 
level: do not use lossy compression; speedscreen progressive display:disabled
            Speedscreen browser acceleration: off; speedscreen flash: enabled - 
all connections; speedscreen multimedia: enabled - buffer = 10.



Thanks again,
Chris
-----------------------------
chris grecsek | centered networks | t:415-294-7776 | f:415-294-7772 | 
chris@xxxxxxxxxxxxxxxxxxxx<mailto:chris@xxxxxxxxxxxxxxxxxxxx>
________________________________
From: thin-bounce@xxxxxxxxxxxxx [mailto:thin-bounce@xxxxxxxxxxxxx] On Behalf Of 
Joe Shonk
Sent: Friday, July 18, 2008 10:35 AM
To: thin@xxxxxxxxxxxxx
Subject: [THIN] Re: Video Performance

It could be a couple of things....  Have you tried removing CSG from the 
equation and isolate if its WI or CSG?  Is your CSG box running in a Virtual 
Machine?  Do you have a LB in front of your WI and/or CSG box?  How are you 
Applying the Ctx Policies for the SpeedScreen/multimedia.

Joe

From: thin-bounce@xxxxxxxxxxxxx [mailto:thin-bounce@xxxxxxxxxxxxx] On Behalf Of 
Chris Grecsek
Sent: Friday, July 18, 2008 10:24 AM
To: 'thin@xxxxxxxxxxxxx'
Subject: [THIN] Video Performance

We provide published desktops via Citrix/terminal services delivered via the 
internet. All users come in through the web interface/CSG via high speed, 
burstable (up to 100mb) connections. Most of the branch offices are <50 users 
so connectivity/maxxing out the pipe is not an issue. We don't use any 
bandwidth restriction policies - users can use as much bandwidth as they'd 
like. We have bandwidth monitoring to look at all traffic and have our 
firewalls setup so that only Citrix traffic is using the burstable connection. 
Latency is sub 20ms and we have no packet loss.

Like everyone else using Citrix we've had challenges with video performance on 
the published desktop (Windows media player works great - Flash/other codecs, 
not so much). We had an interesting discovery today and I was curious if anyone 
could answer why/how this happens...when we connect to a published desktop via 
the WI/CSG (from one of our branch offices) video performance is bad. We ran a 
test today from within the datacenter where we connected to a published 
desktop, locally (from a laptop at the DC), using the Program Neighborhood as 
opposed to coming in over the WI/CSG and low and behold, video performance was 
excellent.

When we look at the amount of session bandwidth being used it's nearly the same 
when coming in via the WI/CSG as it is when testing at the datacenter. So I 
guess the question is, is there something with the way the WI/CSG works that 
would be restricting/degrading the video performance (we ran performance 
counters on the WI/CSG and it doesn't seem to be working very hard), is there a 
difference between using the Program Neighborhood verses the web client, etc.?

We're on Citrix 4.5, latest rollups, latest version of the client, etc.

If anyone has any ideas, it would be much appreciated.

Thank you!

Chris

-----------------------------
chris grecsek | centered networks | t:415-294-7776 | f:415-294-7772 | 
chris@xxxxxxxxxxxxxxxxxxxx<mailto:chris@xxxxxxxxxxxxxxxxxxxx>

GIF image

JPEG image

Other related posts: