Simon Haslam's Oracle Fusion Middleware blog

Oracle Database Appliance X4-2 Released

With Intel's release last September of the Ivy Bridge-EP "v2" processors, very quickly followed by Oracle's Sun Server X4-2, the stage was set for some imminent updates to the whole of Oracle's Engineered Systems range. Last Friday Oracle launched the Oracle Database Appliance X4-2 - or at least that's when there are some blogs about it (Oracle Database Insider Blog and eSTEP Blog) though I didn't see a press release.

So what's in this new version? I've pored over the documentation and here are the differences I've spotted:

Processors

The headline difference for the ODA X4-2 is that the processors used are now E5-2697 v2 2.7 GHz / 12 core (instead of the E5-2690 2.9 GHz / 8 core in the ODA X3-2). This is one of Intel's "tick" releases and they have shrunk the lithography down from 32nm to 22nm.

[1-JAN-14: I re-wrote this section after looking at the benchmark figures]

There are some figures being bandied around like a "50% performance improvement over the previous generation" but that is really just a result of going from 8 to 12 cores. For those of us working in the Oracle world, i.e. usually licensed by cores, that's not particularly relevant (other than from an overall server capacity perspective).

I think the processor's integer performance tends to be the most important for enterprise workloads like SOA. If we look at SPECint CPU2006 data that Oracle has submitted for both X3-2 and X4-2 servers used in the ODAs - the SPECspeed, i.e. single thread performance, result{baseline} for each is:

Note the benchmark result depends on the processor and main memory performance, and compiler optimisations. Both sets of results use the same memory (if you're interested: 256 GB as 16 x 16 GB 2Rx4 PC3-12800R-11, ECC, i.e. 1600MHz RAM).

In other words each processors' single thread performance is comparable (a 3% increase - easily lost in the noise) even though the newer ODA model has a slightly lower clock speed.

Both processor models have the similar TDP (130W for the v2 or 135W for the original) and, whilst this is not exactly power consumption, it does imply that the maximum throughput per W will be significantly better for the v2 (due to the extra cores). Or, looking at it another way, if you're only licensing/running, say, 12 cores per node then the ODA X4-2 servers should have significantly lower power consumption (~35%?) than the X3-2. Of course that saving also has to be put into context with those those 20 spinning disks in each ODA disk array and all the other complexities of power consumption modelling.

[end of update]

I had expected to see the E5-2690 v2 (10 core) in the X4-2 as the E5-2697 v2 (12 core) processor is apparently about 25% more expensive, so the inclusion of the 12 core was a very pleasant surprise given that the ODA X4-2 is the same price as the X3-2 it replaces.

Some of you will know about my involvement with O-box and the new O-box SOA Appliance - these extra cores will come in very useful for our larger configurations. :)

Networking

As regular readers may remember I have pointed out that, having nothing but copper 10GbE ports, the ODA X3-2 networking could be a bit too restrictive for some customers (i.e. those who have fibre, either via SFP+ modules or TwinAx cables, into their fabric extenders or core 10GbE switches).

Anyway, I see that the new ODA X4-2 white paper says "Each server also provides the option for 10GbE SFP+ (fiber) or 10GBase-T (copper) external networking connectivity, ensuring the appliance will be compatible with any data center." The servers themselves have 4 10GbE copper ports onboard, and the X3-2 has a dual port copper 10GbE PCI card, used for the interconnect (via crossover cables). What Oracle has done here is let you move the interconnect onto two of the onboard ports and use fibre in a dual port 10GbE SFP+ NIC instead in the ODA X4-2.

Oracle Database Appliance

What appeared curious to me on first inspection was that, according to the documentation, the interconnect is only moved to the onboard NIC for fibre configurations:

ODA X4-2 interconnect when using fibre

...for non-fibre ones it stays on the PCI card, as per ODA X3-2:

 ODA X4-2 interconnect when using copper

I wondered why Oracle would do something so confusing, especially given the management simplicity goals for ODA. However if you have a look at one of my comments on a previous blog post you'll see why I think Oracle have taken this decision: the slot 1 PCIe card has 16 lanes for the two ports, whereas the one controller for two ports on the onboard NIC has only 8 lanes. The interconnect is almost certainly the most heavily loaded network component on the ODA when pushed to extremes so I reckon that's why Oracle chose to put it on the 16 lane wide PCI link. It would be very interesting to do some benchmark comparisons on this! Certainly for O-box SOA Appliance customers I am happy that Oracle engineers keep performance to the forefront of their minds since we want it to be as quick as possible!

Overall I think this is a smart move by Oracle and will no doubt help larger customers who wish to deploy ODA easily too.

Summary

Other than processors and networking I think everything else for the new X4-2 is the same as before: memory quantity (more would fit in but I expect would be quite expensive as it would mean 32GB modules), storage/disk speed/capacities and, probably most importantly, the price (which is pretty competitive if you compare to HP or IBM equivalents even before you consider the engineered system benefits).

All in all I would say this this is a very useful upgrade and keeps the ODA at the state of the art for these kinds of servers.

Comments:

with respect to "the price is the same", it would be be interesting to see a benchmark on (say) a current ODA with 'n' cores enabled versus an ODA-X4 with 'n' cores enabled. Expandability generally counts for naught unless you intend to use it :-)

Posted by Connor on December 17, 2013 at 02:50 AM GMT #

Agreed, though if the V2 processors deliver anything like what Intel promises then the X4-2 ought to be faster even with the same number of cores enabled (and even though it has a slightly lower clock speed).
When I get my hands on an X4-2 I'll try SLOB 2 and Julian Dyke's CPU benchmark to see what they show compared to an X3-2.
As Kevin Closson pointed out on Twitter of course more cores mean more Oracle licences too...
Thanks for dropping by Connor :)

Posted by Simon Haslam on December 17, 2013 at 11:27 PM GMT #

Post a Comment:
Comments are closed for this entry.

Search this blog

About me
Oracle ACE Director (Middleware and SOA)
Presentation downloads



I'm speaking at Oracle OpenWorld


UKOUG Ambassador Partner
Oracle WebLogic Server 12c Certified Specialist
Oracle WebLogic Server 12c Certified Specialist

Links

Feeds