Groupe des Utilisateurs Altiris Suisses et Francophones Group

It is not the 1st, and probably not last, I got this kind of problems...

But we got a server with a “freezed” Web console (opening from an external IE8 on Windows 7 x64)
about once time per hour, during 3 to 15 mn. When “a long time” freeze, the console “reset” asking again for the Windows credential (Just cancel & refresh the page is OK)...

What’s happen: simple: Cycling the application pool, the W3WP kill himself, and restart a new process ID...

 Log Name:      System
Source:        Microsoft-Windows-WAS
Event ID:      5010
Task Category: None
Level:         Warning
Keywords:      Classic          
Description:
A process serving application pool 'Classic .NET AppPool' failed to respond to a ping. The process id was '13520'. 

Some our customers solve it creating its own “application pool”, but why ? With which options?
And why for some install, and not others?
How can we fix this definitively, and prevent for any further installations?

I see a few nice things in there:
https://www-secure.symantec.com/connect/blogs/creating-new-iis-application-pools-magic-solution-smp-7-slow-console-operations (Thanks Ludovic), but not all answers.
Should we simply create another application pool ?
But for which of the 78 current applications ?
What should we do with the 17 appplications keep under “DefaultAppPool”, and not freezing.
I create a Symantec case escalation (# 414-250-394) for getting confirmation.

We currently install CMS+SMS+AMS 7.1 on a Windows Server 2008 R2 (Build 7600)
IIS is 7.5.7600.16385

Should be a standard Windows installation... Using only one disk C.
Hosted on a standard ESX server.
Using an external SQL database 2008.

IIS activation and settings were defined as standard from the SIM installer (creating the “Classic Pool”?)

(hum, I put the JGP below in att. to see better)

How can we stop this mess ?

I do not find any significative KB related this problem:

• TECH14723 makes me adding the "Fixed nb of request" for 35'000 into the Recycling conditions, without result
• I do not see any entry in log corresponding the TECH22026.

I see only those in Altiris logs:

 Process: w3wp (14092)
Thread ID: 265
Module: w3wp.exe
Source: Altiris.NS.Server.GroupMessaging_Trace
Description: [w3wp.exe:/LM/W3SVC/1/ROOT/Altiris/Console-7-129457842206370590] <_lm_w3svc_1_root_altiris_console_7_129457842206370590_76e993ba7be2480da4e0de569b1b3fe1>:<ReliableMMF> Bad packet received (hash failure). 

I've created a new application pools for the following: Arellia ClientTaskServer InventoryRuleManagement NS PatchManagement SCSAppPool We're running 7.0. But this deals mainly with memory usage, not CPU? How much memory does the W3WP process use before it recycles? Is your CPU maxing out? I posed this quesiton about memory limitations in 7.1 here: https://www-secure.symantec.com/connect/forums/cms-71-mainly-smp-and-memory-usage-capabilities

Hello Pascal,

Split the core applications, and anything that turns to be a memory hog, which can be just about any web-apps in 7.0.

So here are the base candidates for the split:

• Arellia
• Altiris/NS/Agent
• ClitenTaskServer
• ClientTaskAgent
• InventoryRuleManagement

For more it really will come down to usage.

What you will find is that the new pools should start consuming memory and the sum of these pool should be well over the previous pool memory consumption.

I did that last week in production for a customer, and we turned a single w3wp process consuming ~750MiB to 4 process getting closer to ~1GiB.

This is down to .net and memory management. In theory (depending on the object type) a process cannot use more than 1.2 GiB however the limits can be reach much eariler, depending on the count and type of objects.

After log investigating, MySupport suggest a .Net repair, I will try & coming back there :)

apsnet_regiis -i

was not solving this problem :(

I do not find Arrelia, ClientTaskAgent, SCSAppPool (probably 7.1 difference vs 7.0)

Now here we are, after I restart default web site, in case of, all the same probably not required

.. let see ...

Not ITMS installation, but CMS+SMS+AMS:

Would prefer providing a text copy, but seems Symantec not single bad player in area of editors need to "BMP" screenshots, Microsoft join also...

Why the "Save list txt" not there any more, like in previous MMC ?? :(

Pascall Arellia is only avaliable if you have Mobile Management, and it´s not avaliable yet in 7.1 version :)

I did the same configuration for app pools in 7.1 as i previously did on 7.0 and got a better performance regarding CPU usage. Memory consuption was bigger, but no more 100% CPU.

Thanks the info Daniel, I also reproduce for the 1st time this problem on my own LAB, I will split also there.

A VMware machine with a single CPU. One of my colleague told me we should not put 2 CPU if not mandatory from application when ESX using, because can be more degrading perf, as must wait 2 cpu cycle availability from host... So I keep only 1 CPU...

It seems I got improvement on the 1st seing on the machine. I will spend more time on it in 2 weeks. We will see.

If I create 10 pool of applications of 8 applications, do I improve more performance ??? (I guess yes, if upper memory enough, and CPU?). We should build a stress test LAB for measuring such. Hey, who's got a stress test lab on ?

What is "Classic" vs "Integrated" ? I select "classic" because the previous one was same...

Yes, that is question of "pipeline"... I know the one for petrole, but ".Net" one?

I still have overload CPU, but without "cycling"...

 

SMP 7.1 can definitely support more than one CPU if that's what you are asking.

http://www.symantec.com/business/support/index?page=content&id=HOWTO9758

sorry my bad explain - For sure we can use more 1 cpu, and that is highly recommended :)

The question was: When using a VM (ESX 4.1), should be, or not, build 2 CPU machine ??? I receive an info can be more perf using only one, because of not "waiting" 2 cpu cycle from the host.

;-)

Pascall, that 2 cpu overhead is a legend from the ESX Admins, of course most services don´t actually use more then 1 CPU, as the code is build for that, but as Altiris uses multiple application pools i can´t see a reason not to use more then 1 CPU...

Specially when you have inventory processed and other system scheduled tasks.

I went to a very big "friendly discussion" with our virtualization team last week about that :) They gave me a 1 CPU vm for SMP, and i asked for 2.

Anyway i can go more technical if you need, also can gather some info from the Virtualization guys :)

Hello & thanks a lot Daniel: I can confirm you: on ESX server:

• Using local SQL install on a SMP 7.1: should provide 3 CPU, 8 GB RAM or more
• Using SMP 7.1 with external SQL: provide at least 2 CPU,

I can't explain you why, and I must say my VMware expert colleague told me "depending the application". So if your VMware team told you to configure only 1 CPU your Altiris SMP 7.1 host: just explain "Altiris" is one of the exception! This will not work correctly: application pool over cpu (W3W3P) with dead web session can be expected... (WAS recycling into windows Application logs up in this article)

Here the reference Symantec case number if need an "official" one: 414-250-394

You can just answer your VMware team, that "Altiris" is one of those exceptionnal applications, requiring 2 or more CPU.

Moreover, if plan on a production server, you must ask the reserve at least 6 GB with external SQL, 8GB with internal, and higher priority on 2 core CPU...

The App Pools is a good idea but not the fix for "But we got a server with a “freezed” Web console (opening from an external IE8 on Windows 7 x64) about once time per hour, during 3 to 15 mn. When “a long time” freeze, the console “reset” asking again for the Windows credential (Just cancel & refresh the page is OK)..."

Yes, the freeze & session reset, was because of a timeout, overloaded CPU. Adding a CPU remove the "timeout", and remove or reduce the "WAS" Cycling error. I suppose distributing more than one application pool for Altiris, also provides better distribution workload multiple CPU, but not sure about that.