It would be interesting to see if HPE 3Par's consistent performance delivers its promises when controllers have to do garbage collection on SSDs. It was the reason why my customer ran out of space on 8450 during data load... and overfileed the array :(
33 posts • joined 5 Oct 2009
Re: Oh dear
Exactly. And once you want a snapshot / flash copy, you still need to do copy-on-fist-write through back-end FC, transferring blocks back and forth.... all those flash optimizations are wasted with such architecture. Not to mention lack of dedupe.
Re: If you are not already tied in
Two things that you probably misunderstood:
1. Oracle Real Application Clusters is not a "shared-everything" architecture. It's a shared disk architecture.
2. RAC does need replication. It does not offer any database corruption protection / site protection itself. Take a look at Oracle Maximum Availability Architecture. It's RAC _with_ Data Guard.
IB controller for memory scans..?
> M7 on doing SQL memory scan can do 120 GB/sec, whereas an x86 runs queries at ~5Gb/sec (in both cases it is max theoretical speed) .. and this is why in M7/Sonoma CPUs is build IB controller.
I'm sorry, but you probably miss some basic CPU architecture concepts. You do not need IB to do memory scans. In fact you don't do any I/O to access RAM.
You should compare systems carefully instead of believing what Oracle marketing says. Compare socket to socket performance or $ to $ performance. 5Gb -- did you mean 5 gigabits or gigabytes? You can scan more than 5 gigabytes /s on modern notebook.
Re: Still using eMLCs?
Happy potential customer of XtremIO here... it depends on price I guess :)
In Dec 2014 my company tested XtremIO minimal config (I believe it;s the smallest possible) -- 13 disks, with our application transactional workload, Oracle based. Although I can not say what the max storage performance is, because we saturated all server CPUs before killing the array, we managed to achieve 117k IOPS at about 0.4ms latency with 1) majority of 8kB block size, 2) read/write ratio about 70/30. According to array management software, we were at about 45-50% XtremIO utilization. With low I/O workload, the latency was floating around 100-200 usec, same block size. Someone has cited 6000 IOPS with 3PAR 7450. Sure, it's possible, but this is a workload a single SSD drive can do.
The performance is only half of the story. At the same time we had 7.2:1 data reduction rate (3 copies of our application and database, compression ratio was around 2.5:1, dedup ratio 2.8:1, some benefit from thin provisioning).
Have you ever heard of Avamar and DataDomain? how about Mozy?
> For comparison Pure Storage's FA-450 scales to 70TB raw, and a claimed 250TB effective capacity after deduplication.
But... if you have a large database to fit into this capacity, it WILL NOT dedupe at all! Every single 9kB Oracle block is unique.
> All VNX2 models appear to be affected: VNXe3100 and 3300, VNX 5100, 5300, 5500, 5700 and 7500.
OK, but those are all 1st gen VNX models, not VNX2!
Re: nothing special
Really? Which other block array does dedupe?
Re: Ibm marketing...
> But nobody can buy cores
you are obviously wrong. With IBM Power, you pay per core license, which is 80-90% of total core price. So you may have 2 8-core P7s and 11 licenses if you wish. Optimal for system price and for software licenses.
Fastest ^H^H^H^H^H^H^H processor..?
Congrats to Oracle!
They finally beat almost 5-year old IBM TPC-C result for non-clustered configuration. I'm wondering if IBM would benchmark big POWER7 system, but probably they will not.
From the processor speed perspective and Oracle's claims, T5 core is still slower in tpmC than POWER6 core by 33%... Interesting how good IBM's CPU was, especially when you compare POWER6 65nm litography with 28nm litography of T5 :)
> Two of those eight bays are used for a pair of 200GB SSDs (again for the database redo logs)
Who uses SSD for redo? Redo logs are sequentially written, so this is waste of cash. There is no random IOPS! All other storage vendors recommend using 15k HDDs for redo logs. Much chaper, same performance.
Re: Itanium dead, huzzah!
> We have replaced 1.500 HP servers with one single IBM P795 servers.
Could be. But I'm sure you are aware, the 795 has a limit of 1000 logical partitions?
> These 64 cpu monsters gives us more performance for a fraction of the price.
IBM p795 has a maximum of 32 CPUs.
looking from performance and TCO perspective...
I'm wondering who wants to run Oracle on Itanium with ridiculous license core factor of 1.0? Given this and performance metrics, there is no sense at all to stick with itanium platform unless the application is not available on other cpus (which is not the case with Oracle database).
Is there any benchmark on the planet where itanium is faster or equal to POWER or x86?
> The chips support 20 logical partitions (IBM's name for what everyone else calls a virtual machine)
TPM, everyone else? Oracle calls them LDOMs.... :)
> The Power 750+ server has six PCI-Express 2.0 slots
Looks like the PCI-X is finally gone for good!
Re: what is a controller?
1 engine = 2 directors = 2 controllers
8 engines equals 16 directors (16 physical controllers).
Each 40k director may have up to 12 cores. Thus max config for VMAX 40k is 192 cores.
Re: Take no bets
Phil, I've already seen many roadmaps from IBM and Intel, with much more details on upcoming chips.
BTW, you say "SPARC T4 delivers almost 50% faster Siebel performance and 41% faster Java jvm performance". Cool. Now look at the picture with John Fowler and the slide titled Outpacing the Competition. "SPARC: 2x performance each generation". Shouldn't it make you wonder?
On the other hand performance improvement of POWER6 to POWER7 was about 5x... not close to what tis presentation says.
Re: Take no bets
Maybe you're right. Competition is good for all customers.
But look again at the Oracle roadmap. It does not commit to anything. There is no functionality or any basic description of each future chip. What do they commit to? A name of the chip? Do we know at least the number of cores or threads? +2x Throughput? Throughput of what? Of SPARC III? They can make up basically anything to fit this "roadmap".
Re: Smoke and Mirrors anyone?
> No SPEC benchmarks released for the T4. Wonder why.
Hmm... let's think. Same core/core performance as T3?
And another news: so much for Oracle promises. Last year Larry promised T5 processor and systems to be announced in 2012. OpenWorld is over. No traces of T5. 2013?
Re: How ?
> All high clocked cpus must have deep pipelines.
have you checked pipeline and clock frequency of power?
"...and linked with an IBM Power7 SMP chipset. (It is not entirely clear where IBM hides this chipset, but it is possible that the Power7 architecture supports glueless connections across four processor sockets.)"
What chipset? Power7 systems support glueless configurations up to 32 sockets. The same applied to Power6 systems.
purescale on linux...
"There was talk in October 2009, when PureScale was announced as Oracle was Sunning up its Exadata clusters, that PureScale would be ported to Windows and Linux systems, but this has not happened."
Or it did happen? It's good to check before posting the article. DB2 pureScale on System x servers:
"Again, if you are doing it right, you never use floating numbers in finance. Every calculation is done with integers, and you keep track of the number of decimals separately."
Wrong. Read about Decimal Floating Point and there is a chance you'll understand. Available only in POWER and Z.
Power7 in 2008?? Power7+ in 2009???
> A case in point is one of the earlier schedules for the Power processor lineup, which had Power6 coming out in 2006, Power6+ in 2007, Power7 in 2008, and Power7+ in 2009.
That's interesting. I was following such roadmaps and have never seen the one with Power7 coming in 2008 or Power7+ in 2009. Perhaps it's available anywhere?
SPECint_rate is not single strand, but we will see... if only Oracle decides to publish any industry standard benchmark :)
> those customers just change the hardware to either be Sun boxen or x86 derivative
Take a deep breath before next post and just relax :) you make a lot of mistakes.
Niagara T3 did not break any record. It was 108 Niagara T3s. Single one of them is too slow to beat any competitor.
Re: No need to speculate ...
Nice read. Lots of "innovations" by marketing dept:
- out of order, wow!!
- 28nm in T5 (intel, amd and ibm will have 16nm by then) :)
- "we do have a unique situation here where the operating system, processor development, virtualization software, and the applications all under one roof"... unique? except they do not own a single chip fab, so they depend on other's technology (taiwan), Oracle cannot innovate much here. IBM has everything from chip fab to applications.
- "virtualization software that can offer 100s of logical domains" -- reminds me of power5 partitions back in 2004... or mainframe lpars in 1980 :)
> In that new Oracle roadmap above the company is promising "+3x Single Strand" performance, which we take to literally mean that the clock speeds will be three times those of the past Sparc T2+ or now current Sparc T3s. That means between 4.8 GHz and 5 GHz clock speeds
I doubt they can bump the clock that much. In my opinion they will redesign/change multithreading model. With T2 and T3 the single thread performance is poor due to CMT constraints.
Sparc T3s are superior to x64?
> Oracle doesn't just need to offer Solaris 11 as an option on the x64-based systems, but to use the Exadata appliances to prove that Sparc T3s are superior to x64 processors.
The problem is.... they are not faster than x64 :-)
> For instance, this ExaLogic machine costs 25% of IBM POWER 795 - and this ExaLogic machine is faster.
That's very interesting. Where did Larry get 795 performance numbers? Can you find any benchmarks? This system has not been benchmarked yet, but of course Larry had to release marketing bullshit, same as a year ago. I'm wondering why Larry did not mention first TRUE XX MILLION TPC-C record... achieved by only 3 (not 12) IBM POWER7 (10,3 M tpmC) systems using half of the cores of Oracle's fastest configuration (7,6 M tpmC result) and with price/performance twice as low as Oracle's one.
If you are so precise, you should clarify the prices of Exalogic DID NOT INCLUDE SOFTWARE licenses, as was mentioned on the Larry's slide at the very bottom. We all know that Larry charges for software 10x the hardware cost.
Regarding lockins -- remind me how many OS-es run on SPARC? I heard of only one. Linux support is a joke. On the other hand I heard you can run both AIX and Linux on IBM Power, with support for DB2, WebSphere for Linux version.
And here is another question for you: why there is no Exadata software or Exalogic software available for installation any x86 servers? I want to test it, install on my x86 system from HP. Were you telling something about vendor lockin? There is even no Exadata documentation available for general download, not to mention ANY public benchmark result!
One last: tell all of us why Oracle decided to go with x86 for Exalogic and not with SPARC. I heard they were buzzing Niagara was so superior for Java workloads... the fastest chip for Java, yet they decided to go with x86, which has to be obviously a slower CPU. Interesting.
Kebabbert, you need to educate a bit more before posting such a mess.
Clock frequency has little in common with CPU performance in this case. Single Power7 core is faster than single Power6 core, because P7 has out of order processing.
Other assumptions of yours are wrong as well. Reading Sun whitepapers is not good enough.
Power's processor have since a long time missed their release schedule, and only recently did they announce the capacity to have more than two cores with POWER7
Could you please reference some roadmap here?? AFAIR Power7 was schedued for 2010 many years ago.