Re: Tuning IDS for Multicore CPUs
A user asked how to size IDS CPU VPs when moving from single-core UltraSPARC III to multicore UltraSPARC IV+ machines. The advice given: the classic rule is one CPU VP per processor (to minimise context switching), with a rough guideline of about two CPU VPs per three cores, though the best setting depends on chip architecture (e.g. shared L2) and workload. The thread then drifted into IBM licensing, noting Processor Value Unit pricing scales with core count and processor type, and a warning that heavily multi-threaded chips (e.g. Sun T1) can perform poorly for database workloads. No definitive tuning answer or follow-up results were recorded.
Auto-generated by DrWatson from the posts below — may be imperfect; read the full thread.
>From: Claus Samuelsen <csa@dk.ibm.com> >the_omegamon@yahoo.com wrote: > > Greetings All, > > > > I am hoping someone can provide some guidance concerning the tuning > > of > > IDS for multicore CPU architectures. Currently, I have a number Sun > > Ultra > > Sparc III machines that will be replaced with multicore Ultra Sparc IV > > + boxes. [SNIP]>The "official" general rule is one CPU VP per single core processor, but >as you mention, it may be better with other settings. A rule of thumb >for multi core processors is two CPU VPs per three cores. Again it >depends - partly on the architecture (fx shared L2 cache) and partly >(probably most) on the usage of IDS. >One of the ideas with IDS dynamic scalable architecture (DSA) is to >avoid context switching, because thread switching has a lower cost than >process switching. With only one CPU VP per processor context switching >is minimized. > I think you're going to have to rethink the "rule of thumb". At least in terms of Sun's architecture. Intel's multi-core and AMD's multi-core architecture is to put multiple cores (read whole CPUs) on a single die. This means that a quad core machine actually has 4 "regular" CPUs and your "rule of thumb" should work. Sun's multi-cores are a bit of a different beast. here's a link to a news.com article that talks a bit about Sun's multicore cpus... http://news.com.com/Sun+puts+16+cores+on+its+Rock+chip/2100-1006_3-6141961.html So I think that you're on your own, however, you should post back what you find. It should be interesting... How does IBM handle pricing of Sun's multicore cpus. Is a 32 core license twice as expensive as a 16 core processor? Thx -G _________________________________________________________________ Learn.Laugh.Share. Reallivemoms is right place! http://www.reallivemoms.com?ocid=TXT_TAGHM&loc=us
Ian Michael Gumby wrote: > > How does IBM handle pricing of Sun's multicore cpus. Is a 32 core > license twice as expensive as a 16 core processor? > https://www-112.ibm.com/software/howtobuy/passportadvantage/valueunitcalculator/vucalc.wss
"Claus Samuelsen" <csa@dk.ibm.com> wrote in message news:46bc1839$0$2092$edfadb0f@dread15.news.tele.dk... > Ian Michael Gumby wrote: >> >> How does IBM handle pricing of Sun's multicore cpus. Is a 32 core license >> twice as expensive as a 16 core processor? Yes. > https://www-112.ibm.com/software/howtobuy/passportadvantage/valueunitcalculator/vucalc.wss I couldn't find SPARC on here except at the bottom of a cvery long "Other" tree. But anyway, the principle is always that 2n cores of any one processor type will cost twice as much to licence as n cores. IBM has introduced "Processor Value Unit (PVU)" pricing, where the total amount of PVUs required is a function not only of the number of cores, as was the case under the old model, but of the processor type also. For most processors, SPARC included, this has been a non-event, as the licence requirement has remained identical between the old and new models. But for some, notably POWER5 and I think Sun's new T1 range, there can be a software licensing cost advantage under the new scheme.
Hey! Sorry to top post... Its a good site to look at and see comperable pricing. I chose an 8 core T1 vs a quad core intel. Sun = 240 processor units Intel = 200 processor units. As to which will give you better overall performance? Who knows for sure. (I dont even think IBM knows since *NO ONE* has done a published benchmark... ;-) [OK, so maybe IBM has some internal numbers...] For the original poster, it would be nice to know how you did in your tuning efforts. The chip architecture is different enough that its never going to be an apples to apples comparison between server architecture. IMHO, based on the numbers I have seen on the IBM site, I'd say that IBM does know roughly the performance values. For some situations, the T1 may be a better buy, depending on the overall configuration and application. (Financial Services applications as an example....) Interesting stuff.... -G >From: Claus Samuelsen <csa@dk.ibm.com> >To: informix-list@iiug.org >Subject: Re: Tuning IDS for Multicore CPUs >Date: Fri, 10 Aug 2007 09:48:07 +0200 >MIME-Version: 1.0 >Received: from perform.iiug.org ([216.177.38.211]) by >bay0-mc4-f2.bay0.hotmail.com with Microsoft SMTPSVC(6.0.3790.2668); Fri, 10 >Aug 2007 00:50:26 -0700 >Received: from localhost (localhost [127.0.0.1])by perform.iiug.org >(Postfix) with ESMTP id 53E05B1E4;Fri, 10 Aug 2007 03:50:21 -0400 (EDT) >Received: from perform.iiug.org ([127.0.0.1])by localhost (perform.iiug.org >[127.0.0.1]) (amavisd-new, port 10024)with ESMTP id SF7Jb9aw3mIT; Fri, 10 >Aug 2007 03:50:15 -0400 (EDT) >Received: by perform.iiug.org (Postfix, from userid 60001)id C64E9B1D8; >Fri, 10 Aug 2007 03:50:14 -0400 (EDT) >Received: from perform.iiug.org (localhost [127.0.0.1])by perform.iiug.org >(Postfix) with ESMTP id E5966B1C8;Fri, 10 Aug 2007 03:50:09 -0400 (EDT) >X-Message-Delivery: Vj0zLjQuMDt1cz0wO2k9MDtsPTA7YT0w >X-Message-Info: >ZLIeWzn802Q0y/GqifAj8lfz/xWL8Kb7QgM2YIUyUQuZiQWNTwWyr2Hs6FmysHm245F/9fvsb3O4Ww4ziTbJvg== >X-Virus-Scanned: amavisd-new at iiug.org >Path: >nnrp.xmission!xmission!nntp.infostrada.it!newsfeed01.sul.t-online.de!newsfeed00.sul.t-online.de!t-online.de!news.tele.dk!feed118.news.tele.dk!not-for-mail >User-Agent: Thunderbird 2.0.0.4 (X11/20070625) >Newsgroups: comp.databases.informix >References: <mailman.590.1186696058.13675.informix-list@iiug.org> >Lines: 7 >Organization: TDC Totalloesninger >NNTP-Posting-Host: 80.166.21.170 >X-Trace: 1186732089 dread15.news.tele.dk 2092 80.166.21.170:64039 >X-Complaints-To: abuse@post.tele.dk >Xref: nnrp.xmission comp.databases.informix:198798 >X-BeenThere: informix-list@iiug.org >X-Mailman-Version: 2.1.6 >Precedence: list >List-Id: "comp.databases.informix" <informix-list.iiug.org> >List-Unsubscribe: ><http://www.iiug.org/mailman/listinfo/informix-list>,<mailto:informix-list-request@iiug.org?subject=unsubscribe>; >List-Archive: <http://www.iiug.org/pipermail/informix-list>; >List-Post: <mailto:informix-list@iiug.org> >List-Help: <mailto:informix-list-request@iiug.org?subject=help> >List-Subscribe: ><http://www.iiug.org/mailman/listinfo/informix-list>,<mailto:informix-list-request@iiug.org?subject=subscribe>; >Errors-To: informix-list-bounces@iiug.org >Return-Path: informix-list-bounces@iiug.org >X-OriginalArrivalTime: 10 Aug 2007 07:50:26.0662 (UTC) >FILETIME=[1D619060:01C7DB23] > >Ian Michael Gumby wrote: > > > > How does IBM handle pricing of Sun's multicore cpus. Is a 32 core > > license twice as expensive as a 16 core processor? > > > >https://www-112.ibm.com/software/howtobuy/passportadvantage/valueunitcalculator/vucalc.wss >_______________________________________________ >Informix-list mailing list >Informix-list@iiug.org >http://www.iiug.org/mailman/listinfo/informix-list _________________________________________________________________ Tease your brain--play Clink! Win cool prizes! http://club.live.com/clink.aspx?icid=clink_hotmailtextlink2
>From: "Neil Truby" <neil.truby@ardenta.com> >IBM has introduced "Processor Value Unit (PVU)" pricing, where the total >amount of PVUs required is a function not only of the number of cores, as >was the case under the old model, but of the processor type also. For most >processors, SPARC included, this has been a non-event, as the licence >requirement has remained identical between the old and new models. But for >some, notably POWER5 and I think Sun's new T1 range, there can be a >software >licensing cost advantage under the new scheme. > The sparc is under the risc tree. ;-) Yes the value per core is the same 8 vs 16 vs whatever. This works against Sun in their first generation of chips since the cores are a bit "light". In their second generation, things will improve. But in the example I ran, a Quad Core Intel chip was 200 units, while the 8 core was 240. My guess is that they are fairly equivelent and IBM has leveled the playing field to a degree. For my customers, I think building out a quad core box (MSI for example) would be a better general computing box, than the sun 8-core. However, I can see for certain applications that the Sun would be a better choice. (As a vendor/reseller there are other factors which make the Intel based boxes better for me. Can you say "margin" ;-) Again it would be interesting to get some independent research and do a write up. -Mike _________________________________________________________________ Tease your brain--play Clink! Win cool prizes! http://club.live.com/clink.aspx?icid=clink_hotmailtextlink2
On Aug 10, 6:42 am, "Ian Michael Gumby" <im_gu...@hotmail.com> wrote: > Hey! > Sorry to top post... > > Its a good site to look at and see comperable pricing. > > I chose an 8 core T1 vs a quad core intel. > Sun = 240 processor units > Intel = 200 processor units. > > As to which will give you better overall performance? Who knows for sure. > (I dont even think IBM knows since *NO ONE* has done a published > benchmark... ;-) > [OK, so maybe IBM has some internal numbers...] > > For the original poster, it would be nice to know how you did in your tuning > efforts. > > The chip architecture is different enough that its never going to be an > apples to apples comparison between server architecture. > > IMHO, based on the numbers I have seen on the IBM site, I'd say that IBM > does know roughly the performance values. > > For some situations, the T1 may be a better buy, depending on the overall > configuration and application. (Financial Services applications as an > example....) > > Interesting stuff.... > > -G > > > > >From: Claus Samuelsen <c...@dk.ibm.com> > >To: informix-l...@iiug.org > >Subject: Re: Tuning IDS for Multicore CPUs > >Date: Fri, 10 Aug 2007 09:48:07 +0200 > >MIME-Version: 1.0 > >Received: from perform.iiug.org ([216.177.38.211]) by > >bay0-mc4-f2.bay0.hotmail.com with Microsoft SMTPSVC(6.0.3790.2668); Fri, 10 > >Aug 2007 00:50:26 -0700 > >Received: from localhost (localhost [127.0.0.1])by perform.iiug.org > >(Postfix) with ESMTP id 53E05B1E4;Fri, 10 Aug 2007 03:50:21 -0400 (EDT) > >Received: from perform.iiug.org ([127.0.0.1])by localhost (perform.iiug.org > >[127.0.0.1]) (amavisd-new, port 10024)with ESMTP id SF7Jb9aw3mIT; Fri, 10 > >Aug 2007 03:50:15 -0400 (EDT) > >Received: by perform.iiug.org (Postfix, from userid 60001)id C64E9B1D8; > >Fri, 10 Aug 2007 03:50:14 -0400 (EDT) > >Received: from perform.iiug.org (localhost [127.0.0.1])by perform.iiug.org > >(Postfix) with ESMTP id E5966B1C8;Fri, 10 Aug 2007 03:50:09 -0400 (EDT) > >X-Message-Delivery: Vj0zLjQuMDt1cz0wO2k9MDtsPTA7YT0w > >X-Message-Info: > >ZLIeWzn802Q0y/GqifAj8lfz/xWL8Kb7QgM2YIUyUQuZiQWNTwWyr2Hs6FmysHm245F/9fvsb3O4Ww4ziTbJvg== > >X-Virus-Scanned: amavisd-new at iiug.org > >Path: > >nnrp.xmission!xmission!nntp.infostrada.it!newsfeed01.sul.t-online.de!newsfeed00.sul.t-online.de!t-online.de!news.tele.dk!feed118.news.tele.dk!not-for-mail > >User-Agent: Thunderbird 2.0.0.4 (X11/20070625) > >Newsgroups:comp.databases.informix > >References: <mailman.590.1186696058.13675.informix-l...@iiug.org> > >Lines: 7 > >Organization: TDC Totalloesninger > >NNTP-Posting-Host: 80.166.21.170 > >X-Trace: 1186732089 dread15.news.tele.dk 2092 80.166.21.170:64039 > >X-Complaints-To: ab...@post.tele.dk > >Xref: nnrp.xmissioncomp.databases.informix:198798 > >X-BeenThere: informix-l...@iiug.org > >X-Mailman-Version: 2.1.6 > >Precedence: list > >List-Id: "comp.databases.informix" <informix-list.iiug.org> > >List-Unsubscribe: > ><http://www.iiug.org/mailman/listinfo/informix-list>,<mailto:informix-list-requ...@iiug.org?subject=unsubscribe>; > >List-Archive: <http://www.iiug.org/pipermail/informix-list>; > >List-Post: <mailto:informix-l...@iiug.org> > >List-Help: <mailto:informix-list-requ...@iiug.org?subject=help> > >List-Subscribe: > ><http://www.iiug.org/mailman/listinfo/informix-list>,<mailto:informix-list-requ...@iiug.org?subject=subscribe>; > >Errors-To: informix-list-boun...@iiug.org > >Return-Path: informix-list-boun...@iiug.org > >X-OriginalArrivalTime: 10 Aug 2007 07:50:26.0662 (UTC) > >FILETIME=[1D619060:01C7DB23] > > >Ian Michael Gumby wrote: > > > > How does IBM handle pricing of Sun's multicore cpus. Is a 32 core > > > license twice as expensive as a 16 core processor? > > >https://www-112.ibm.com/software/howtobuy/passportadvantage/valueunit... > >_______________________________________________ > >Informix-list mailing list > >Informix-l...@iiug.org > >http://www.iiug.org/mailman/listinfo/informix-list > > _________________________________________________________________ > Tease your brain--play Clink! Win cool prizes!http://club.live.com/clink.aspx?icid=clink_hotmailtextlink2 Beware of multi-threaded chip architectures! Applications with high cpu intensity and low icache misses (like database engines) run poorly on many multi-threaded chip architectures. Not all multi-threaded chip architectures are alike! I suggest due diligence (try googling) for other's experience, there lots of information out there. Until the DBMS engines catch up to the multi-threaded chip architectures and the multi-threaded chip architectures make some changes, some applications will run very poorly so stick with the older chips for now.