IDS and P Series
Question: does IDS exploit SMT on IBM POWER7 (4 hardware threads per core), and should each thread be counted as a CPU? Consensus from the respondents: don't size CPU VPs by hardware threads — count cores (IBM also licenses Informix per core), using the rule of thumb of roughly one CPU VP per 450–500 MHz. Scaling with SMT threads is a curve, not linear, so you can't multiply cores by 2 or 4. Noted exception: Sun Niagara/T-class chips lack prefetch queues and suit Informix poorly, where one VP per core worked best. Advice ended with: benchmark your own workload.
Auto-generated by DrWatson from the posts below — may be imperfect; read the full thread.
Hello ... I have been asked by my management if IDS will take advantage of the new P Series chips that are multi threaded. From what I understand a single core will have 4 threads that run on it. They are calling it SMT. I'm going to guess that IDS would see each thread as a processor .. but wanted to make sure. Thanks in advance for any info. I think these are the IBM P 7 series chips ... Peter Logan Senior Database Administrator Phone: 616/878-8309
I have experience with one client P7 system and a couple of P6 systems. The P7s are an improvement over the P6. That said I would ignore the hardware threading and just count cores for configuring Informix. The rule of thumb stands, even on these machines, one CPU VP per 450-500 MHZ. A big hint is that IBM licenses Informix by the number of cores, not threads! Art On Aug 26, 2011 12:07 PM, "Peter_Logan@spartanstores.com" < Peter_Logan@spartanstores.com> wrote: > Hello ... > > I have been asked by my management if IDS will take advantage of the new P > Series chips that are multi threaded. From what I understand a single > core will have 4 threads that run on it. They are calling it SMT. I'm > going to guess that IDS would see each thread as a processor .. but wanted > to make sure. > > Thanks in advance for any info. I think these are the IBM P 7 series > chips ... > > Peter Logan > Senior Database Administrator > Phone: 616/878-8309 > > > ******************************************************************************* > Forum Note: Use "Reply" to post a response in the discussion forum. > --90e6ba212393f2226a04ab6b7e49
I would not ignore threads... but in the end it may not be different from your rule. P7 can be configure in SMT-4 and SMT-2. And although not all internal CPU units are in pairs (or more) in each core many of them are. What I blindly agree is that you cannot multiply the number of cores by 4 in SMT-4 configured systems. As you could/should not multiply by 2 in SMT-2 P6. If you create a graphic of scalability that shows performance vs threads, it will not be a straight line... It will be a curve with quick growth between 1 and 2 threads, and after that the growth will slow down. Different workloads will influence this. For example Java != databases Regards On Fri, Aug 26, 2011 at 6:01 PM, Art Kagel <art.kagel@gmail.com> wrote: > I have experience with one client P7 system and a couple of P6 systems. The > P7s are an improvement over the P6. That said I would ignore the hardware > threading and just count cores for configuring Informix. The rule of thumb > stands, even on these machines, one CPU VP per 450-500 MHZ. A big hint is > that IBM licenses Informix by the number of cores, not threads! > > Art > On Aug 26, 2011 12:07 PM, "Peter_Logan@spartanstores.com" < > Peter_Logan@spartanstores.com> wrote: > > Hello ... > > > > I have been asked by my management if IDS will take advantage of the new > P > > > Series chips that are multi threaded. From what I understand a single > > core will have 4 threads that run on it. They are calling it SMT. I'm > > going to guess that IDS would see each thread as a processor .. but > wanted > > > to make sure. > > > > Thanks in advance for any info. I think these are the IBM P 7 series > > chips ... > > > > Peter Logan > > Senior Database Administrator > > Phone: 616/878-8309 > > > > > > > > > ******************************************************************************* > > > Forum Note: Use "Reply" to post a response in the discussion forum. > > > > --90e6ba212393f2226a04ab6b7e49 > > > > ******************************************************************************* > Forum Note: Use "Reply" to post a response in the discussion forum. > > -- Fernando Nunes Portugal http://informix-technology.blogspot.com My email works... but I don't check it frequently... --20cf3002555a9b816204ab6f374e
I would add the rule of thumb doesn't appear to be valid with latest Sun processors, when I tested last I saw the greatest transaction through put with a one to one relationship with cores and CPU VPs Cheers Paul -----Original Message----- From: ids-bounces@iiug.org [mailto:ids-bounces@iiug.org] On Behalf Of Fernando Nunes Sent: Friday, August 26, 2011 4:28 PM To: ids@iiug.org Subject: Re: IDS and P Series [24741] I would not ignore threads... but in the end it may not be different from your rule. P7 can be configure in SMT-4 and SMT-2. And although not all internal CPU units are in pairs (or more) in each core many of them are. What I blindly agree is that you cannot multiply the number of cores by 4 in SMT-4 configured systems. As you could/should not multiply by 2 in SMT-2 P6. If you create a graphic of scalability that shows performance vs threads, it will not be a straight line... It will be a curve with quick growth between 1 and 2 threads, and after that the growth will slow down. Different workloads will influence this. For example Java != databases Regards On Fri, Aug 26, 2011 at 6:01 PM, Art Kagel <art.kagel@gmail.com> wrote: > I have experience with one client P7 system and a couple of P6 systems. The > P7s are an improvement over the P6. That said I would ignore the hardware > threading and just count cores for configuring Informix. The rule of thumb > stands, even on these machines, one CPU VP per 450-500 MHZ. A big hint is > that IBM licenses Informix by the number of cores, not threads! > > Art > On Aug 26, 2011 12:07 PM, "Peter_Logan@spartanstores.com" < > Peter_Logan@spartanstores.com> wrote: > > Hello ... > > > > I have been asked by my management if IDS will take advantage of the new > P > > > Series chips that are multi threaded. From what I understand a single > > core will have 4 threads that run on it. They are calling it SMT. I'm > > going to guess that IDS would see each thread as a processor .. but > wanted > > > to make sure. > > > > Thanks in advance for any info. I think these are the IBM P 7 series > > chips ... > > > > Peter Logan > > Senior Database Administrator > > Phone: 616/878-8309 > > > > > > > > > **************************************************************************** *** > > > Forum Note: Use "Reply" to post a response in the discussion forum. > > > > --90e6ba212393f2226a04ab6b7e49 > > > > **************************************************************************** *** > Forum Note: Use "Reply" to post a response in the discussion forum. > > -- Fernando Nunes Portugal http://informix-technology.blogspot.com My email works... but I don't check it frequently... --20cf3002555a9b816204ab6f374e **************************************************************************** *** Forum Note: Use "Reply" to post a response in the discussion forum. _____ avast! Antivirus <http://www.avast.com>; : Outbound message clean. Virus Database (VPS): 110826-1, 08/26/2011 Tested on: 8/26/2011 4:34:40 PM avast! - copyright (c) 1988-2011 ALWIL Software.
Sun's threads are "lighter" than Power... More threads tend to be lighter, meaning that in many workloads (database) the curve I mentioned earlier has a slower "start" Regards On Fri, Aug 26, 2011 at 10:34 PM, Paul Watson <paul@oninit.com> wrote: > I would add the rule of thumb doesn't appear to be valid with latest Sun > processors, when I tested last I saw the greatest transaction through put > with a one to one relationship with cores and CPU VPs > > Cheers > Paul > > -----Original Message----- > From: ids-bounces@iiug.org [mailto:ids-bounces@iiug.org] On Behalf Of > Fernando Nunes > Sent: Friday, August 26, 2011 4:28 PM > To: ids@iiug.org > Subject: Re: IDS and P Series [24741] > > I would not ignore threads... but in the end it may not be different from > your rule. > P7 can be configure in SMT-4 and SMT-2. And although not all internal CPU > units are in pairs (or more) in each core many of them are. > What I blindly agree is that you cannot multiply the number of cores by 4 > in > SMT-4 configured systems. As you could/should not multiply by 2 in SMT-2 > P6. > > If you create a graphic of scalability that shows performance vs threads, > it > will not be a straight line... It will be a curve with quick growth between > 1 and 2 threads, and after that the growth will slow down. > > Different workloads will influence this. For example Java != databases > > Regards > > On Fri, Aug 26, 2011 at 6:01 PM, Art Kagel <art.kagel@gmail.com> wrote: > > > I have experience with one client P7 system and a couple of P6 systems. > The > > P7s are an improvement over the P6. That said I would ignore the hardware > > threading and just count cores for configuring Informix. The rule of > thumb > > stands, even on these machines, one CPU VP per 450-500 MHZ. A big hint is > > that IBM licenses Informix by the number of cores, not threads! > > > > Art > > On Aug 26, 2011 12:07 PM, "Peter_Logan@spartanstores.com" < > > Peter_Logan@spartanstores.com> wrote: > > > Hello ... > > > > > > I have been asked by my management if IDS will take advantage of the > new > > P > > > > > Series chips that are multi threaded. From what I understand a single > > > core will have 4 threads that run on it. They are calling it SMT. I'm > > > going to guess that IDS would see each thread as a processor .. but > > wanted > > > > > to make sure. > > > > > > Thanks in advance for any info. I think these are the IBM P 7 series > > > chips ... > > > > > > Peter Logan > > > Senior Database Administrator > > > Phone: 616/878-8309 > > > > > > > > > > > > > > > > > **************************************************************************** > *** > > > > > Forum Note: Use "Reply" to post a response in the discussion forum. > > > > > > > --90e6ba212393f2226a04ab6b7e49 > > > > > > > > > > **************************************************************************** > *** > > Forum Note: Use "Reply" to post a response in the discussion forum. > > > > > > -- > Fernando Nunes > Portugal > > http://informix-technology.blogspot.com > My email works... but I don't check it frequently... > > --20cf3002555a9b816204ab6f374e > > > **************************************************************************** > *** > Forum Note: Use "Reply" to post a response in the discussion forum. > > _____ > > avast! Antivirus <http://www.avast.com>; : Outbound message clean. > > Virus Database (VPS): 110826-1, 08/26/2011 > Tested on: 8/26/2011 4:34:40 PM > avast! - copyright (c) 1988-2011 ALWIL Software. > > > > ******************************************************************************* > Forum Note: Use "Reply" to post a response in the discussion forum. > > -- Fernando Nunes Portugal http://informix-technology.blogspot.com My email works... but I don't check it frequently... --001517592c5a76b9d804ab702b58
You are absolutelyu right Paul. The Sun T-class or Niagra processors are brain-dead. They were specifically designed to run web servers and other Java or C++ style multi-threaded applications that run many many very short run threads that spend most of their time waiting for other threads to coordinate. To accomplish efficiently Sun had to eliminate the pre-fetch queues that made modern SMT processors so good at running databases and other long running threads/processes. With no pre-fetch queue Informix's style of multithreading just is not efficient. Indeed, Oracle has the same problem running on Niagra processors - DB2 does best but still does not run as well on the T-class processors as it does on the traditional SMT oriented M-class Sparc processors. While IBM built similar hardware threading into the P6 and P7 processors as that on the Niagra cores, they did not do away with the pre-fetch queues and so performance is comparable to or better than a non-threaded P5 processor core running at the same speed. Art Art S. Kagel Advanced DataTools (www.advancedatatools.com) Blog: http://informix-myview.blogspot.com/ Disclaimer: Please keep in mind that my own opinions are my own opinions and do not reflect on my employer, Advanced DataTools, the IIUG, nor any other organization with which I am associated either explicitly, implicitly, or by inference. Neither do those opinions reflect those of other individuals affiliated with any entity with which I am affiliated nor those of the entities themselves. On Fri, Aug 26, 2011 at 5:34 PM, Paul Watson <paul@oninit.com> wrote: > I would add the rule of thumb doesn't appear to be valid with latest Sun > processors, when I tested last I saw the greatest transaction through put > with a one to one relationship with cores and CPU VPs > > Cheers > Paul > > -----Original Message----- > From: ids-bounces@iiug.org [mailto:ids-bounces@iiug.org] On Behalf Of > Fernando Nunes > Sent: Friday, August 26, 2011 4:28 PM > To: ids@iiug.org > Subject: Re: IDS and P Series [24741] > > I would not ignore threads... but in the end it may not be different from > your rule. > P7 can be configure in SMT-4 and SMT-2. And although not all internal CPU > units are in pairs (or more) in each core many of them are. > What I blindly agree is that you cannot multiply the number of cores by 4 > in > SMT-4 configured systems. As you could/should not multiply by 2 in SMT-2 > P6. > > If you create a graphic of scalability that shows performance vs threads, > it > will not be a straight line... It will be a curve with quick growth between > 1 and 2 threads, and after that the growth will slow down. > > Different workloads will influence this. For example Java != databases > > Regards > > On Fri, Aug 26, 2011 at 6:01 PM, Art Kagel <art.kagel@gmail.com> wrote: > > > I have experience with one client P7 system and a couple of P6 systems. > The > > P7s are an improvement over the P6. That said I would ignore the hardware > > threading and just count cores for configuring Informix. The rule of > thumb > > stands, even on these machines, one CPU VP per 450-500 MHZ. A big hint is > > that IBM licenses Informix by the number of cores, not threads! > > > > Art > > On Aug 26, 2011 12:07 PM, "Peter_Logan@spartanstores.com" < > > Peter_Logan@spartanstores.com> wrote: > > > Hello ... > > > > > > I have been asked by my management if IDS will take advantage of the > new > > P > > > > > Series chips that are multi threaded. From what I understand a single > > > core will have 4 threads that run on it. They are calling it SMT. I'm > > > going to guess that IDS would see each thread as a processor .. but > > wanted > > > > > to make sure. > > > > > > Thanks in advance for any info. I think these are the IBM P 7 series > > > chips ... > > > > > > Peter Logan > > > Senior Database Administrator > > > Phone: 616/878-8309 > > > > > > > > > > > > > > > > > **************************************************************************** > *** > > > > > Forum Note: Use "Reply" to post a response in the discussion forum. > > > > > > > --90e6ba212393f2226a04ab6b7e49 > > > > > > > > > > **************************************************************************** > *** > > Forum Note: Use "Reply" to post a response in the discussion forum. > > > > > > -- > Fernando Nunes > Portugal > > http://informix-technology.blogspot.com > My email works... but I don't check it frequently... > > --20cf3002555a9b816204ab6f374e > > > **************************************************************************** > *** > Forum Note: Use "Reply" to post a response in the discussion forum. > > _____ > > avast! Antivirus <http://www.avast.com>; : Outbound message clean. > > Virus Database (VPS): 110826-1, 08/26/2011 > Tested on: 8/26/2011 4:34:40 PM > avast! - copyright (c) 1988-2011 ALWIL Software. > > > > ******************************************************************************* > Forum Note: Use "Reply" to post a response in the discussion forum. > > --90e6ba21241d1ebe7604ab70445e
Am 26.08.2011 18:06, schrieb Peter_Logan@spartanstores.com: > Hello ... > > I have been asked by my management if IDS will take advantage of the new P > Series chips that are multi threaded. From what I understand a single > core will have 4 threads that run on it. They are calling it SMT. I'm > going to guess that IDS would see each thread as a processor .. but wanted > to make sure. > > Thanks in advance for any info. I think these are the IBM P 7 series > chips ... > > Peter Logan > Senior Database Administrator > Phone: 616/878-8309 > > > ******************************************************************************* > Forum Note: Use "Reply" to post a response in the discussion forum. > > Hello Peter, if this is a posible answer to management, then ask for a side-by-side comparative PoC P7 vs EX5 including SSD and memory expansion options. This will show very good, that: 'If you want more performance, you must learn to spend less money' But then, who on earth really wants to know such things? dic_k -- Richard Kofler SOLID STATE EDV Dienstleistungen GmbH Vienna/Austria/Europe
Test, test and then test again :-) > You are absolutelyu right Paul. The Sun T-class or Niagra processors are > brain-dead. They were specifically designed to run web servers and other > Java or C++ style multi-threaded applications that run many many very > short > run threads that spend most of their time waiting for other threads to > coordinate. To accomplish efficiently Sun had to eliminate the pre-fetch > queues that made modern SMT processors so good at running databases and > other long running threads/processes. With no pre-fetch queue Informix's > style of multithreading just is not efficient. Indeed, Oracle has the same > problem running on Niagra processors - DB2 does best but still does not > run > as well on the T-class processors as it does on the traditional SMT > oriented > M-class Sparc processors. > > While IBM built similar hardware threading into the P6 and P7 processors > as > that on the Niagra cores, they did not do away with the pre-fetch queues > and > so performance is comparable to or better than a non-threaded P5 processor > core running at the same speed. > > Art > > Art S. Kagel > Advanced DataTools (www.advancedatatools.com) > Blog: http://informix-myview.blogspot.com/ > > Disclaimer: Please keep in mind that my own opinions are my own opinions > and > do not reflect on my employer, Advanced DataTools, the IIUG, nor any other > organization with which I am associated either explicitly, implicitly, or > by > inference. Neither do those opinions reflect those of other individuals > affiliated with any entity with which I am affiliated nor those of the > entities themselves. > > On Fri, Aug 26, 2011 at 5:34 PM, Paul Watson <paul@oninit.com> wrote: > >> I would add the rule of thumb doesn't appear to be valid with latest Sun >> processors, when I tested last I saw the greatest transaction through >> put >> with a one to one relationship with cores and CPU VPs >> >> Cheers >> Paul >> >> -----Original Message----- >> From: ids-bounces@iiug.org [mailto:ids-bounces@iiug.org] On Behalf Of >> Fernando Nunes >> Sent: Friday, August 26, 2011 4:28 PM >> To: ids@iiug.org >> Subject: Re: IDS and P Series [24741] >> >> I would not ignore threads... but in the end it may not be different >> from >> your rule. >> P7 can be configure in SMT-4 and SMT-2. And although not all internal >> CPU >> units are in pairs (or more) in each core many of them are. >> What I blindly agree is that you cannot multiply the number of cores by >> 4 >> in >> SMT-4 configured systems. As you could/should not multiply by 2 in SMT-2 >> P6. >> >> If you create a graphic of scalability that shows performance vs >> threads, >> it >> will not be a straight line... It will be a curve with quick growth >> between >> 1 and 2 threads, and after that the growth will slow down. >> >> Different workloads will influence this. For example Java != databases >> >> Regards >> >> On Fri, Aug 26, 2011 at 6:01 PM, Art Kagel <art.kagel@gmail.com> wrote: >> >> > I have experience with one client P7 system and a couple of P6 >> systems. >> The >> > P7s are an improvement over the P6. That said I would ignore the >> hardware >> > threading and just count cores for configuring Informix. The rule of >> thumb >> > stands, even on these machines, one CPU VP per 450-500 MHZ. A big hint >> is >> > that IBM licenses Informix by the number of cores, not threads! >> > >> > Art >> > On Aug 26, 2011 12:07 PM, "Peter_Logan@spartanstores.com" < >> > Peter_Logan@spartanstores.com> wrote: >> > > Hello ... >> > > >> > > I have been asked by my management if IDS will take advantage of the >> new >> > P >> > >> > > Series chips that are multi threaded. From what I understand a >> single >> > > core will have 4 threads that run on it. They are calling it SMT. >> I'm >> > > going to guess that IDS would see each thread as a processor .. but >> > wanted >> > >> > > to make sure. >> > > >> > > Thanks in advance for any info. I think these are the IBM P 7 series >> > > chips ... >> > > >> > > Peter Logan >> > > Senior Database Administrator >> > > Phone: 616/878-8309 >> > > >> > > >> > > >> > >> > >> > >> >> **************************************************************************** >> *** >> > >> > > Forum Note: Use "Reply" to post a response in the discussion forum. >> > > >> > >> > --90e6ba212393f2226a04ab6b7e49 >> > >> > >> > >> > >> >> **************************************************************************** >> *** >> > Forum Note: Use "Reply" to post a response in the discussion forum. >> > >> > >> >> -- >> Fernando Nunes >> Portugal >> >> http://informix-technology.blogspot.com >> My email works... but I don't check it frequently... >> >> --20cf3002555a9b816204ab6f374e >> >> >> **************************************************************************** >> *** >> Forum Note: Use "Reply" to post a response in the discussion forum. >> >> _____ >> >> avast! Antivirus <http://www.avast.com>; : Outbound message clean. >> >> Virus Database (VPS): 110826-1, 08/26/2011 >> Tested on: 8/26/2011 4:34:40 PM >> avast! - copyright (c) 1988-2011 ALWIL Software. >> >> >> >> > ******************************************************************************* >> Forum Note: Use "Reply" to post a response in the discussion forum. >> >> > > --90e6ba21241d1ebe7604ab70445e > > > ******************************************************************************* > Forum Note: Use "Reply" to post a response in the discussion forum. > -- Paul Watson Tel: +1 913-674-0360 Mob: +1 913-387-7529 Web: www.oninit.com www.advancedatatools.com Failure is not as frightening as regret. If you want to improve, be content to be thought foolish and stupid.