Why benchmarks are important....

While those IBM sock puppets who defend IBM at any cost (Hi Serge) keep saying that Benchmarks don't sell databases, the truth is quite the opposite. Sure, no one buys a machine based on a benchmark. But its the benchmark that puts the solution on the customer's radar. Its also column fodder that can push you from column A to column B, C, or even D. Its the thing that keeps the salesman honest about the capability of their product. (Don't take my word for it. Learn Solution Selling Methodology and you'll understand. But then again, Serge et al are not customer facing sales people in S&D. Sure there are a couple IT Specialists who post, but then again, they don't manage the sales process, just answer 'technical' questions and keep the IT minions happy. They are also the cannon fodder who get sacrificed when the shite hits the fan.) Yet I digress. Sure IBM doesn't play well in the TPC.org arena. I mean they sure wouldn't want IDS to show up their venerable DB2. ;-) Nor do they want to start a war on the TPC-E tests where its now a pure hardware play since Microsoft is the defacto standard of that benchmark.... So we look at other 'industry' standards where IBM does play. Here's a blurb from el Reg: http://www.theregister.co.uk/2009/12/09/sun_sap_tests/ If you take the time to read the article, you see Sun/Oracle 10g against HP/Microsoft/SAP and then IBM (AIX/DB2). I'll cut to the chase. DB2 on AIX was a distant third when it came to horsepower. Also I wonder what the price was of each of the boxes, the article didn't say. Just the Sun pricing. For hardware and not the pricing for Oracle 10g and SAP. And didn't IBM go through a 'partnership' with SAP and specifically tune DB2 to run SAP better than the competition? (Oh! That's got to hurt.) Before anyone points out the obvious that SAP no longer supports IDS, thanks to JANET P., if you look at the configuration, many cores, lots of memory, IDS would have a distinct advantage. IDS may have issues with multiple threads per core, but even with that handicap, IDS's infrastructure should tend to give it an advantage. And yes, with some effort, I'm sure that getting beyond the multiple thread issue per core is not an impossible task. The whole point of this... 1) IBM has a product that has an advantage in competing with other players in the high multi-core environment. 2) IBM needs to promote their products by performing the benchmarks. If only to establish their name in this arena. 3) Unlike DB2, IDS does have customers who run on other machines than IBM Hardware. Sun, Linux, etc ... Just food for thought. But hey! What do I know? Ooops, my coffee cup is empty, got to run... ;-) -G _________________________________________________________________ Get gifts for them and cashback for you. Try Bing now. http://www.bing.com/shopping/search?q=xbox+games&scope=cashback&form=MSHYCB&publ=WLHMTAG&crea=TEXT_MSHYCB_Shopping_Giftsforthem_cashback_1x1

On 9 Dec, 13:45, Ian Michael Gumby <im_gu...@hotmail.com> wrote: > While those IBM sock puppets who defend IBM at any cost (Hi Serge) keep saying that Benchmarks don't sell databases, the truth is quite the opposite. > > Sure, no one buys a machine based on a benchmark. But its the benchmark that puts the solution on the customer's radar. Its also column fodder that can push you from column A to column B, C, or even D. Its the thing that keeps the salesman honest about the capability of their product.  (Don't take my word for it. Learn Solution Selling Methodology and you'll understand. But then again, Serge et al are not customer facing sales people in S&D. Sure there are a couple IT Specialists who post, but then again, they don't manage the sales process, just answer 'technical' questions and keep the IT minions happy. They are also the cannon fodder who get sacrificed when the shite hits the fan.) > > Yet I digress. > > Sure IBM doesn't play well in the TPC.org arena. > I mean they sure wouldn't want IDS to show up their venerable DB2. ;-) Nor do they want to start a war on the TPC-E tests where its now a pure hardware play since Microsoft is the defacto standard of that benchmark.... > > So we look at other 'industry' standards where IBM does play. > Here's a blurb from el Reg:http://www.theregister.co.uk/2009/12/09/sun_sap_tests/ > > If you take the time to read the article, you see Sun/Oracle 10g against HP/Microsoft/SAP and then IBM (AIX/DB2). > > I'll cut to the chase. DB2 on AIX was a distant third when it came to horsepower. Also I wonder what the price was of each of the boxes, the article didn't say. Just the Sun pricing. For hardware and not the pricing for Oracle 10g and SAP. And didn't IBM go through a 'partnership' with SAP and specifically tune DB2 to run SAP better than the competition? (Oh! That's got to hurt.) > > Before anyone points out the obvious that SAP no longer supports IDS, thanks to JANET P., if you look at the configuration, many cores, lots of memory, IDS would have a distinct advantage. > > IDS may have issues with multiple threads per core, but even with that handicap, IDS's infrastructure should tend to give it an advantage. And yes, with some effort, I'm sure that getting beyond the multiple thread issue per core is not an impossible task. > > The whole point of this... > > 1) IBM has a product that has an advantage in competing with other players in the high multi-core environment. > 2) IBM needs to promote their products by performing the benchmarks. If only to establish their name in this arena. > 3) Unlike DB2, IDS does have customers who run on other machines than IBM Hardware. Sun, Linux, etc ... > > Just food for thought. > > But hey! What do I know? Ooops, my coffee cup is empty, got to run... ;-) > > -G > > _________________________________________________________________ > Get gifts for them and cashback for you. Try Bing now.http://www.bing.com/shopping/search?q=xbox+games&scope=cashback&form=... And interestingly the next IDS patch for IDS 11.50 will contain performance enhancements for 40 or 50 cores! http://www-01.ibm.com/support/docview.wss?uid=swg1IC62462 IC62462: SPINLOCK AIOSHCB LOCK CAUSES PERFORMANCE PROBLEM IN PDQ SCAN When doing a full table scan on a table with 40 fragments,which were scanned in parallel on 40 cpu vps. The problem is caused by contention on a spin lock named 'AIOSHCB lock'. Running the command 'onstat -g spi |grep AIOSHCB' repeatedly will show a rapid increase in the AIOSHCB lock requests if this problem is being encountered. The problem is first fixed in IDS 11.50.xC6. http://www-01.ibm.com/support/docview.wss?uid=swg1IC61072 IC61072: PREFORMANCE DEGRADATION DUE TO CONTENTION ON THE DBS_PARTN MUTEX. A test with 50+ cpu vps was seeing high contention on the dbs_partn mutex, resulting in a performance bottleneck. The throughput improves by 10-20%, after changing the dbs_partn mutex to a read-write mutex. Problem has been fixed in IDS 10.00.xC11 and 11.50.xC4. So Neil Truby how is your box with 128 cpus doing?

<david@smooth1.co.uk> wrote in message news:b055d830-2707-4159-a411-02bf1d0b2b7c@m3g2000yqf.googlegroups.com... >> And interestingly the next IDS patch for IDS 11.50 will contain >> performance enhancements for 40 or 50 cores! ... http://www-01.ibm.com/support/docview.wss?uid=swg1IC61072 IC61072: PREFORMANCE DEGRADATION DUE TO CONTENTION ON THE DBS_PARTN MUTEX. A test with 50+ cpu vps was seeing high contention on the dbs_partn mutex, resulting in a performance bottleneck. The throughput improves by 10-20%, after changing the dbs_partn mutex to a read-write mutex. Problem has been fixed in IDS 10.00.xC11 and 11.50.xC4. >> So Neil Truby how is your box with 128 cpus doing? Mmm, 16 cpus. 64 cores. 128 "threads". FC11 isn't out yet, so not so well ;-) Seriously? The IBM labs are putting a lot of effort into analysis of this area, with more testing due this week. I am not sure whether I will be free to discuss the results, as the work is being done under a specific customer's PMR. But, it is an interesting area, and IBM is showing determination that IDS will be well positioned to exploit the new technology.

> From: neil.truby@ardenta.com > Subject: Re: Why benchmarks are important.... > Date: Wed, 9 Dec 2009 22:00:57 +0000 > To: informix-list@iiug.org > > <david@smooth1.co.uk> wrote in message > news:b055d830-2707-4159-a411-02bf1d0b2b7c@m3g2000yqf.googlegroups.com... [SNIP] > Mmm, 16 cpus. 64 cores. 128 "threads". > FC11 isn't out yet, so not so well ;-) > > Seriously? The IBM labs are putting a lot of effort into analysis of this > area, with more testing due this week. I am not sure whether I will be free > to discuss the results, as the work is being done under a specific > customer's PMR. But, it is an interesting area, and IBM is showing > determination that IDS will be well positioned to exploit the new > technology. > Which goes directly to my point. IDS has the capability, yet IBM? Not so much. Meaning that they don't know how to sell anything that isn't on the check list. But you can only scale an smp box so far, in terms of cores and memory. So you need to scale out. IDS? Not really planned. By the time IBM figures out which end is up, they will yet again be behind the curve, watching revenues drop in SWG while IGS gears up selling everything *but* IBM's products. ;-) _________________________________________________________________ Windows Live Hotmail gives you a free,exclusive gift. http://www.microsoft.com/windows/windowslive/hotmail_bl1/hotmail_bl1.aspx?ocid=PID23879::T:WLMTAGL:ON:WL:en-ww:WM_IMHM_7:092009

"Ian Michael Gumby" <im_gumby@hotmail.com> wrote in message news:mailman.357.1260404868.6236.informix-list@iiug.org... > From: neil.truby@ardenta.com > Subject: Re: Why benchmarks are important.... > Date: Wed, 9 Dec 2009 22:00:57 +0000 > To: informix-list@iiug.org > > <david@smooth1.co.uk> wrote in message > news:b055d830-2707-4159-a411-02bf1d0b2b7c@m3g2000yqf.googlegroups.com... [SNIP] > Mmm, 16 cpus. 64 cores. 128 "threads". > FC11 isn't out yet, so not so well ;-) > > Seriously? The IBM labs are putting a lot of effort into analysis of this > area, with more testing due this week. I am not sure whether I will be > free > to discuss the results, as the work is being done under a specific > customer's PMR. But, it is an interesting area, and IBM is showing > determination that IDS will be well positioned to exploit the new > technology. > > Which goes directly to my point. IDS has the capability, yet IBM? Not so much. Meaning that they don't know how to sell anything that isn't on the check list. But you can only scale an smp box so far, in terms of cores and memory. So you need to scale out. > IDS? Not really planned. Well ... a. That's a little unfair. The architecture features of IDS that *may* mean that an SMT chips OS needs a little tuning to optimise for IDS (or DB2. Or Oracle ....) pre-dates the SMT technology. b. You can't of course, as you'll no doubt point out, stand about crying over the fact that the world has moved on. And *my* point is that IBM is, so far as I can see (and I am well-positioned) taking this matter very seriously. ( ... "and I am well-positioned ....". F**k! I'm turning into you ....!).

> <david@smooth1.co.uk> wrote in message > news:b055d830-2707-4159-a411-02bf1d0b2b7c@m3g2000yqf.googlegroups.com... > >>> And interestingly the next IDS patch for IDS 11.50 will contain >>> performance enhancements for 40 or 50 cores! Describing bug fixes as "performance enhancements" is jolly loyal of you ... ;-)

> From: neil.truby@ardenta.com > Subject: Re: Why benchmarks are important.... > Date: Thu, 10 Dec 2009 00:44:02 +0000 > To: informix-list@iiug.org > > Well ... > > a. That's a little unfair. The architecture features of IDS that *may* mean > that an SMT chips OS needs a little tuning to optimise for IDS (or DB2. Or > Oracle ....) pre-dates the SMT technology. > b. You can't of course, as you'll no doubt point out, stand about crying > over the fact that the world has moved on. And *my* point is that IBM is, > so far as I can see (and I am well-positioned) taking this matter very > seriously. > > ( ... "and I am well-positioned ....". F**k! I'm turning into you ....!). > Errr yes, sort of. IBM purchased Sequent many moons ago, hence their expertise in SMP boxes. At around the same time, Informix was building out XPS. Shared Nothing, *near* *linear* in scalability. Now there's more to this but the point is that if you look at the architecture of the engine 5-6-7 (Ok 6 wasn't really in the market...) you start to see how IDS was engineered with Informix writing their own shared memory and thread libraries etc ... I don't know that you could say that prior to 7 that they had the pieces in place. Sequent and others were doing SMP. (Ok, you said SMT == Symentric multi-threading?) Same concept as SMP. You have to give credit where credit was due and Informix's engineering was light years ahead of their Oracle and IBM counterparts. Of course Informix didn't have the boat anchor of mainframe legacy that Oracle and IBM had. You can only scale an SMP/SMT box so large and so fast. IBM , Sun and even Intel and AMD are looking at scaling back on the speed of the cpu, but increasing the number of cores and threads to the cores. Of all of the databases that exist on the market, IDS has the shortest path to market when it comes to taking advantage of this. MMP shared nothing? IBM didn't take advantage of it. Project Arrowhead (Informix's version) would have put IDS in a unique situation. And yes, the world is moving on. Unfortunately IBM doesn't see this and even if they did, they are not in a position to capture market share. The reason you see bigtable (Google) and Hadoop (Yahoo) gaining traction is that you don't have to pay for DB licensing and you can write your own analytics. Hmmm seems IDS's extensibility makes even more sense. I saw a number of $30,000 a month to run a 2000 cluster network, so if you had to put EE licenses on each of those clusters, how much do you think it would cost? And while you're talking to customers who have 128 virtual core machines, I'm talking to customers who want to walk away from RDBMSs on tasks because of their licensing costs and lack of scalability. IBM put IDS and DB2 out in the cloud. How did that go? Are you starting to sound like me? Then you'd understand why IBM's IM marketing and sales leads need to get either a swift kick in the pants, or the working end of a cattle prod. Unfortunately that would be going against IBM's culture, in spite of their brand marketing. And getting back to the story, IBM is a distant third behind Sun/Oracle and HP/Microsoft. I wonder if things would be different if they were to do a Sun/IDS or HP/Linux/IDS or IBM/IDS. So yeah, you know what you know and can't talk about it. I know what I know and can't talk about it. And what you know should make you mad as hell. IDS is a fucking great product no one ever hears about. So who do you blame for that? _________________________________________________________________ Windows 7: Unclutter your desktop. Learn more. http://www.microsoft.com/windows/windows-7/videos-tours.aspx?h=7sec&slideid=1&media=aero-shake-7second&listid=1&stop=1&ocid=PID24727::T:WLMTAGL:ON:WL:en-US:WWL_WIN_7secdemo:122009

> From: theharlequin36@hotmail.com > Subject: Re: Why benchmarks are important.... > Date: Thu, 10 Dec 2009 00:45:33 +0000 > To: informix-list@iiug.org > > > <david@smooth1.co.uk> wrote in message > > news:b055d830-2707-4159-a411-02bf1d0b2b7c@m3g2000yqf.googlegroups.com... > > > >>> And interestingly the next IDS patch for IDS 11.50 will contain > >>> performance enhancements for 40 or 50 cores! > > Describing bug fixes as "performance enhancements" is jolly loyal of you ... > ;-) > I think if you apply this translation... he's just being polite. David said (translated): "Gee! Nobody thought that there would be such a thing as a need to go beyond 32 VCPUs so we didn't do things in a more efficient manner and didn't care if things started to thrash apart. So I guess we decided to put down the donuts and start to fix these darn things." Like I said. IDS can fit the massive SMT chips better than Oracle, DB2 or others. Only nobody knows about it. When you "brown neckers" come up for air, think about how better things could be if IDS were actually being marketed. -G BTW, there's this college project to create HBaseDB. That is a cloud of HBase machines where each node has a postgres database underneath it. Unfortunately its a bunch of college students so the quality isn't quite there, but the concept holds water. ;-) _________________________________________________________________ Windows Live Hotmail is faster and more secure than ever. http://www.microsoft.com/windows/windowslive/hotmail_bl1/hotmail_bl1.aspx?ocid=PID23879::T:WLMTAGL:ON:WL:en-ww:WM_IMHM_1:092009