Re: Performance question
My SLES 10.1 zLinux servers have been notorious for not making much use of the swap space, even when we lower the main (virtualized) in the USER DIRECTORY. We may have something just not configured optimally somewhere, but our luck has gone with adding main Linux memory and subsequently having to add ZVM memory. David Dean Information Systems *bcbstauthorized* From: The IBM z/VM Operating System [mailto:[EMAIL PROTECTED] On Behalf Of Mary Anne Matyaz Sent: Monday, September 29, 2008 11:31 AM To: IBMVM@LISTSERV.UARK.EDU Subject: Performance question If your VM system is at 101% memory usage, and you are overcommitted by about 14%, is it worthwhile to add a vdisk to a linux for swap space, or better just to add main memory to the linux? MA (Looking for opinions, thoughts, rationalizations, whatever. :) Please see the following link for the BlueCross BlueShield of Tennessee E-mail disclaimer: http://www.bcbst.com/email_disclaimer.shtm
Re: Performance question
Best practices is to use Vdisk for swap, and reduce linux virtual machine sizes - not to buy more REAL z/VM memory unless you really need it. 101% memory useage means almost nothing. It is not relevant to performance or capacity, and thus shouldn't have business decisions or performance decisions decided based on that number. Dean, David (I/S) wrote: My SLES 10.1 zLinux servers have been notorious for not making much use of the swap space, even when we lower the main (virtualized) in the USER DIRECTORY. We may have something just not configured optimally somewhere, but our luck has gone with adding main Linux memory and subsequently having to add ZVM memory. David Dean Information Systems *bcbstauthorized* From: The IBM z/VM Operating System [mailto:[EMAIL PROTECTED] On Behalf Of Mary Anne Matyaz Sent: Monday, September 29, 2008 11:31 AM To: IBMVM@LISTSERV.UARK.EDU Subject: Performance question If your VM system is at 101% memory usage, and you are overcommitted by about 14%, is it worthwhile to add a vdisk to a linux for swap space, or better just to add main memory to the linux? MA (Looking for opinions, thoughts, rationalizations, whatever. :) Please see the following link for the BlueCross BlueShield of Tennessee E-mail disclaimer: http://www.bcbst.com/email_disclaimer.shtm
Re: Performance question
Can you give me any insight into making more use of the swap; see below? We have had systems significantly speed up after adding memory, even though swap was not being used. I realize that swap / disk is going to be slower, but with cache and buffers shouldn't it be at least close? lnx057:~ # free -l total used free sharedbuffers cached Mem:763208 758572 4636 0 77092 592676 Low:763208 758572 4636 High:0 0 0 -/+ buffers/cache: 88804 674404 Swap: 10798963001079596 Thank you for any help. David Dean Information Systems *bcbstauthorized* -Original Message- From: The IBM z/VM Operating System [mailto:[EMAIL PROTECTED] On Behalf Of Barton Robinson Sent: Monday, September 29, 2008 1:23 PM To: IBMVM@LISTSERV.UARK.EDU Subject: Re: Performance question Best practices is to use Vdisk for swap, and reduce linux virtual machine sizes - not to buy more REAL z/VM memory unless you really need it. 101% memory useage means almost nothing. It is not relevant to performance or capacity, and thus shouldn't have business decisions or performance decisions decided based on that number. Dean, David (I/S) wrote: My SLES 10.1 zLinux servers have been notorious for not making much use of the swap space, even when we lower the main (virtualized) in the USER DIRECTORY. We may have something just not configured optimally somewhere, but our luck has gone with adding main Linux memory and subsequently having to add ZVM memory. David Dean Information Systems *bcbstauthorized* From: The IBM z/VM Operating System [mailto:[EMAIL PROTECTED] On Behalf Of Mary Anne Matyaz Sent: Monday, September 29, 2008 11:31 AM To: IBMVM@LISTSERV.UARK.EDU Subject: Performance question If your VM system is at 101% memory usage, and you are overcommitted by about 14%, is it worthwhile to add a vdisk to a linux for swap space, or better just to add main memory to the linux? MA (Looking for opinions, thoughts, rationalizations, whatever. :) Please see the following link for the BlueCross BlueShield of Tennessee E-mail disclaimer: http://www.bcbst.com/email_disclaimer.shtm Please see the following link for the BlueCross BlueShield of Tennessee E-mail disclaimer: http://www.bcbst.com/email_disclaimer.shtm
Re: Performance question
I know I'm probably going to regret this, but, how can that be? I said VM memory usage, right? Not Linux MA On Mon, Sep 29, 2008 at 1:23 PM, Barton Robinson [EMAIL PROTECTED] wrote: Best practices is to use Vdisk for swap, and reduce linux virtual machine sizes - not to buy more REAL z/VM memory unless you really need it. 101% memory useage means almost nothing. It is not relevant to performance or capacity, and thus shouldn't have business decisions or performance decisions decided based on that number.
Re: Performance question
It appears there's alot of cache usage. What's running on this machine? -- Rich Smrcina VM Assist, Inc. Phone: 414-491-6001 Ans Service: 360-715-2467 rich.smrcina at vmassist.com http://www.linkedin.com/in/richsmrcina Catch the WAVV! http://www.wavv.org WAVV 2009 - Orlando, FL - May 15-19, 2009 Dean, David (I/S) wrote: Can you give me any insight into making more use of the swap; see below? We have had systems significantly speed up after adding memory, even though swap was not being used. I realize that swap / disk is going to be slower, but with cache and buffers shouldn't it be at least close? lnx057:~ # free -l total used free sharedbuffers cached Mem:763208 758572 4636 0 77092 592676 Low:763208 758572 4636 High:0 0 0 -/+ buffers/cache: 88804 674404 Swap: 10798963001079596 Thank you for any help. David Dean Information Systems *bcbstauthorized* -Original Message- From: The IBM z/VM Operating System [mailto:[EMAIL PROTECTED] On Behalf Of Barton Robinson Sent: Monday, September 29, 2008 1:23 PM To: IBMVM@LISTSERV.UARK.EDU Subject: Re: Performance question Best practices is to use Vdisk for swap, and reduce linux virtual machine sizes - not to buy more REAL z/VM memory unless you really need it. 101% memory useage means almost nothing. It is not relevant to performance or capacity, and thus shouldn't have business decisions or performance decisions decided based on that number. Dean, David (I/S) wrote: My SLES 10.1 zLinux servers have been notorious for not making much use of the swap space, even when we lower the main (virtualized) in the USER DIRECTORY. We may have something just not configured optimally somewhere, but our luck has gone with adding main Linux memory and subsequently having to add ZVM memory. David Dean Information Systems *bcbstauthorized* From: The IBM z/VM Operating System [mailto:[EMAIL PROTECTED] On Behalf Of Mary Anne Matyaz Sent: Monday, September 29, 2008 11:31 AM To: IBMVM@LISTSERV.UARK.EDU Subject: Performance question If your VM system is at 101% memory usage, and you are overcommitted by about 14%, is it worthwhile to add a vdisk to a linux for swap space, or better just to add main memory to the linux? MA (Looking for opinions, thoughts, rationalizations, whatever. :) Please see the following link for the BlueCross BlueShield of Tennessee E-mail disclaimer: http://www.bcbst.com/email_disclaimer.shtm Please see the following link for the BlueCross BlueShield of Tennessee E-mail disclaimer: http://www.bcbst.com/email_disclaimer.shtm
Re: Performance question
It is predominately a file server +600GIG. The OS is SUSE 10.1 (Novell). David Dean Information Systems *bcbstauthorized* -Original Message- From: The IBM z/VM Operating System [mailto:[EMAIL PROTECTED] On Behalf Of Rich Smrcina Sent: Monday, September 29, 2008 12:46 PM To: IBMVM@LISTSERV.UARK.EDU Subject: Re: Performance question It appears there's alot of cache usage. What's running on this machine? -- Rich Smrcina VM Assist, Inc. Phone: 414-491-6001 Ans Service: 360-715-2467 rich.smrcina at vmassist.com http://www.linkedin.com/in/richsmrcina Catch the WAVV! http://www.wavv.org WAVV 2009 - Orlando, FL - May 15-19, 2009 Dean, David (I/S) wrote: Can you give me any insight into making more use of the swap; see below? We have had systems significantly speed up after adding memory, even though swap was not being used. I realize that swap / disk is going to be slower, but with cache and buffers shouldn't it be at least close? lnx057:~ # free -l total used free sharedbuffers cached Mem:763208 758572 4636 0 77092 592676 Low:763208 758572 4636 High:0 0 0 -/+ buffers/cache: 88804 674404 Swap: 10798963001079596 Thank you for any help. David Dean Information Systems *bcbstauthorized* -Original Message- From: The IBM z/VM Operating System [mailto:[EMAIL PROTECTED] On Behalf Of Barton Robinson Sent: Monday, September 29, 2008 1:23 PM To: IBMVM@LISTSERV.UARK.EDU Subject: Re: Performance question Best practices is to use Vdisk for swap, and reduce linux virtual machine sizes - not to buy more REAL z/VM memory unless you really need it. 101% memory useage means almost nothing. It is not relevant to performance or capacity, and thus shouldn't have business decisions or performance decisions decided based on that number. Dean, David (I/S) wrote: My SLES 10.1 zLinux servers have been notorious for not making much use of the swap space, even when we lower the main (virtualized) in the USER DIRECTORY. We may have something just not configured optimally somewhere, but our luck has gone with adding main Linux memory and subsequently having to add ZVM memory. David Dean Information Systems *bcbstauthorized* From: The IBM z/VM Operating System [mailto:[EMAIL PROTECTED] On Behalf Of Mary Anne Matyaz Sent: Monday, September 29, 2008 11:31 AM To: IBMVM@LISTSERV.UARK.EDU Subject: Performance question If your VM system is at 101% memory usage, and you are overcommitted by about 14%, is it worthwhile to add a vdisk to a linux for swap space, or better just to add main memory to the linux? MA (Looking for opinions, thoughts, rationalizations, whatever. :) Please see the following link for the BlueCross BlueShield of Tennessee E-mail disclaimer: http://www.bcbst.com/email_disclaimer.shtm Please see the following link for the BlueCross BlueShield of Tennessee E-mail disclaimer: http://www.bcbst.com/email_disclaimer.shtm Please see the following link for the BlueCross BlueShield of Tennessee E-mail disclaimer: http://www.bcbst.com/email_disclaimer.shtm
Re: Performance question
Dean, David (I/S) wrote: My SLES 10.1 zLinux servers have been notorious for not making much use of the swap space, even when we lower the main (virtualized) in the USER DIRECTORY. Linux uses extra memory for file cache buffers. When the kernel detects memory is short, it just allocates less of them and spends more time accessing the disk. Linux thinks it is its own virtual memory manager. It does not, one imagines, co-operate optimally w/r/t VM. -- Jack J. Woehr# Self-delusion is http://www.well.com/~jax # half the battle! http://www.softwoehr.com # - Zippy the Pinhead
Re: Performance question
Thanks to all for some really good input. So the tuning legend that the Linux should just touch swap is true? But if Linux is going to eat all for file caching, would it not Always eventually swap? Thanks again. Anything that is not a mystery is guesswork. David Dean Information Systems *bcbstauthorized* From: The IBM z/VM Operating System [mailto:[EMAIL PROTECTED] On Behalf Of Jack Woehr Sent: Monday, September 29, 2008 12:55 PM To: IBMVM@LISTSERV.UARK.EDU Subject: Re: Performance question Dean, David (I/S) wrote: My SLES 10.1 zLinux servers have been notorious for not making much use of the swap space, even when we lower the main (virtualized) in the USER DIRECTORY. Linux uses extra memory for file cache buffers. When the kernel detects memory is short, it just allocates less of them and spends more time accessing the disk. Linux thinks it is its own virtual memory manager. It does not, one imagines, co-operate optimally w/r/t VM. -- Jack J. Woehr# Self-delusion is http://www.well.com/~jax # half the battle! http://www.softwoehr.com # - Zippy the Pinhead Please see the following link for the BlueCross BlueShield of Tennessee E-mail disclaimer: http://www.bcbst.com/email_disclaimer.shtm
Re: Performance question
Dean, David (I/S) wrote: Thanks to all for some really good input. So the tuning legend that the Linux should just touch swap is true? But if Linux is going to eat all for file caching, would it not Always eventually swap? Swap and file caching are two sides of the same thing. In Solaris, they are the same thing, not quite so in Linux. Basically, Linux will use extra mem for file caching. It will always use *some* mem for file caching. It will use less if mem is constrained. Linux kernel for running on VM could be designed a little different from Linux kernel for PC in order to behave better with VM. I'm not sure to what extent it actually is different. -- Jack J. Woehr# Self-delusion is http://www.well.com/~jax # half the battle! http://www.softwoehr.com # - Zippy the Pinhead
Re: Performance question
On Monday, 09/29/2008 at 01:41 EDT, Jack Woehr [EMAIL PROTECTED] wrote: Linux kernel for running on VM could be designed a little different from Linux kernel for PC in order to behave better with VM. I'm not sure to what extent it actually is different. This is not just a z/VM problem. For any virtualization platform that overcommits memory, Linux memory usage will be problematic. I hope that, someday, Linux will have a generalized ability to sense its surroundings and constrain itself according to the wishes of the hypervisor. I.e. learn whether or not it is sharing the CPU, memory, and I/O, and know the relative value of each. For example, knowing that on System z, I/O is not a Bad Thing as it is in Intel (this is what drives Linux' fanatical use of cache - I/O is evil) would lead to different biases in the cache management subsystem. That bias could be further influenced by communication between the hypervisor and Linux (similar to CMM on z/VM). Alan Altmark z/VM Development IBM Endicott
Re: Performance question
z/VM Memory usage, what do you think it means? If a page of a virtual machine is in storage, but has not been referenced in 10 minutes, is that part of your percent used? Likely you don't know the answer and the source of your information doesn't either. So if that's the case, what information are you using to make decisions? Mary Anne Matyaz wrote: I know I'm probably going to regret this, but, how can that be? I said VM memory usage, right? Not Linux MA On Mon, Sep 29, 2008 at 1:23 PM, Barton Robinson [EMAIL PROTECTED] wrote: Best practices is to use Vdisk for swap, and reduce linux virtual machine sizes - not to buy more REAL z/VM memory unless you really need it. 101% memory useage means almost nothing. It is not relevant to performance or capacity, and thus shouldn't have business decisions or performance decisions decided based on that number.
Re: Performance question
If the page has not been referenced in 10 minutes, but is not paged out, I would expect it to be included in the 101%. Try not to focus so much on the extraneous info and address the question, if I am using a huge amount of memory, is it more helpful to use vdisk or guest memory? MA On Mon, Sep 29, 2008 at 3:32 PM, Barton Robinson [EMAIL PROTECTED] wrote: z/VM Memory usage, what do you think it means? If a page of a virtual machine is in storage, but has not been referenced in 10 minutes, is that part of your percent used? Likely you don't know the answer and the source of your information doesn't either. So if that's the case, what information are you using to make decisions? Mary Anne Matyaz wrote: I know I'm probably going to regret this, but, how can that be? I said VM memory usage, right? Not Linux MA On Mon, Sep 29, 2008 at 1:23 PM, Barton Robinson [EMAIL PROTECTED] wrote: Best practices is to use Vdisk for swap, and reduce linux virtual machine sizes - not to buy more REAL z/VM memory unless you really need it. 101% memory useage means almost nothing. It is not relevant to performance or capacity, and thus shouldn't have business decisions or performance decisions decided based on that number.
Re: Performance question
Alan Altmark wrote: On Monday, 09/29/2008 at 01:41 EDT, Jack Woehr [EMAIL PROTECTED] wrote: Linux kernel for running on VM could be designed a little different from Linux kernel for PC in order to behave better with VM. I'm not sure to what extent it actually is different. This is not just a z/VM problem. Didn't mean to imply it was. For any virtualization platform that overcommits memory, Linux memory usage will be problematic. I hope that, someday, Linux will have a generalized ability to sense its surroundings and constrain itself according to the wishes of the hypervisor. It may be more pratical to modify the kernel so that it implements some of its functionality directly in terms of what the hypervisor provides rather than by assuming it controls all memory. This is the dialectic of modern virtualization scheme ... VMWare is like z/VM, the guest knows nothing, vs. Xen, the guest is modified to support the hypervisor. The former is cleaner and more secure, the latter more efficient execution. -- Jack J. Woehr# Self-delusion is http://www.well.com/~jax # half the battle! http://www.softwoehr.com # - Zippy the Pinhead
Re: Performance question
Best practices is to use Vdisk for swap Mary Anne Matyaz wrote: If the page has not been referenced in 10 minutes, but is not paged out, I would expect it to be included in the 101%. Try not to focus so much on the extraneous info and address the question, if I am using a huge amount of memory, is it more helpful to use vdisk or guest memory? MA On Mon, Sep 29, 2008 at 3:32 PM, Barton Robinson [EMAIL PROTECTED] wrote: z/VM Memory usage, what do you think it means? If a page of a virtual machine is in storage, but has not been referenced in 10 minutes, is that part of your percent used? Likely you don't know the answer and the source of your information doesn't either. So if that's the case, what information are you using to make decisions? Mary Anne Matyaz wrote: I know I'm probably going to regret this, but, how can that be? I said VM memory usage, right? Not Linux MA On Mon, Sep 29, 2008 at 1:23 PM, Barton Robinson [EMAIL PROTECTED] wrote: Best practices is to use Vdisk for swap, and reduce linux virtual machine sizes - not to buy more REAL z/VM memory unless you really need it. 101% memory useage means almost nothing. It is not relevant to performance or capacity, and thus shouldn't have business decisions or performance decisions decided based on that number.
Re: Performance Question
I see messages like this from time to time on my VM Lpars that only run guest operating systems and have little to no CMS users. I think it may be because the VM scheduler can't really see what a transaction is inside the GOS. I think it may just be an artifact of VM having an earlier design point of support a lot of concurrent users that have a lot of think time, rather than a farm of GOSes. From: The IBM z/VM Operating System on behalf of Martin, Terry R. (CMS/CTR) (CTR) Sent: Wed 9/24/2008 10:54 PM To: IBMVM@LISTSERV.UARK.EDU Subject: Re: Performance Question The REAL processors are running about 85% per (5 IFLs) and the virtual processors for this Linux host are running an average of about 90% per (4 Logicals). There is no CPU queuing that I can tell. I am not running any of the products you mentioned. This may not be a bad indicator I am just trying to find out what it really means since the default shipped is 1.000 elapse time not sure if that means much! Thanks, Terry From: The IBM z/VM Operating System [mailto:[EMAIL PROTECTED] On Behalf Of David Kreuter Sent: Wednesday, September 24, 2008 9:03 PM To: IBMVM@LISTSERV.UARK.EDU Subject: Re: Performance Question hard to say for certain. Not a number I look at that often. How busy are the processors? Is there CPU queuing? The scheduler tries to classify around 80% of work as class 1; so fluctuations are rather common. Are you per chance running WebSphere, WPS or Domino servers? David Kreuter From: The IBM z/VM Operating System on behalf of Martin, Terry R. (CMS/CTR) (CTR) Sent: Wed 9/24/2008 3:42 PM To: IBMVM@LISTSERV.UARK.EDU Subject: [IBMVM] Performance Question Hi I am trying to get a handle on all of the z/VM tuning knobs if you will and have a couple of questions: I see the following during heavy processing on the particular Linux host: FCXPER315A Cl1 time slice 8.939 exceeds limit 1.000 (Q1=02 Qx=26) I am using the default alert of 1.000 that is set in the PTK hence the alert. My question, is the fact that elapsed time which a class 1 user can spend in the dispatch list being 8.939 a high number and what is it really telling me? Does it represent a potential bottle neck? Thank You, Terry Martin Lockheed Martin - Information Technology z/OS z/VM Systems - Performance and Tuning Cell - 443 632-4191 Work - 410 786-0386 [EMAIL PROTECTED]
Re: Performance Question
Hi I know there are some performance folks out there (Velocity, IBM (Omegamon)) any thoughts on this from you all? Thanks Terry -Original Message- From: The IBM z/VM Operating System [mailto:[EMAIL PROTECTED] On Behalf Of Quay, Jonathan (IHG) Sent: Thursday, September 25, 2008 7:19 AM To: IBMVM@LISTSERV.UARK.EDU Subject: Re: Performance Question I see messages like this from time to time on my VM Lpars that only run guest operating systems and have little to no CMS users. I think it may be because the VM scheduler can't really see what a transaction is inside the GOS. I think it may just be an artifact of VM having an earlier design point of support a lot of concurrent users that have a lot of think time, rather than a farm of GOSes. From: The IBM z/VM Operating System on behalf of Martin, Terry R. (CMS/CTR) (CTR) Sent: Wed 9/24/2008 10:54 PM To: IBMVM@LISTSERV.UARK.EDU Subject: Re: Performance Question The REAL processors are running about 85% per (5 IFLs) and the virtual processors for this Linux host are running an average of about 90% per (4 Logicals). There is no CPU queuing that I can tell. I am not running any of the products you mentioned. This may not be a bad indicator I am just trying to find out what it really means since the default shipped is 1.000 elapse time not sure if that means much! Thanks, Terry From: The IBM z/VM Operating System [mailto:[EMAIL PROTECTED] On Behalf Of David Kreuter Sent: Wednesday, September 24, 2008 9:03 PM To: IBMVM@LISTSERV.UARK.EDU Subject: Re: Performance Question hard to say for certain. Not a number I look at that often. How busy are the processors? Is there CPU queuing? The scheduler tries to classify around 80% of work as class 1; so fluctuations are rather common. Are you per chance running WebSphere, WPS or Domino servers? David Kreuter From: The IBM z/VM Operating System on behalf of Martin, Terry R. (CMS/CTR) (CTR) Sent: Wed 9/24/2008 3:42 PM To: IBMVM@LISTSERV.UARK.EDU Subject: [IBMVM] Performance Question Hi I am trying to get a handle on all of the z/VM tuning knobs if you will and have a couple of questions: I see the following during heavy processing on the particular Linux host: FCXPER315A Cl1 time slice 8.939 exceeds limit 1.000 (Q1=02 Qx=26) I am using the default alert of 1.000 that is set in the PTK hence the alert. My question, is the fact that elapsed time which a class 1 user can spend in the dispatch list being 8.939 a high number and what is it really telling me? Does it represent a potential bottle neck? Thank You, Terry Martin Lockheed Martin - Information Technology z/OS z/VM Systems - Performance and Tuning Cell - 443 632-4191 Work - 410 786-0386 [EMAIL PROTECTED]
Re: Performance Question
On Thu, Sep 25, 2008 at 2:04 PM, Martin, Terry R. (CMS/CTR) (CTR) [EMAIL PROTECTED] wrote: I know there are some performance folks out there (Velocity, IBM (Omegamon)) any thoughts on this from you all? The transaction as observed by CP relates to applications running in CMS. It has no meaning in the context of the transactions in the application running in Linux. However, when the virtual machine is using minimal CPU resources and CP still does not recognize transactions, you may conclude that the application is polling and thus hurting scalability (since memory management is based on dropping from queue when idle). Rob -- Rob van der Heij Velocity Software http://velocitysoftware.com/
Re: Performance Question
hard to say for certain. Not a number I look at that often. How busy are the processors? Is there CPU queuing? The scheduler tries to classify around 80% of work as class 1; so fluctuations are rather common. Are you per chance running WebSphere, WPS or Domino servers? David Kreuter From: The IBM z/VM Operating System on behalf of Martin, Terry R. (CMS/CTR) (CTR) Sent: Wed 9/24/2008 3:42 PM To: IBMVM@LISTSERV.UARK.EDU Subject: [IBMVM] Performance Question Hi I am trying to get a handle on all of the z/VM tuning knobs if you will and have a couple of questions: I see the following during heavy processing on the particular Linux host: FCXPER315A Cl1 time slice 8.939 exceeds limit 1.000 (Q1=02 Qx=26) I am using the default alert of 1.000 that is set in the PTK hence the alert. My question, is the fact that elapsed time which a class 1 user can spend in the dispatch list being 8.939 a high number and what is it really telling me? Does it represent a potential bottle neck? Thank You, Terry Martin Lockheed Martin - Information Technology z/OS z/VM Systems - Performance and Tuning Cell - 443 632-4191 Work - 410 786-0386 [EMAIL PROTECTED]
Re: Performance Question
The REAL processors are running about 85% per (5 IFLs) and the virtual processors for this Linux host are running an average of about 90% per (4 Logicals). There is no CPU queuing that I can tell. I am not running any of the products you mentioned. This may not be a bad indicator I am just trying to find out what it really means since the default shipped is 1.000 elapse time not sure if that means much! Thanks, Terry From: The IBM z/VM Operating System [mailto:[EMAIL PROTECTED] On Behalf Of David Kreuter Sent: Wednesday, September 24, 2008 9:03 PM To: IBMVM@LISTSERV.UARK.EDU Subject: Re: Performance Question hard to say for certain. Not a number I look at that often. How busy are the processors? Is there CPU queuing? The scheduler tries to classify around 80% of work as class 1; so fluctuations are rather common. Are you per chance running WebSphere, WPS or Domino servers? David Kreuter From: The IBM z/VM Operating System on behalf of Martin, Terry R. (CMS/CTR) (CTR) Sent: Wed 9/24/2008 3:42 PM To: IBMVM@LISTSERV.UARK.EDU Subject: [IBMVM] Performance Question Hi I am trying to get a handle on all of the z/VM tuning knobs if you will and have a couple of questions: I see the following during heavy processing on the particular Linux host: FCXPER315A Cl1 time slice 8.939 exceeds limit 1.000 (Q1=02 Qx=26) I am using the default alert of 1.000 that is set in the PTK hence the alert. My question, is the fact that elapsed time which a class 1 user can spend in the dispatch list being 8.939 a high number and what is it really telling me? Does it represent a potential bottle neck? Thank You, Terry Martin Lockheed Martin - Information Technology z/OS z/VM Systems - Performance and Tuning Cell - 443 632-4191 Work - 410 786-0386 [EMAIL PROTECTED]
Re: performance question
Huegel, Thomas [EMAIL PROTECTED] writes: Tom, I'm not sure I completely follow the question. I am not aware of any way to get just the cost of the SCEXIT from existing data. You could at least bound it by looking at total costs in the TCP/IP stack machine to understand what the costs may be. One thought is to add something to the exit that grabs perhaps QUERY TIME or something before at entering and leaving the exit. Bill Bitner - VM Performance Evaluation - IBM Endicott - 607-429-3286
Re: performance question
Title: RE: performance question Bill, I wasn't concerned with just the SCEXIT itself, if I thought that was the issue I would rewrite it in assembler. I was interested in the whole DIAL/DROP path. From the time TCP/IP first saw the connection until VSE/VTAM saw the terminal and how much overhead there was to keep dialing and dropping the same terminal. Then my next step will be to try and find what goes on in VSE with VTAM, CICS autoinstalls, lostterm clean ups ... Tom -Original Message- From: The IBM z/VM Operating System [mailto:IBMVM@LISTSERV.UARK.EDU]On Behalf Of Bill Bitner Sent: Thursday, October 12, 2006 3:31 PM To: IBMVM@LISTSERV.UARK.EDU Subject: Re: performance question Huegel, Thomas [EMAIL PROTECTED] writes: Tom, I'm not sure I completely follow the question. I am not aware of any way to get just the cost of the SCEXIT from existing data. You could at least bound it by looking at total costs in the TCP/IP stack machine to understand what the costs may be. One thought is to add something to the exit that grabs perhaps QUERY TIME or something before at entering and leaving the exit. Bill Bitner - VM Performance Evaluation - IBM Endicott - 607-429-3286 __ ella for Spam Control has removed VSE-List messages and set aside VM-List for me You can use it too - and it's FREE! http://www.ellaforspam.com
Re: performance question
Huegel, Thomas [EMAIL PROTECTED] writes: Thank you, I better understand your problem. There is no trivial way I know of to break this apart as you would like. My best suggestion is: - isolate this processing to a TCP/IP stack that does this alone - use either SCEXIT or cons logs to track rate of dial/drop. - record total usage of the TCP/IP stack from Performance Toolkit - correlate the CPU to dial/drop rate to get an average cost of cost in the TCP/IP stack - you could also perhaps create a dummy guest for test purposes and measure that part of it without skewing from other work. Ugly, but all I have to offer. And would give you ballpark numbers. Not sure you need any significant precision. Bill Bitner - VM Performance Evaluation - IBM Endicott - 607-429-3286
Re: performance question
Title: performance question Tom, although I couldn't open the stuff you tried to provide in your append, the data you sent me by separate mail gave me the required information to tell what's wrong: - Your RESET specifications are fine. This isevidenced by the report headers From 2006/04/25 00:14:00 To 2006/04/26 00:00:00 For 85560 Secs 23:46:00 i.e. you really get averages for the whole day, as expected. (I assume that the late start time in the example was due to PerfKit being activated only at 00:14.) - I assume that when you wrote '.. all my reports start at 12:01:00..' you referred to the 'by time' logs you printed. Their headers tell you that the averages shown are also for the whole day, so they're based on the correct reset time, but the detail lines start only with the sample periodending at 12:01. This is working as designed: What you see is the effect of a documented restriction that only allows PerfKit to hold a maximum of 720detail lines in its 'by time' log buffers, so when the chosen granularity for the logs is 1 minute you'll only see the detail lines for the last 12 hours (see description of the 'FC MONCOLL REDISP' command). To see detail lines for the whole 24 hours change the 'by time' interval to 2 minutes (FC SET BYTIME 2), or to any other higher value. If you always want to print logs for the whole day then I'd suggest to set the BYTIME interval to 10 or 15 minutes. That way the output size is reduced to something more easily manageable. Eginhard Jaeger - Original Message - From: Huegel, Thomas To: IBMVM@LISTSERV.UARK.EDU Sent: Wednesday, April 26, 2006 3:42 PM Subject: Re: performance question I tried this in my $profile.. I get this display. My report heading looks like this. BUT my reports still start at 12:00:00 and end at 23:59:00.. Any ideas what I may have wrong??? Thank you. Tom -Original Message-From: The IBM z/VM Operating System [mailto:[EMAIL PROTECTED]On Behalf Of Eginhard JaegerSent: Monday, April 24, 2006 9:05 AMTo: IBMVM@LISTSERV.UARK.EDUSubject: Re: performance question Can anybody tell me how to get reports from PERFKIT that start at 00:01:00 and go to 23:59:00 ? I have tried everything I can find and all my reports start at 12:01:00 and end at 23:59:00. What am I missing? The times when counters should be reset in PerfKit are defined by arguments of the command 'FC MONCOLL RESET ...'. Entering the command without further arguments will tell you what your current settings are, and my guess is you have defined a reset at noon. Note that each automated 'print' request (i.e. P-suffix after the time) will automatically reset the corresponding counters, i.e. it will become the start of a new reporting period unless it is superseded by a later explicit 'hh:mm:ssR_P' reset specification. For automatic resetting at 00:01 and printing of reports at 23:59 you'd need just the statement 'FC MONCOLL RESET 00:01:00R_P 23:59:00P' Eginhard Jaeger ella for Spam Control has removed 3853 VSE-List messages and set aside 2172 VM-List for meYou can use it too - and it's FREE!www.ellaforspam.com
Re: performance question
Title: performance question I tried this in my $profile.. I get this display. My report heading looks like this. BUT my reports still start at 12:00:00 and end at 23:59:00.. Any ideas what I may have wrong??? Thank you. Tom -Original Message-From: The IBM z/VM Operating System [mailto:[EMAIL PROTECTED]On Behalf Of Eginhard JaegerSent: Monday, April 24, 2006 9:05 AMTo: IBMVM@LISTSERV.UARK.EDUSubject: Re: performance question Can anybody tell me how to get reports from PERFKIT that start at 00:01:00 and go to 23:59:00 ? I have tried everything I can find and all my reports start at 12:01:00 and end at 23:59:00. What am I missing? The times when counters should be reset in PerfKit are defined by arguments of the command 'FC MONCOLL RESET ...'. Entering the command without further arguments will tell you what your current settings are, and my guess is you have defined a reset at noon. Note that each automated 'print' request (i.e. P-suffix after the time) will automatically reset the corresponding counters, i.e. it will become the start of a new reporting period unless it is superseded by a later explicit 'hh:mm:ssR_P' reset specification. For automatic resetting at 00:01 and printing of reports at 23:59 you'd need just the statement 'FC MONCOLL RESET 00:01:00R_P 23:59:00P' Eginhard Jaeger ella for Spam Control has removed 3853 VSE-List messages and set aside 2172 VM-List for meYou can use it too - and it's FREE!www.ellaforspam.com
Re: performance question
Title: performance question Can anybody tell me how to get reports from PERFKIT that start at 00:01:00 and go to 23:59:00 ? I have tried everything I can find and all my reports start at 12:01:00 and end at 23:59:00. What am I missing? The times when counters should be reset in PerfKit are defined by arguments of the command 'FC MONCOLL RESET ...'. Entering the command without further arguments will tell you what your current settings are, and my guess is you have defined a reset at noon. Note that each automated 'print' request (i.e. P-suffix after the time) will automatically reset the corresponding counters, i.e. it will become the start of a new reporting period unless it is superseded by a later explicit 'hh:mm:ssR_P' reset specification. For automatic resetting at 00:01 and printing of reports at 23:59 you'd need just the statement 'FC MONCOLL RESET 00:01:00R_P 23:59:00P' Eginhard Jaeger
