* Posts by Jesper Frimann

478 publicly visible posts • joined 8 Oct 2008

Page:

Oracle hammered as hardware sales soften

Jesper Frimann
Headmaster

Sales of Power.

Well what fuels POWER7 sales right now is the fact that POWER5(+) kit is seeing EOL, in terms of typical company depreciation. So people are going from Power5(+) -> POWER7.

Which surely is stupid cause it's much cheaper to do n+1 upgrades, rather than n+2 forklift upgrades.

Now the recent upgrades to the 770 and the 780 are actually pretty major ones. Both the POWER 770-MMB and the 780-MHB IMHO suffered from having to much processor punch compared to in system unit IO and Memory.

So I think the new 770-MMC and 780-MHC are more balanced, and you should be able to keep all your adapters in the system units, if you do proper IO virtualization, and 4TB ram to 64 cores is more like it.

// Jesper

Jesper Frimann
Devil

Hmm..

I kind off have some of the same experiences as Matt here, but again I have bad experiences with pretty much every vendor .)=

Some more than others. But I do remember the "Performance tuning gurus" from SUN, I once worked with together on a client, not really being that guru like.

Basically there were more interested in blaming the disk vendor (think it was EMC and it wasn't the DMX that was the problem), than fixing the problem.

// Jesper

Oracle, Cisco crow new database flash dash record

Jesper Frimann
Headmaster

Yeah Yeah

Well it also still got beat by a 2 chip 8 core POWER7 submission, that is around 20% for 2/3 the number of cores.

But that's what these big companies do everytime they submit something, and Oracle is perhaps the worst of them, it's a WORLD record... geee.

// Jesper

Oracle fires Itanium countersuit at HP

Jesper Frimann
Holmes

Well...

Oracle is currently doing all it can to destroy UNIX, as we know it. Shame on them.

But one thing is sure, there are quite a few migrating Oracle on HPUX -> Oracle on something else projects being spawned.

// Jesper

Virtualisation: just a lot of extra software licences?

Jesper Frimann
Headmaster

Eh ?

Perhaps one should then buy some decent boxes or start to use some decent virtualization technology.

One of the real money gainers of virtualization is reducing the number of ports, and raising the utilization of these. I mean fewer converged 10Gbit ports perhaps running at 30-50% average utilization is much much better than having several hundreds of SAN and 1GBit Network ports running at max 5% average utilization.

// Jesper

Oracle gives Solaris 11 final spit and polish

Jesper Frimann
Trollface

Check your facts.

Actually AIX 5.3 runs on Common Hardware Reference Platform (CHRP) machines. This includes old machines that are RS64 based and even 604e based machines like the F50.

Come on now..

// Jesper

IBM big iron OSes treated to spit and polish

Jesper Frimann
Trollface

easy now..

Keb...

Sure Alisons comparison with regards to threads is fair. He is comparing current shipping products to shipping products. Now you saying that this will be changed in a few months is also fair.

As for the 100.000 virtual servers, I have no doubt that a mainframe could do that, but hey is it relevant?

I guess not just as relevant as having a single (non supercomputer) OS image using 16.384 threads.

And as for Oracle FUD and marketing lies.. do you really want a list ?

// Jesper

IBM preps boost for Power Systems

Jesper Frimann
Gimp

Yes.

To be quite frank, without the help of redpieces and redbooks, we would be lost trying to figure out how to design a solution.

The normal IBM websites are today to infested with marketing bullsh*t and fancy flash stuff that doesn't really help you. And to be honest, although they have greatly improved, the manuals do contain all to many circular references. And it can usually take som time to actually locate what you need. Much easier in the old days where there just were a PDF file that contained it all.

// Jesper

Jesper Frimann
Terminator

Gear.

Well, when I look around the home office.. it says HP printer, HP scanner, Philips screen, Toshiba laptop, hp calculator (I did use to study physics) .

And well I don't know if the new system's announced are actually POWER7+ (32nm), or just a 'we've gotten better at making POWER7' version of the chip I don't really know. *shrug*

Well the real change is the increase in RAM x2 for the POWER 770 is a real game changer for us, it basically means x2 the amount of load on the machine. So for RL performance this is good. Furthermore the announcements talk about increased IO bandwidth, I supposed that means that the P5IOC chips that only ran at GX+ speeds have been updated. That should give the machine x2 IO when only using the system units IO and 33% more when using IO drawers.

The POWER 780 now supposedly goes to 16 sockets, according to the IBM homepage, So basically this means that the machine has quite an unused potential of IO and memory bandwidth.

An option like the IBM has on the x3850 and x3690 with extra memory in an addon box would be great.

But with 16 sockets, the POWER 780 actually surpasses the M9000 with 64 sockets on the specint2006rate high score.

// jesper

Jesper Frimann
Thumb Up

Actually it's already here (I guess) .....

in the form of new POWER 770 and POWER 780 systems.

The new POWER 770 (which is the one that I normally use) will at

64 cores@3.3 GHz and 4TB of RAM doing 4,132 Watts at 100% utilization

versus the old one which will at

64 cores@3.1 GHz and 2TB of RAM doing 4,472 Watts at 100% utilization.

So 5% more oomph with x2 the amount of RAM using 8% less POWER.

Sure reminds me of the Jump from POWER5 to POWER5+.

// Jesper

So where is IBM's Power7+ CPU?

Jesper Frimann
Mushroom

Well they did do something... TPM

Cause if you take a 1 System Unit POWER 770-MMB with 3.1 GHz Power7 cores and ships it with 512 GB of the same RAM as you ship in a POWER 770-MMC with 3.3GHz POWER7 cores.

Then the MMB uses 1,118 Watt and the MMC uses.. 1033 Watt.

Try it our yourself:

http://www-912.ibm.com/see/EnergyEstimator

So cranking up the Processor speed with 200 MHz makes each chip use 44 watt less.

Hmm.. Well guess we'll get wiser with time.

// Jesper

Oracle's Sparc T4 prices mask improved value

Jesper Frimann
Headmaster

Capcaity versus throughput

Well, the capacity seems to be around the same for the T3 and the T4.

Where capacity is the SUM of throughput for all the threads at the optimum number of threads.

Now the behaviour of the T4, I guess, will be much more like POWER7, when you add threads and surely that will mean that a single thread per core will have the full benefit of the core cause it'll be running SMT and not fine grained multithreading.

So.. T4 is a huge step away from what Oracle have been singing for these last years. So hopefully there are a lot of people who didn't go the T1,T2,T3 way who now will have a much smother migration onto new hardware.

// Jesper

Ellison: 'There'll be nothing left of IBM once I'm done'

Jesper Frimann

To the people cheering while Larry turns Oracle into IBM of the 70ties

I really think it's amazing that people are cheering while Larry turns Oracle into something that looks more like the IBM of the 70ies or 80ties. Nobody today in their right mind would be doing that.

At some point in time Larry will stand all alone, just like IBM did in the late 80ies. The problem for Larry is that back then IBM was still the biggest player and had the time, competitive space and money to turn it's business around. The problem for Larry is that he has clients desperately trying to get off his platform, the media have labelled him the 'bad guy' and two bigger competitors that are focusing more and more on him, cause he is behaving like a dork.

So IMHO what Larry should do is focusing on running his business, rather than trying to destroy others.

// Jesper

Oracle previews Solaris 11, due in November

Jesper Frimann
Trollface

Snap me there snap me here....

Eh.. you do know that AIX have had snapshots since AIX 5.2, that is late 2002 early 2003 right ?

// Jesper

Jesper Frimann
Trollface

First hit on google:

http://www-03.ibm.com/systems/power/software/aix/v71/editions.html

And I was actually wrong... 256 physical cores maps to 1024 logical cores that can then map to 10240 virtual cores. But I don't think you'll get AIX to go beyond 1024, but the virtualization layer will hence you can run 10 virtual servers each with 1024 virtual cores ontop of a machine with 256 physical cores.... *cough* *cough* but you can then also use Wpars and then subdivide that again.. *cough* *cough*

I'd better stop now.

Well the whole thread thing, it's really the throughput of Business operations that count, not so much the thread

// Jesper

Jesper Frimann
Trollface

funny how...

"For instance, recently, AIX needed to be rewritten to be able to handle as "many" as 256 cores in the P795. Solaris 10 handled more that, many years ago."

You are contradictiong yourself, and cannot even see it. You write just above:

"Today Solaris 10 only scales to 512 virtual CPUs - vCPUs. "

You do know that 256 physical cores on the p795 equals 1024 virtual CPU's right ?

So basically in your book with all your Solaris 10 is more scalable than AIX cause it supports more threads have been crap. You do realise that ? Cause you are actually saying it yourself here.

That's actually funny.

// Jesper

Oracle rises for Unix server push

Jesper Frimann
Devil

Old times...

"anyone else, and I often get very worked up by their support processes. "

Yes, the worst response is still "Works as designed". And the whole 'We are measured on closing, not necessary solve, as many problems as possible, that seems to be the common trend in the industry, also makes blood preasure heat up.

"When IBM entered the Open Systems world in 1990, they were regarded as the Big Enemy by many UNIX people, myself included, but I think that they did actually prove themselves."

Isn't that a bit of an understatement ? I do remember how I felt myself, after growing up on VAX running BSD, HP9000 systems also running BSD and motorola 680XX based SUN systems running Sunos. I remember my first real UNIX job in the mid nineties, where I desperately tried to get my employer to use HP systems rather than RS/6000 machines, until I actually kind of started acting like a professional and gave it a chance. And AIX was actually a fresh breath of air. LVM was clearly superior than the rather arcane ways of many other UNIX'es, and SMIT was a nice easy way to get started on the OS. But it took some years to get rid of the heretic label, that my friends who most of them worked with Solaris put on me.

Today where I have very little hands on, that's the price you pay for going the IT-architect way, my perspective have changed, today it's all about money, and less about technical 'stuff'. But I have found that enabling technical excellence is perhaps the best way to save money.

// jesper (The BSD devil for old times sake)

Jesper Frimann
Headmaster

Yeah Yeah

"1) x86 is catching up fast in terms of performance?"

Nahh... I caught up quite a few years ago. The problem is that the field seems to be narrowed down to INTEL x86 and POWER, when you talk about 'performance'. The rest including SPARC is pretty much looking like legacy and niche products.

"2) Intel is adding RAS to Intel Xeon cpu quite quick"

Jup. And the price is rising. But Xeon -EX'es are really the sweet spot for Intel. The highend x86 marked is tiny compared to for example the rest of the highend marked.

"3) The price difference between x86 servers and POWER7 servers is smaller than ever?

Jup. POWER7 rocks.

"4) IBM has officially said that they are going to replace AIX with Linux, in the future?

Yawn. IBM also said once that only 4 computers were needed in the world. Come on.

// Jesper

Jesper Frimann
Trollface

Keb. I don't care about IBM/SUN/ORACLE whatever anonymous fanbois that post strange things. Neither should you, and neither does anybody else.

The problem with you is that you echo without any reality check what soever, Oracle marketing material. And the material is pretty desperate at the moment cause to be quite frank Oracle is under enormous pressure in the UNIX marked.

// Jesper

Jesper Frimann
Facepalm

Braiiins... Braiiinss...

You go all zombie when you read Oracle marketing material don't you ?

SPECjEnterprise2010

Yes, this is the fastest submission, sure it it is. But if you really wanna conclude anything, then it should be that WebLogic is a damn fast application server. That will run fast when you throw a lot of flash disk and and and... after the benchmark.

Again this is a benchmark that really haven't many submissions.

TPC-H

The oracle submission is... NOT.. I REPEAT NOT the world record, there are several other servers that are not only faster but also a good deal cheaper on the price per transaction.

The Oracle E-Business Suite R12.1.2 XL

Yes, world record.. why.. cause T4 is the only submitted result.

Oracle PeopleSoft Enterprise Payroll (N.A.) 9.1

Lots of results here, but T4 is the only one using the newest version. What is mentioned is that there actually is a mainframe result where the T4 with 32 cores 256GB of RAM and Flash drives and native Oracle software stack, only manages to outperform a older generation IBM z10 mainframe with 9 cores and 32 GB of RAM using DB2 (not a native platform for the APP) which isn't using flash drives with a factor of 2.8. Meaning that the per core performance of the mainframe is 25% better than the T4.

How sad for you Keb, a old mainframe being faster....

Oracle PeopleSoft HRMS Self-Service 9.1.

Whenever I find a link.. it's broken.. so it's kind of hard to actually verify anything. But again it's oracle's own software stack.. in a new version .. bla bla..

Oracle JD Edwards EnterpriseOne

I've covered that one.

The rest isn't really world records or anything just claims about how good the machine is.. Kind of not very serious.

So what was your point Keb ?

// jesper

Jesper Frimann
Facepalm

@Kebbie

"Never mind about the T3 results. What is your comment regarding the T4 results? Here are some world records, with more down below this one.

http://blogs.oracle.com/BestPerf/entry/20110927_sparc_t4_2_jdedwards"

Are you really such a sucker for marketing material ?

The T4 benchmark that Oracle have run is not even in the same category as the IBM iSeries result mentioned. You can't really compare them.

Furthermore the iSeries result is not an official result of any kind, if you had bothered following the link to the IBM website, it's an internal IBM test of the JD edwards stack, it's not even an official submitted benchmark result.

Lets just say that the Oracle marketing guys must have looked a long time for this one. Geeezes..

And again this is 60GB of RAM and 48 normal disks, with a CPU utilization of 60% running with DB2 as the database (Not a the optimized Oracle stack for JD Edwards) versus a fully native software stack running on 3 machines with x10 the RAM, internal flash disks and 2 flash Storage servers.

Gee, comparing this is.. to put it mildly not serious. If you can't see that .. well.. you have a problem IMHO.

// Jesper

Jesper Frimann
Holmes

T4.

First of all, it's nice to see that Oracle have managed to get a decent processor out. That is truely good news for the competitive landscape and the UNIX marked.

Now I would have liked Oracle to actually have come up with some more benchmark results that were in the category, "Now we can really compare to something". Rather than the usual suspects where there either isn't anything to compare against or it's their own 100% controlled software stack with a new version that nobody else is running.

But in benchmarks and war you should only pick the fights you can win. And Oracle is very good at avoiding Westmere-EX and POWER7 where they know they can't win.

Lets have a look.

TPC-H.

This is truly a great benchmark result. It's a real banger. But to be honest I think that Oracle is the only vendor who has truly cracked this benchmark. If you look at all the submissions except the Oracle DB on SPARC then they have much longer run time for Q1, Q9,Q13, Q18 and Q21.

The recent Oracle benchmark on the M8000 and the T4-4 actually runs faster than the HP DL980 with 80 Westmere-EX cores.

Now on a benchmark like SPECINTRATE the M8000 does 882@64 cores where as the DL980 with 80 cores does 2080. So IMHO it's the software stack and perhaps setup that allows for Oracle to excel in this benchmark, not so much the hardware. And what really points toward the setup and that they have cracked the benchmark is that the SUPERDOME 2 submission with the Oracle DB shows the same pattern as the other non Oracle on Oracle hardware. I would have presumed that the SD2 was at least 25% faster than the M8000.

Which kind of makes me better understand why HP is so pissed at Oracle. If HP suspects Oracle not helping out the max, when not doing submissions on their own hardware... well

As for the POWER7 based POWER 780 submission. Come on it's Sybase on Linux, that's not really a serious measure of what the platform is capable off.

jEnterprise2010.

Well, as I have said before this is a benchmark that nobody really heard about before Oracle started using it in their marketing material.

Now there is no denying that Oracle also here have made a good benchmark .

If you compare the setup of the two submission that Kebabbert are pointing to you can clearly see the difference:

http://www.spec.org/jEnterprise2010/results/res2010q3/jEnterprise2010-20100628-00012.png

versus

http://www.spec.org/jEnterprise2010/results/res2011q3/jEnterprise2010-20110907-00027.jpg

But you have to acknowledge that the Oracle submission uses Flash disks, 8 times the number of storage nodes, 12 times the memory, 16 times the network bandwidth, 8 times the storage bandwidth. And last sure they only use x4 the number of cores. But CPU is not always the bottleneck.

Then there is the JDedwards benchmark (http://blogs.oracle.com/BestPerf/entry/20110701_sparc_t3_1_jd)

Get real Keb. You are comparing Oracle running ontop of System i formerly known as OS400, to a UNIX platform. If you knew anything about AS400/os400, you would know better than to deduce anything about Oracle performance on AIX based upon those results. But I guess you don't.

So to conclude nice benchmarks but it's still to cherry picked to really conclude much about the true performance of T4 compared to the other players, but that it's a huge step in the right direction from Oracle, if it turns out that it has a good single threaded throughput. I just don't understand why they had to go through T1-T3 to get to T4. Now compared to T3 I would say that the throughput is most likely the same.

If we look at the jEnterprise2010, result then we will see that the T4 result gives pretty much x4 throughput with x4 the number of nodes.

So Oracle have managed to keep throughput while, speeding up (which is still to be seen by how much) the single threaded throughput. That is great, but IMHO still not good enough.

// Jesper taking on his flame resistant hat.

Oracle revs up Sparc, speeds up roadmap

Jesper Frimann
Headmaster

RE BILL

I don't think that saving a small amount of money on the hardware on Exadata/Exalogic, by using x86 machines rater than SPARC based boxes is an issue.

T series servers most likely aren't that expensive to make, and it's all inhouse hence you won't be paying profit to Intel and others. Hence the price difference will be much much smaller than one might think.

I think one of the primary reasons why Exadata/Exalogic are still x86 based are historical and practical reasons. It takes time and effort to move to another platform.

Personally i think that if T4 or it's successor performs well enough, and their roadmap for SPARC is long enough, then it makes great sense for Oracle to migrate these solutions onto SPARC.

For example I think that IBM made a mistake when something like the SVC and SONAS was implemented on x86 rather than POWER.

// Jesper

Oracle defies the economy – and the curse of Sun

Jesper Frimann
Holmes

"Regarding this Oracle benchmark I posted, I thought it was relevant because it shows how much performance x86 servers can deliver, winning POWER7 servers, including P795. This is a relevant piece of fact, in an discussion about x86 vs POWER/SPARC. We see that x86 can in some cases be faster already today."

The problem Keb, is that given a benchmark that is enough embarrassingly parallel, a room full of mobile phones could beat a POWER 795. Hence the relevance of your comparison. There is no doubt that the Westmere-EX boxes and POWER7 based machines shipping today are really the two top dogs to look out for in the server business.

But they mostly do play in two different areas of the server marked, POWER in the midrange to highend, and Westmere based servers in the lowend to midrange.

Many years from now there will still be Itanium based servers, Mainframes, SPARC and POWER servers running. So perhaps you should cut down on the rhetoric, if you want people to listen to you.

// Jesper

Jesper Frimann
Headmaster

@Kebabbert

I must admit I fail to see what your clustered SAP SD 2-TIER benchmark has to do with x86 scalability.

But perhaps you should have a look at your good friend Larry's statement about his x86 business, before you get all hyped up on x86.

http://www.eweekeurope.co.uk/news/oracle-happy-to-see-its-low-end-business-tumble-40363

With regards to Ivy Bridge. Then there seems to be a big difference between what is released, and what you think is released. You won't see a Ivy Bridge -EX processor in 2012 if INTEL keeps the normal pace of releasing -EX models after -EP models.

Furthermore POWER7+ will be the processor shipping POWER servers when Ivy Bridge enters the marked. But hey, shouldn't you rather do your normal comparing against POWER6 HW from 2007.

// Jesper

Jesper Frimann
Headmaster

RE:Hardware Margins

Woow 54% gross profit margin on Hardware ?

Now that is alot, that is what 14% more than IBM, which has a much larger portion of it's sales in the highend than Oracle has, where the profit margins are higher.

They surely must have cut to the bone on sales, and/or charge to much for their kit.

// Jesper

Jesper Frimann
Holmes

Re:Oracle betting on clusters

Sure they are... If you read the benchmark text then the SAP app code ran on the

Now what does an Oracle Enterprise edition with RAC cost running on 480 Cores ?

Lets see the per core factor for the cores is 0,5 with 480 cores.

the cost of the Oracle license is 47,500 $

the cost of RAC is 23.000 $

Hence the cost is 16.920.000 $ + 3.722.400$ per year in support. And that is just for the database.

And we haven't even started piling up other database features, like DB management etc. and hardware and ....

This is why Oracle is pushing clusters.

Furthermore SD2 parallel is well, almost an irrelevant benchmark. Why, cause it's a embarrassingly parallel benchmark run on a cluster.

Each node will be able to pull around ~25.000 users at 99% utilization. if we look at the other vendors's 8 socket boxes. Now the Oracle submission runs at 92% utilization. Hence ~25.000 x 6 x0, 92 =138.000 is the theoretical max performance for 6 nodes. Now the Oracle benchmark delivers 134.080 which is 97%+ of the theoretical peak performance at 92% utilization. Geee..

Now if they did a SAP SD3Tier parallel it would tell you something about the DB layer. Not the 2Tier, that is just marketing bull.

@AC with regards to the POWER 740 being the best POWER b ox, nope use the 770. If you have a POWER 570 based POWER6 then a MESS upgrade to the 48 core version is actually a good deal, but remember to get a good discount. Don't let IBM charge you the same as for a 64 core upgrade.

A 48 core machine will still run out of RAM, if used right, before it runs out of processor power IMHO.

// Jesper

Sparc T4 chips: Oracle makes stealthy spec change

Jesper Frimann
Holmes

Smoking while posting

Whatever you are smoking can't be legal.

// Jesper

Oracle's Sparc T4 chip: Will you pay Larry's premium?

Jesper Frimann
Holmes

have one's cake and eat it too

The increase in single threaded throughput is great, but I fear that it comes at the cost of the total chip throughput. I mean the whole idea with Tx chips up until now have been to sacrifice single threaded throughput for chip throughput.

And a factor of 5 on single threaded throughput still isn't enough IMHO to match most of the competition.

// Jesper

IBM yanks chain on 'Blue Waters' super

Jesper Frimann
Pint

Re:re Jesper

Hi AC.

First, I must admit that my knowledge of the POWER 775 is kind of like on the car magazine level. I haven't seen any material on any IBM sites about the box. There haven't been written any redbooks, there aren't any manuals online yet. So....

But I am a bit puzzled by what you write about the machine being split up into 8 images. I know that if you order the machine with preloaded images. That is what you'll get, from eConfig per default if you don't specify something else. But if you specify one Image.. eConfig doesn't protest. But you can verify that pretty quickly if you can have one image that can span a whole machine, by simply having a look at the HMC, and see what it'll let you do.

Now AIX scales to 1024 threads quite nicely, we got our first POWER 795 here some time ago.. and it got test booted up with one virtual machine with 1024 virtual cores. Wroom Wroom.

With regards to QCM's, you do mean DCM right ?

Now with regards to tuning code. All I've done is to read this paper:

https://www.power.org/events/Power7/Performance_Guide_for_HPC_Applications_on_Power_755-Rel_1.0.1.pdf

mostly out of curiosity. I haven't done HPC work for... like 6-7 years since I was a consultant. Today being an architect it's all about spreadsheets and power point and explaining to managers how things really work in the real world.

With regards to turning SMT on and off.. yeah, half or 3/4th of you processors will disappear, so that you'll have to fix any scripts etc. that assumes a continuous list of processors, so no for x in `lsdev -C....` but hey... :)=

But to use SMT or not to use SMT on HPC computing depends on what you are doing.. if you are only doing FP and that is what you need to do as fast as possible, then it's hard to get to much benefit :)=

// Jesper

Jesper Frimann
Holmes

re:AC

Quite a good explanation of the differences of workloads in HPC.

But with regards to POWER7 and the POWER 775 versus the 575 I think you are wrong. Sure POWER6 is faster clocked than POWER7 is.

But Judging from the benchmarks released, then POWER7 is at least as fast on a per thread level as POWER6, and most likely a good bit faster. But it's not huge, and if you have badly compiled code that runs in GHz then sure :)=

Now comparing the POWER 575 with it's newer version the 775 then it's x8 the memory and Processor cores.. so I guess your numbers aren't really that accurate.

// Jesper

Oracle bestows Sparc T4 beta on 'select' customers

Jesper Frimann
Thumb Down

Yawn.

Funny how you keep pulling 4 year old hardware out of your hat when you have to compare against your favourite Oracle products. It's to be quite honest ridiculous.

And your link here is to a benchmark made by Oracle, controlled by Oracle...

And then there is the TPC-C cluster benchmark again and again that you keep comparing to non clustered results. You aren't being serious.

Again Yawn.

// jesper

Jesper Frimann
Holmes

re:Joerg

Joerg.

With regards to Round robin and thread switching on the Niagara.

Well you are quite right that the newest version of the Niagara family of processor cores won't do round robin in the way that they will idle 7/(2x4=8) of the time if only a single thread is active.

They will do a LRU (which they call Least Recently Fetched (LRF)) which is a round robin, if more than one thread is active, dispatch to active threads on each of the two execution units.

You can even call it a bit of a hybrid fine and coarse grained multithreading.

But the Niagara core (pre T4) is still a much much simpler core than for example the POWER7 core, and one thread cannot dynamically take up all the resources in the core, as a single thread can on the POWER7 core, if it needs to. Which leads me to your remark.

Yes you are quite right, that the average per thread throughput of a POWER7 will be the throughput divided by the number of cores divided by the number of threads. Sure, this is something that we have discussed before.

BUT you still can get really really good single threaded throughput, if nobody else is using the resources in the core, I would estimate that the single threaded throughput of the POWER7 is around 26 specint2006, if you divde out and adjust for SMT giving around 80% or something. And you don't need to disable SMT or or or.. you just need the condition where one thread is executing alone on a physical core.

And that is the real difference between POWER7 and T1/T2/T3 cores. You get both the good throughput and the good single threaded throughput. Sure you don't get both at the same time looking at a per core level.

// Jesper

Jesper Frimann
Headmaster

re:Joerg Moellenkamp

Well you are doing a bit to much number magic here.

Single strand going up to 5 timers faster is not the same as taking the per core performance (8 threads/strands) of the T3 and then multiplying that number with 5. It's taking the performance of one thread/strand of the T3 and multiplying that by 5.

IMHO the Maximum per thread performance of the T4 will most likely be in the range of 12-15 SPECintRate2006. (that is 5 times the performance of the max throughput of a single T3 thread (which again AFAIR is ~x2 the throughput of a single thread when all threads are running))

Which might sound really not good if you look at your examples with other architectures. But... I actually think that is kind of OK. Why ?

Because the single strand performance of a Westmere-EX with Hyperthreading enabled isn't the same as the per core performance either. And the same goes for POWER7 with SMT enabled.

Both processors when running with SMT/Hyperthreading enabled does favour throughput over single strand throughput. And there is a price to pay for that. So their single thread/strand throughput isn't equal to specintrate2006 divided by they number of cores either.

Try having a look at the specint2006 score of POWER6, where there actually are numbers that you can use, cause the specint2006 scores didn't use autopar.

IBM System p 570 (4.7 GHz, 1 core-1 Thread) 21.6 specint 2006

Now running on 2 cores with specint2006rate (4 threads) the score becomes:

IBM System p 570 (4.7 GHz, 2 core-2 Threads) 60.9 specint2006 rate.

Hence doubling the number of cores gives almost a factor of 3 in performance. So using your math this would have given POWER6@4.7GHz a specint2006 score of 30.5. (without autopar) Which it clearly doesn't have.

The difference is (at least for POWER7 I haven't read up on the latest enhancements for Hyperthreading so I'll keep from making faulty statements on that one) that POWER7 is able to allocate all the resources (execution slots) to a single thread cause it's running a fairly clever implementation of SMT, and thus be able to reach pretty close to the MAX per thread throughput it is actually capable of. The fine grained (round robin) way of the T3 can't do that trick, but from what I understand about Yosemite Falls (T4) it can do much the same thing as POWER6/POWER7 and thus get both good per thread throughput and good per chip throughput.

Now on the other hand IMHO it's still to little to late...

// Jesper

Jesper Frimann
Devil

Binary compability of SPARC

I think perhaps that Allison Park took his mouth a bit to full. There is AFAIK full forward going binary comparability between the Tx processors and SPARC64 or the other SPARC processors.

But, from what I hear from our Solaris group, is that you need to retune your software stack when going from SPARC64 or SPARC IV or .. to the Niagara, and sometimes it's a .. real.. shitty deal to put it blunt.

Again that is not so much the fault of the Niagara architecture, which is kind of beautiful if used for the right workload, but more the fault of SUN and Oracle sales teams who have sold an architecture to their clients which didn't really fit in.

Furthermore you have the whole Solaris 11 dropping support for rather new "old" SPARC processors. So it's not like customers doesn't have enough to be angry about.

// Jesper

Jesper Frimann
Flame

Actually you are being too kind.

It's actually 3.6 times the performance. So you are being too kind.

And the current T3 chip does not match the best of the pack, like POWER7 and Westmere-EX, unless the workload can take advantage of the buildin accelerators in the T3.

Furthermore you might say that the T4 is to little to late, and this slippage, only makes it worse. Cause there is nothing about improved throughput on a per chip level on the Oracle roadmap for the T4, so the performance gap will only get larger.

Furthermore yet another SPARC roadmap slip is not what Oracle needs right now. If the roadmap that Larry put our less than a year ago is already starting to slip, and slip for processors that were supposed to be shipping this year. Then I guess that HP and IBM sales guys will have a field day.

// Jesper

IBM heaves new System z minis at mainframe shops

Jesper Frimann
Facepalm

*boggle*

"I work in Finance, and we never round off. We always use whole numbers and book keep the number of decimals separately. We never round off. If Mainframes are rounding off, it is another reason to not use them in finance."

You lack of knowledge in the field where you claim expertise is stunning.

// Jesper who actually have written finance applications many years ago.

Jesper Frimann
Holmes

Re:Kebabbert.

You didn't really read and understand what I wrote now did you ?

My point was that raw processor performance is just one of many factors that governs how many 'business transactions' you can do on a server.

// Jesper

Jesper Frimann
Holmes

Jup, they will make him.. mainframe capacity planner.

Most likely not...

But when you've been involved in 'downsizing' projects where people have moved from HP3000 or IBM Mainframes or older platforms, you get a certain respect for these old platforms.

I've seen people size based upon the assumption that Java code could do just as many business transactions as highly tuned old Cobol would be able to do.

But then again, IT infrastructure is simple to people who do by looking at product briefs.

// jesper

IBM 'Blue Waters' super node washes ashore in August

Jesper Frimann
Pint

Yeah but..

half the number of cores and double the RAM, and we'd have a new prime candidate for our 'shared'/cloud like environment.

Wroom Wroom.

// Jesper beer cause it weekend.

PS: TPM a type seems to have sneaked into the article "for a total of 246 cores, on a single, massive motherboard", should have been 256 :)=

Unix still data center darling, says survey

Jesper Frimann
Holmes

Sure keb...

"I certainly dont agree with you. I mean, IBM offerings has scaling problems and does not scale as well as Solaris (TPC-C, AIX scaling was rewritten to handle P795 with a measly 256 cores)."

Ehh ? What the f word are you talking about. The POWER server platform actually has good scaling. Well at least compared to anything Oracle can muster.

Lets have a look at perhaps the easiest scalable benchmark SPECINTrate2006.

Here going from 128 cores to 256 cores the new SPARC64+ based M9000 has a scaling factor of 91.6% compared to an ideal x2, now the POWER 795 going all the way from the 32 cores to 256 cores has a scaling factor of 98.7%.

But we can of cause also look at pure raw performance numbers, the 256 core M9000 with the 2.88 GHz SPARC VII on SAP2Tier does 175,600 SAPS, the POWER 795 does 688,630 SAPS. I mean even the 64 core POWER 780 does 202,180 SAPS.

Come on... Don't throw stones when you live in a glass house filled with china.

"IBM AIX is copying Solaris DTrace and renaming it Probevue."

Well, just cause others are doing neat stuff doesn't mean that you shouldn't do it yourself.

"IBM for many years trash talked Sun Niagara and said that 1-2 cores at high clock speed such as 5-6GHz is the future, because data bases like strong cores. To use many cores at lower speeds is just a bad idea, said IBM. One strong core is the future. "

You do know that POWER have managed to do all three things. Put more cores on a chip AND increase the per core throughput and socket throughput.

SPECJBB2005 for example:

POWER 780 with 8 chips and 64 POWER7 cores@3.86 GHz does 5,087,469 BOBS.

POWER 595 with 32 chips and 64 POWER6 cores@5.00 GHz does 3,435,485 BOBS.

That is a difference of 48% per core throughput.

and lets have a look at SPECINT2006rate

POWER 780 with 8 chips and 64 POWER7 cores@3.86 GHz does 2740

POWER 595 with 32 chips and 64 POWER6 cores@5.00 GHz does 2160

that is a difference in 27% in per core throughput.

And it not even with the highest clocked POWER7.

Again you have absolutely no clue what so ever.

"And now today POWER7 does not have 1 core at 8-9GHz, but instead it has many lower clocked cores, just like Niagara. Sun realized that GHz race will shift to many core race, but IBM did not understand that until POWER7. POWER6 was 5GHz and 2 cores. POWER7 is not 6-7GHz and 1-2 cores. So, the future is not 1-2 strong cores. Back in the Sun days, 8 cores in a cpu was just crazy, no one had that, except Sun. Today Oracle aim for 512 threads in one cpu, which is crazy today. But tomorrow everyone will have it. IBM will copy that many threads too."

Eh.. again with your GHz.. you sound like a IBM Mainframe sales guy.

And you don't get it again, in servers it's about having as good as possible combination of per thread, core and per socket throughput.

And to be quite blunt SUN/Oracle have in recent years failed being in the top pack with more than one of these at the same time. POWER on the other hand have managed to increase all three things: the per core, per thread, per chip and per socket throughput from generation to generation.

And you seriously need a history lesson. Cause you do seem to totally ignore facts.

Now comparing SPARC and POWER history wise.

More than one core per chip.

2001 POWER4 2 cores per chip

2005 Xenon 3 cores per chip.

2006 Cell 1+8 cores per chip

2007 BlueGene/P chip 4 cores per chip

2010 POWER7 chip 8 cores per chip.

2004 UltraSPARC IV 2 cores per chip.

2005 UltraSPARC T1 8 cores per chip.

2007 SPARC64 VI 2 cores per chip.

2008 SPARC64 VII 4 cores per chip

2009 SPARC64 VIIIfx 8 cores per chip

2010 SPARC T3 16 cores per chip

More than one chip per socket.

2001 POWER4 MCM modules for the p690 resulting in 4 chips/8 cores/8 threads per socket.

2004 POWER5 MCM modules for the p595 resulting in 4 chips / 8 cores per/16 threads socket.

2005 POWER5+ QCM modules for the p520/550/560 resulting in 2 chips/4 cores/8 threads per socket

2011 POWER7 MCM modules for the POWER 775 resulting in 4 chips/32 cores/128 threads per socket

SUN/Oracle doesn't have/use this technology.

Multithreading:

2000 RS64 IV implemented 2 way Coarse Grained MultiThreading.

2004 POWER5 implemented 2 way Simultaneous MultiThreading.

2010 POWER7 implemented 4 way Simultaneous MultiThreading.

2005 UltraSPARC T1 implemented 4 way Fine Grained MultiThreading.

2007 UltraSPARC T2 implemented 8 way Fine Grained MultiThreading.

2007 SPARC64 VI implemented 2 way Coars Grained MultiThreading.

2008 SPARC64 VI implemented 2 way Simultaneous MultiThreading.

Do I really need to say more ?

"And for instance Solaris ZFS, I dont know of any IBM storage solution that protects your data as well as ZFS does. It would not surprise me if IBM copy ZFS too, soon."

I think the guy that put this best was Linus, when he asked why does a filesystem have to do that ?

And cool as it is.. I have to say I agree with Linus, this is perhaps taking the role of the filesystem one step to far.

As for putting computers into a container... well... I would hardly call that innovation. We have them.. it's stupid in most cases IMHO, but as a hack to have variable capacity that you can move from location to location it's ok.

"etc etc etc."

So that is basically your argument ? You couldn't

"What techniques has IBM created, that Solaris copied? You talk about "recent years". Can you give an example?"

You have to be kidding ? You forget that IBM is a hundred year old company that pretty much invented the Computer Industry together with companies like NCR, Xerox, bell labs ... and later DEC and ...

And to be honest Oracle (SUN) is kind of new in that perspective. So what about.....

The RISC processor ?

The hard disk ?

The tape drive ?

Logical Volume Manager.

journalised filesystem.

Software package management system.

Solaris Jumpstart (AIX NIM)

And as far as I understand Solaris 11 will kind of start to use a ODM system like the one found in AIX.

The Hypervisor used on the T series ?

All due respect to Oracle (SUN), but they are not one of the inventors of the computer business, they will always have to build on stuff made by others, simply cause they are a much younger company.

And the rest is just your normal B******, I think people are pretty tired of hearing it. I mean even 'serious' IT people who really dig Solaris and SPARC (and there are a lot of those) are distancing them from you. Perhaps U need to find a new tune to play.

// jesper

Jesper Frimann
Linux

Re:Mark 65

Oh.. sure.. you just take your 40 core and 700GB RAM virtual machine and Vmotion it from one machine to another .. over your 4x10GBit link aggregated management network.

Now there are people who are really driving a Ford F150/F250/F350 when they say that they drive a truck, and then there are people that are driving an eighteen wheeler.

Now that doesn't mean that they aren't really both right, and that the guy who drives the eighteen wheeler doesn't have a F150 at home. But there is a difference.

And btw Linux runs just fine on POWER :)=

// Jesper writing from a Linux box.

Cisco may slash 10,000 jobs

Jesper Frimann
Facepalm

It's a self for filling prophecy

I work quite a lot with offshored people. I have a project manager right now from India. He is F****** good at his job. Sure it's not optimal that he is sitting in India, and is only up here when we really really need him to be here.

And that is a clear problem. But mostly what I've seen around in the business is management that doesn't really understand the business they are in. That don't understand that you cannot take a something that perhaps requires 1-2 very skilled people and just offshore it to 10 entry level people. You'll foobar whatever you are trying to cut cost on. Sure you can find very skilled people in the typical offshoring countries, sure you can. But the prices are different.

And I can assure you, that you for example don't find many 'mainframe people' in the Ukraine, russia or China with 15 years of experience.

And when you offshore without knowing what you are doing, your local guys end up having to clean up, and the cost of cleaning up goes on your local budget, which makes you more inefficient which leads to more offshoring which leads to more cleaning up which ..........

Note that I'm not saying that you shouldn't have people sitting in other countries.. But you damn well have to know what you are doing.

// Jesper

IBM chases Itanium shops with 'Breakfree' deals

Jesper Frimann
Headmaster

It is what it is.

Eh ?

"You are making the unsubstantiated conclusion that any IBM gains were due to competitive take-outs"

No I am not. I have made no such claim. I have no way of knowing that.

"whereas IBM's gains int eh quarter were more likley to have been old Power upgrades to newer Power6 (which was reduced to clear) and Power7. "

Yes I agree, that this has to be a major factor. We are doing massive migrations from POWER4,POWER5 and ..yes also POWER6 to POWER7. And furthermore we are more or less only using POWER 770's as target platform for that. That is one sweet machine when it comes to TCO. Ok I have gotten my first POWER 795 Monster.

But as much of the installations is actually downsizing 690's 595's etc that becomes 770's, which are quite cheap compared to the old machines, and the fact that IBM increased UNIX sales in raw $ numbers with more than the rest of the vendors put together. They they have to have been shipping a sh*tload of tin. And just replacing existing machines is IMHO not the whole story.

So to quote a famous coach. "It is what it is".

// jesper

Jesper Frimann
Holmes

Selective reading.

Matt... You kind of read that selective. HP grew their UNIX sales by 6.7% in a marked that rose 20.7%. IBM's UNIX revenue increased by 25.6% and Oracle had a stunning quarter with sales rising 34.4%. It wasn't HP's markedshare that rose with 6.7%.

So HP lost a fair chunk of markedshare, which combined with Oracle's raise in markedshare shows that their foul play towards HP is working. But calculating a bit back then in raw numbers then HP sold for 40 MUSD more in Q1 2011 compared to Q1 2010, Oracle for 156MUSD and IBM for 243 MUSD.

And IBM is pretty close to being as big as HP and Oracle put together in the UNIX marked. They actually might be that looking at Full year 2011.

I don't particular like the fact that Oracle's tactics with drawing support for platforms seem to work.

// Jesper

Do we really want 100Gig Ethernet?

Jesper Frimann
Headmaster

RE:10G territory?

We are slowly getting there.

The standard machines we deploying today are still 1Gbit based, but we are going for 10Gbit on our large clients, on their demands btw.

We still have some 100Mbit connections, but that is normally for appliance type of connections.

// Jesper

Profit piranhas want a bite of HP

Jesper Frimann
Headmaster

Sharks..

Well, Larry drew some blood from HP with his whole "cancel Oracle software stack for Itanium" thing. And when there is blood in the water, the sharks from Wall Street comes sniffing, for some cheap meat. And he did put the fork in the BCS crown jewels, so to speak.

But IMHO there is not a lot of meat to be had. Cause although HP's BCS division is pretty much showing a flat revenue, I personally don't think that HP can back out of that business, as the BCS revenue indirectly gives HP a lot of revenue in other parts of the company

// Jesper

Oracle Solaris 11 to abandon elderly servers

Jesper Frimann
Facepalm

re:re: Lifetime of a server....

With regards to solaris, then historically it has normally had a new release every 2-3 years, the five -six years between Solaris 10 and 11 seems to be a bit of the exception that confirms the rule. This picture illustrates it well:

http://upload.wikimedia.org/wikipedia/en/timeline/3a506e57d970f4bcdc0b3b673b6dd694.png

Oracle is a software company, software companies want to keep short release cycles between major releases to milk the clients. I've put our Oracle business partner dude on it, lets see what he finds out. Most likely gonna be NDA so..

With regards to the hardware, I didn't buy it, back in 2008/2009 I would have bought Mseries, if only to get two vendors supplying the same machines, to play them out against each other. But back then I was working with the competition to where I work now. But I know one thing, back then your SUN sales person was pushing E25K to the end, all while whispering about Rock in your ears. And there were a lot of people that bet on simply skipping 'the temporary fix' which many were led to believe that the Fujitsu made APL line of servers were to be, and then jump directly from SPARC IV+ to Rock.

How wrong they were.

"You are already a generation behind. You should consider upgrading your hardware by 2013-2014 anyway."

A machine normally has a lifespan of 5-6 years. Hence a machine bought in 2009 should be ready for replacement in 2014-2015.

" Your current platform was released in 2002!!! If I recall correctly, the 25K was EOL'd in 2009... "

The E25K is from 2004 and it was not EOL'ed in 2009. SUN stopped selling it in 2009, there is a VERY big difference. You are thinking like a sales guy. You don't buy a server and then 2 1/2 years later OS support for the machine stops. That is simply treating your clients badly.

You must expect that the client buying the machine will depreciate the value over the machine for 3-5 years, hence you keep supporting the machine for at least 5 years (6 is better) after you stopped selling it.

You simply don't do a thing like that, we have machines that will have almost 50% of their deprecation value left when Solaris 11 (non express) tapes out, and we have to look through our contracts to figure out if there is a "within 1 year of a release you have to have the OS upgraded" clause. Cause by then there is still 30% left of the depreciation value of the machine.

And that is somewhere between 1-1.5MUSD for a E25K. So if we have to go and buy new HW, that is actually a loss to us, formally that is. One one account that is a bloody 8 machines.

So forget internet trolls and teasing about products and I like XXX better than YYY.. This is serious real life business. And we are a F****** large Oracle customer, again just in our little tiny country I counted 500+ servers, with 10% of them being high-end machines. So my aim is to mobilize our Oracle relationship dudes and get those guys to put pressure on Oracle to revoke the removal of support for SPARC IV+.

"Oracle support a new OS for an EOL product?"

See above. I hope you get it.

And btw thanks to TPM for taking this up, I would have missed it until the release of Solaris 11, if I hadn't read this article.

// Jesper

Jesper Frimann
Facepalm

re:re: Yet another reason to hate Oracle

AC you are missing the point.

Again SPARC IV+ systems were sold up until Q1 2009, and then an OS released aprox 2 1/2 years later then doesn't support the hardware sold only 2 1/2 years earlier.

If you do want to compare with AIX 7.1 that would correspond to AIX 7.1 not being supported on the POWER5+ based p570, and oh wait AIX 7.1 does actually support POWER4. What AIX 7.1 doesn't support is RS64, which IBM stopped selling back in 2002, that is 8 years for AIX 7.1 (Btw 6.1 didn't support RS64 either, but again you had 5+ years from when they stopped selling the server til they dropped OS support for it.)

2 1/2 years is simply being nasty to your clients, and forcing them to upgrade prematurely.

And you need to get your facts right with regards to SPARC IV+ release dates. Here is for V890

- Announced support for 1.5GHz UltraSPARC IV+ processors in September 2005

- Announced support for 1.8GHz UltraSPARC IV+ processors in August 2006

- Announced support for 2.1GHz UltraSPARC IV+ processors in April 2007

And there is no hate here.. only )/#)(Y(U¤# cause this is gonna cost us money.

// Jesper

Jesper Frimann
Facepalm

Lifetime of a server....

Well I have done a quick scan of our CMDB, and we have around 15 x 25K's with Panther SPARC's in them. And this means that these machines can't be upgraded to Solaris 11. Which is not a problem AFAIR our contracts says current OS version -1 as a minimum. BUT if Oracle then releases Solaris 12 in late 2013 or Early 2014, then we need to upgrade, to at least Solaris 11, within a specified amount of time. And as some of our 25K's were originally expected to last until late 2014/ mid 2015, we will need to shorten the lifespan of some of these servers. Hence this is most likely gonna cost us money, as we will have to depreciate some of these servers faster than expected. Now this could be a loss of up to 1 MUSD per server.

Honestly.. this sucks.

// Jesper

Page: