Using complete CPU core instead thread

Hi, is it possible to configure informix to use the whole core per VP instead of only one thread using affinity ? Let's say that now we have cpu=6,aff=(0-5),noage Condidering that there are 2 threads per core, if I configure informix to use only one thread per core, would it use the whole core ? For example: cpu=6,aff=(0-10/2),noage or the other thread would be used sometimes by another application on the machine Or bring offline the every second cpu thread of the core that has a thread attached to informix VP ? Thank you.

Are you using Linux on Intel? In which case, hyper-threading manifests itself as two different core numbers (check /proc/cpuinfo) which you can use with affinity. However you'd need to test whether having two CPU VPs per core in this way is faster than one. If not, what platform (OS and hardware combination) are you using? Ben.

Hi Benjamin, I forgot those important details: I use Linux on Intel. Unfortunately I don't have where to test so I tried to ask here Thank you.

Hi, I'm not sure I completely understood the original question. Anyway, here's some (not too much in-depth) explanation that may be helpful: Informix has a homegrown implementation of multi threading. (This is mainly because at the time the Posix thread implementation first was not and then not widely available on the many different platforms where Informix was running (Sequent, Iris, Data General Unix, SINIX, to name a few). Furthermore, it is also of some advantage that this homegrown multi threading is non-preemptive as this allows thread switches at times when they are opportune and less costly.) An Informix Virtual Processor (VP) is a UNIX/Linux process (of the oninit binary) that normally has several threads, but runs only one of these threads at a time. Therefore, a single VP will utilize only a single CPU at any one time. If there are more CPUs available in the system, then the way to make Informix utilize them is to start more VPs, especially the so-called CPU VPs as they are doing most of the normal work by running the so-called "sqlexec" threads. With the threads running in a VP being of the homegrown multi threading implementation, the OS does not really know about them and therefore does not schedule them concurrently. The OS only knows about the VPs (that are processes) and schedules those. Bringing in the concept of Intel HyperThreading (and very similar also Simultaneous Multi Threading - SMT - on Power) does not change anything of the above. With HyperThreading enabled, the OS basically sees two CPUs for every processor core. Therefore it can schedule twice as many processes concurrently. In order to fully take advantage of this, one would have to re-configure Informix to run twice as many processes (i.e. VPs). Now, thing with HyperThreading (as well as SMT) is, that - obviously - it does not magically create twice as many "real CPUs". What it does is to offer the OS the possibility to schedule twice as many processes concurrently ... and then runs certain operations executed by two concurrent processes in parallel, taking advantage of arcane details of the processor and system architecture (instruction execution pipeline, etc.). Therefore, the two processes that are seemingly running concurrently on the two CPUs that the HyperThreading offers for the same core ... only run partially in parallel. How well this pseudo-parallelism works out (i.e. how much performance improvement it creates) depends a lot on how the application is implemented and therefore normally cannot be predicted with much accuracy. With an application like the Informix database server it may even considerably depend on the database applications. With all this, it hopefully becomes a bit clearer that and why: - a single Informix VP does not utilize two CPUs concurrently as it is a single process from the perspective of the OS. - the Informix affinity configuration cannot make a single VP to utilize two CPUs concurrently. (Binding a VP to one or several CPUs basically tells the OS to restrict the scheduling of the process to the specified CPUs.) - switching on HyperThreading does not automatically double the performance. - with HyperThreading active, Informix probably should be re-configured to run more (CPU) VPs in order to take advantage of the HyperThreading. - doubling the number of Informix VPs for HyperThreading probably is not the ideal configuration because the processes only run partially in parallel. Doubling the number of VPs may well overload the system and thus cause some negative effect on performance. - the ideal Informix configuration for a system with HyperThreading should be determined by careful tuning of the individual system using the intended workload. Regards, Martin -- Martin Fuerderer Informix Development Germany e-mail: martin.furderer@hcl.com HCL Technologies Germany GmbH Frankfurter Ring 17 80807 Munich, Germany http://www.hcltech.com/de --DISCLAIMER-- -------------------------------------------------------------------------------- ---------------------------- This document is intended for transmission to the named recipient only. If you are not that person, you should note that legal rights reside in this document and you are not authorized to access, read, disclose, copy, use or otherwise deal with it and any such actions are prohibited and may be unlawful. The views expressed in this document are not necessarily those of HCL Technologies Ltd. Notice is hereby given that no representation, contract or other binding obligation shall be created by this e-mail, which must be interpreted accordingly. Any representations, contractual rights or obligations shall be separately communicated in writing and signed in the original by a duly authorized officer of the relevant company. -------------------------------------------------------------------------------- ---------------------------- ________________________________ From: ids-bounces@iiug.org <ids-bounces@iiug.org> on behalf of R MICHAEL <mboncalo@gmail.com> Sent: Tuesday, February 20, 2018 14:58 To: ids@iiug.org Subject: Re: Using complete CPU core instead thread [40717] Hi Benjamin, I forgot those important details: I use Linux on Intel. Unfortunately I don't have where to test so I tried to ask here Thank you. ******************************************************************************* Forum Note: Use "Reply" to post a response in the discussion forum. ::DISCLAIMER:: -------------------------------------------------------------------------------- -------------------------------------------------------------------------------- -------------------------------------------------------------------------------- -------------------------------------- The contents of this e-mail and any attachment(s) are confidential and intended for the named recipient(s) only. E-mail transmission is not guaranteed to be secure or error-free as information could be intercepted, corrupted, lost, destroyed, arrive late or incomplete, or may contain viruses in transmission. The e mail and its contents (with or without referred errors) shall therefore not attach any liability on the originator or HCL or its affiliates. Views or opinions, if any, presented in this email are solely those of the author and may not necessarily reflect the views or opinions of HCL or its affiliates. Any form of reproduction, dissemination, copying, disclosure, modification, distribution and / or publication of this message without the prior written consent of authorized representative of HCL is strictly prohibited. If you have received this email in error please delete it and notify the sender immediately. Before opening any email and/or attachments, please check them for viruses and other defects. -------------------------------------------------------------------------------- -------------------------------------------------------------------------------- -------------------------------------------------------------------------------- --------------------------------------

Hi Martin, Thanks for your detailed response. Why I asked this question? Is because We have lots of CPUs available( real ones) and I can use only a small percentage because of the 3TB database is stored on only 2 disks and I can't take advantage of all those CPUs. So I wanted to use, somehow, the whole CPU core instead of a thread. I needed to know if I take offline every second CPU thread, the remaining online thread will have the computation power of the whole CPU core so that I could use it as Informix VP. Thank you, Michael.

Michael, It is better to Hyper-threading off in the BIOS then from the command line but this is how we do it. Here are some notes from our Informix Performance Tuning class on how to do this. To See CPU usage: cat /proc/cpuinfo Look for the CPUs that have the same "core id", you want to switch off the second of each pair And a script to disable it on our systems. ############################################################################# ## Module: @(#)informix.env 2.0 Date: 01/01/2015 ## Author: Lester Knutsen Email: lester@advancedatatools.com ## Advanced DataTools Corporation ############################################################################# ## Disable CPU 4-7 to turn off Hyperthreading echo 0 > /sys/devices/system/cpu/cpu4/online echo 0 > /sys/devices/system/cpu/cpu5/online echo 0 > /sys/devices/system/cpu/cpu6/online echo 0 > /sys/devices/system/cpu/cpu7/online ############################################################################# Make sure you are disabling the second of each pair... the cpu numbers may be different on your system. Make sure you test, depending on your system it may or may not be faster. One our training servers, with 4 Physical Cores and 8 Virtual Cores, running 8 oninit CPUVP's on the 4 Physical Cores can be faster then running 8 oninit CPUVP's on 8 Virtual Cores, it all depends on your workload. Regards - Lester On 2/20/18 5:28 PM, R MICHAEL wrote: > Hi Martin, > > Thanks for your detailed response. > Why I asked this question? Is because We have lots of CPUs available( real > ones) and I can use only a small percentage because of the 3TB database is > stored on only 2 disks and I can't take advantage of all those CPUs. So I > wanted to use, somehow, the whole CPU core instead of a thread. > I needed to know if I take offline every second CPU thread, the remaining > online thread will have the computation power of the whole CPU core so that I > could use it as Informix VP. > > Thank you, > Michael. > > > ******************************************************************************* > Forum Note: Use "Reply" to post a response in the discussion forum. > -- ______________________________________________________________________ Lester Knutsen lester@advancedatatools.com Advanced DataTools Corporation Voice: 703-256-0267 x102 Visit our Web page: http://www.advancedatatools.com ______________________________________________________________________

I agree with Lester that it is better to turn off HyperThreading rather than turning off individual CPUs (i.e. those additional CPUs that the OS shows as a result from the HyperThreading being active). If you have some other application(s) runnning on the system for which you want HyperThreading to be on ... then you may have no other choice than to individually turn off the "HyperThreading CPUs" of those cores that you want to give to the Informix Server. I'm not sure what exactly this then means in terms of process scheduling (from the perspective of the OS). I think you somehow will have to test. There are many factors that may influence performance, beginning with the Linux kernel version, and ending with the nature of the application workload. If your system is I/O bound because there is a bottle neck with those two disks containing 3 TB of data, then optimizing CPU utilization probably has very little effect on overall performance. If the Informix threads doing the CPU intensive work are waiting for I/O of data, then they will just do that - wait for the data (and in the meantime not burn CPU cycles in a useless manner, fortunately). Regards, Martin -- Martin Fuerderer Informix Development Germany e-mail: martin.furderer@hcl.com HCL Technologies Germany GmbH Frankfurter Ring 17 80807 Munich, Germany http://www.hcltech.com/de --DISCLAIMER-- -------------------------------------------------------------------------------- ---------------------------- This document is intended for transmission to the named recipient only. If you are not that person, you should note that legal rights reside in this document and you are not authorized to access, read, disclose, copy, use or otherwise deal with it and any such actions are prohibited and may be unlawful. The views expressed in this document are not necessarily those of HCL Technologies Ltd. Notice is hereby given that no representation, contract or other binding obligation shall be created by this e-mail, which must be interpreted accordingly. Any representations, contractual rights or obligations shall be separately communicated in writing and signed in the original by a duly authorized officer of the relevant company. -------------------------------------------------------------------------------- ---------------------------- ________________________________ From: ids-bounces@iiug.org <ids-bounces@iiug.org> on behalf of Lester Knutsen <lester@advancedatatools.com> Sent: Wednesday, February 21, 2018 20:26 To: ids@iiug.org Subject: Re: Using complete CPU core instead thread [40726] Michael, It is better to Hyper-threading off in the BIOS then from the command line but this is how we do it. Here are some notes from our Informix Performance Tuning class on how to do this. To See CPU usage: cat /proc/cpuinfo Look for the CPUs that have the same "core id", you want to switch off the second of each pair And a script to disable it on our systems. ############################################################################# ## Module: @(#)informix.env 2.0 Date: 01/01/2015 ## Author: Lester Knutsen Email: lester@advancedatatools.com ## Advanced DataTools Corporation ############################################################################# ## Disable CPU 4-7 to turn off Hyperthreading echo 0 > /sys/devices/system/cpu/cpu4/online echo 0 > /sys/devices/system/cpu/cpu5/online echo 0 > /sys/devices/system/cpu/cpu6/online echo 0 > /sys/devices/system/cpu/cpu7/online ############################################################################# Make sure you are disabling the second of each pair... the cpu numbers may be different on your system. Make sure you test, depending on your system it may or may not be faster. One our training servers, with 4 Physical Cores and 8 Virtual Cores, running 8 oninit CPUVP's on the 4 Physical Cores can be faster then running 8 oninit CPUVP's on 8 Virtual Cores, it all depends on your workload. Regards - Lester On 2/20/18 5:28 PM, R MICHAEL wrote: > Hi Martin, > > Thanks for your detailed response. > Why I asked this question? Is because We have lots of CPUs available( real > ones) and I can use only a small percentage because of the 3TB database is > stored on only 2 disks and I can't take advantage of all those CPUs. So I > wanted to use, somehow, the whole CPU core instead of a thread. > I needed to know if I take offline every second CPU thread, the remaining > online thread will have the computation power of the whole CPU core so that I > could use it as Informix VP. > > Thank you, > Michael. > > > ******************************************************************************* > Forum Note: Use "Reply" to post a response in the discussion forum. > -- ______________________________________________________________________ Lester Knutsen lester@advancedatatools.com Advanced DataTools Corporation Voice: 703-256-0267 x102 Visit our Web page: https://apac01.safelinks.protection.outlook.com/?url=http%3A%2F%2Fwww.advancedat atools.com&data=02%7C01%7Cmartin.furderer%40hcl.com%7C65f34282e7ae4164dc2208d579 611983%7C189de737c93a4f5a8b686f4ca9941912%7C0%7C0%7C636548380326823313&sdata=w9H IRN1ADeoV8rzM3UyjXvR6KOowB60EqSY%2FXiBql0k%3D&reserved=0 ______________________________________________________________________ ******************************************************************************* Forum Note: Use "Reply" to post a response in the discussion forum. ::DISCLAIMER:: -------------------------------------------------------------------------------- -------------------------------------------------------------------------------- -------------------------------------------------------------------------------- -------------------------------------- The contents of this e-mail and any attachment(s) are confidential and intended for the named recipient(s) only. E-mail transmission is not guaranteed to be secure or error-free as information could be intercepted, corrupted, lost, destroyed, arrive late or incomplete, or may contain viruses in transmission. The e mail and its contents (with or without referred errors) shall therefore not attach any liability on the originator or HCL or its affiliates. Views or opinions, if any, presented in this email are solely those of the author and may not necessarily reflect the views or opinions of HCL or its affiliates. Any form of reproduction, dissemination, copying, disclosure, modification, distribution and / or publication of this message without the prior written consent of authorized representative of HCL is strictly prohibited. If you have received this email in error please delete it and notify the sender immediately. Before opening any email and/or attachments, please check them for viruses and other defects. -------------------------------------------------------------------------------- -------------------------------------------------------------------------------- -------------------------------------------------------------------------------- --------------------------------------

Hi Lester, My ideea was to run the oninit processes on 10 cores by turning off every second thread (1,3,5,7..) and run ids with affinity 0-18/2 and leave the other CPU cores threaded to better handle the other small linux/applicationprocesses. and turning off hyper-threading would be a hastle since we're not in charge of that and I don't have enough time to test this Thank you, Michael.

Hi Martin, You are completely right but unfortunately we have to work with what we receive and I must work with this design. Probably we'll have to add more resources to counter the I/O bottleneck, but this CPU tuning ideea could be a first small step and that's why I wanted opinions on this since I've never heard about this kind of "tuning" Thank you, Michael.