Elisabeth Stahl on Benchmarking and IT Optimization

The National Security on the T5-4 and Big Data

with 5 comments

There’s been a lot of talk the last few days on Big Data and when it’s “right” to capture and use it. Some say it’s a real invasion of privacy. Others realistically point out that it is the best way to counter terrorism.

Whichever you believe, the important thing is that Big Data is being discussed not just in geeky meetings with IT managers but by everybody. When your neighbor across the street stops trimming his tree branches just to talk to you about it, you know it’s hot stuff.

So I was particularly interested to see that Oracle just published a new TPC-H data benchmark result on the SPARC T5-4.

And here is what hits you like a train.

  • Why is this publish at only the 3TB size when all the talk these days is on much larger amounts of data?
  • Why is the Total Storage to Database Size ratio a whopping 29? Talk about overkill on storage to achieve performance. This number is many times the ratio we’ve seen from other results.
  • Why is the memory to database size % a whopping 66.6? Again, much more than you should need and what we normally see.
  • Why are there 192 query streams needed? Most results use many, many fewer. That’s because TPC-H has a limited number of query variations; so when you run a lot of streams, you have a high probability that the same queries will be requested more than once. Oracle is greatly increasing the probability that they will have the results of the queries stored in their cache — which may not be representative of how their product would perform in a truly ad hoc query environment.
  • Why isn’t the configuration available now? Because key elements of the storage are not ready.
  • Why did Oracle once again include extremely minimal support in their pricing? Does $2300 a year sound like what you are paying for software “incident server support” . . . ? You don’t even need to answer this one.

Comments are welcome at your own risk.

(1) Oracle TPC-H of 409,721 QphH@3000GB,$3.94 per QphH,Availability 09/24/13,Oracle Database 11g R2 Enterprise Edition w/Partitioning,SPARC T5 3.6 GHz; Total # of Processors: 4,Total # of Cores: 64,Total # of Threads: 512.
Source: Results current as of 6/12/13.
TPC-C ,TPC-H, and TPC-E are trademarks of the Transaction Performance Processing Council (TPPC).

The postings on this site solely reflect the personal views of the author and do not necessarily represent the views, positions, strategies or opinions of IBM or IBM management.

technorati tags: , , ,,,,,,,,,,,,,,



Written by benchmarkingblog

June 12, 2013 at 3:36 pm

Posted in SPARC T5, TPC-H

Tagged with , , ,

5 Responses

Subscribe to comments with RSS.

  1. Good thing you didn’t do any comparisons to the Power780 results! And why hasn’t IBM published *any* TPC-H results in the last 2+ years??? No Power7+, no high end Power7? Hmmm. Do Power7+ servers perform poorly on datawarehouses?


    June 13, 2013 at 11:59 am

    • I can see why you would want to pick the 780 server to compare yourself against. They are the bootstraps which you are trying to pull yourself up to compare against. As a IBM business partner who works with “competitive” customers almost daily I would not need a 780 to compete against any of the T5 offerings. I would consider the 16 or 32 core Power7+ servers called 740 and 750 respectively. The 740 provides 16 cores and up to 1,280 threads while the 750 delivers 32 cores and up to 2,560 threads……that must be important because Oracles webpage for the T5 products says the T5-8 has 1024 threads. Don’t mis-understand me…..the T5 is finally respectable. It took 5 generations to get there. However, it lacks a robust, flexible and scalable hypervisor with RAS features more akin to a x86 server than good ol’ Big Iron RISC based servers. Also, not only are my two servers less expensive to acquire, but also for customers who run “real world” workloads and not the artificial benchmarks from all manufacturers it has a lower TCO. Power customers regularly run dozens (and dozens) of VM’s (ie LPAR’s) on their servers guaranteeing each VM a quality of service, each securely isolated from one another while giving the user the ability to dynamically adjust up or down cpu, memory and I/O as needed not to mention the ability to use the “EASY” button to live migrate VM’s (called Live Partition Mobility or LPM) to other Power servers for upgrades, when the electrician says “We have to move this rack so shut everything down”, etc. Use LPM from a 710 to a 795 server or from a 720 to a Flex p460 node…..doesn’t matter. To summarize, buy SPARC servers if you want discrete, expensive servers because you love Solaris. That is fine. But, don’t try to compare your entry level T5 servers to a Enterprise Power7+ 64 or 128 core 780 server running at 4.42 GHz. Any time Larry wants to send you out to the midwest to race your best vs my best Power server just drop me a note. Please know that I will brag to everybody I know about the results! No lawyers allowed – just customer workloads – OLTP, DW (yes, I said it), BI, in-memory DB!! When we are done I’ll buy the first round so you know we can still be friends! 🙂

      Brett Murphy

      September 6, 2013 at 9:04 pm

  2. This is incredibly disingenuous. Here was IBM using 10,000+ disks for their TPC-C benchmark on p595 and now you are nitpicking? The amount of memory doesn’t bother me, if its cheap then go ahead and use it. Secondly, should one be nitpicking on the 80MB cache in Power7+ processor? What matters is the total cost for getting the job done, each vendor goes about it their own way to highlight their offering. I’d like to see IBM deliver some benchmarks, for what they are worth. I for one use benchmarking to understand each vendors approach.


    June 18, 2013 at 1:20 am

    • Benchmarks are not a good indicator of TCO. They are a artificial measure of a servers capabilities often propped up by benchmark trickery (not meant with malice but with with experience to tweak results that are not realistic in the real world). You see it in one way or another from all manufacturers. Oracle uses unrealistic maintenance/support plans (per incident) or short term Oracle licenses vs perpetual that most customers buy. I recently looked at some HP 3PAR benchmark results and was laughing at what they did to achieve their results to support their claim of superiority.

      This is why we like to do proof of concepts with customers so they can apply the technology against their actual workloads. From a Power perspective, how does a benchmark show you a 8 core SPARC M5000 running at 100% utilization with the same workload on .5 (yes, 1/2 of a core) on a Power7 server. Or, how does a benchmark demonstrate that a Oracle EBS project required 300 x86 cores compared to 28 on a Power7+ server. The Power server runs all of the workloads on one server while the x86 solution required over a dozen discrete servers….and the math for those 2 & 4 socket servers adds up to 300 – Yes, Cisco wasn’t happy when they learned they were competing against Power. Benchmarks don’t show this and the TCO results would thus not reflect this either.

      Brett Murphy

      September 6, 2013 at 9:19 pm

  3. @philoracle – instead of deflecting the point of the article by trying to draw in other products, can you answer the questions. Why did Oracle cherry pick this benchmark using unrealistic features?

    Brett Murphy

    July 24, 2013 at 1:38 pm

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Google+ photo

You are commenting using your Google+ account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )


Connecting to %s

%d bloggers like this: