Oracle’s Timeline, Copious Benchmarks And Internal Deployments Prove Exadata Is The Worlds First (Best?) OLTP Machine – Part 1.5

In Part I of this series about Oracle OLTP/ERP on Exadata versus non-Exadata, I took a moment to point out the inaccuracies of a particular piece of Oracle marketing literature. In a piece aimed at chronicling Oracle Corporation history, the marketing department went way out of line by making the following claim regarding Exadata:

[...] wins benchmarks against key competitors [..]

Please don’t get me wrong, those five words appearing in any random sentence wouldn’t pose any sort of  a problem. However, situated as they are in this particular sentence does create a  problem because the statement is utterly false.  Exadata has not won a single benchmark against any competitor–”key” or otherwise.

Along For The Ride
In Part I of this series I pointed out the fact HP Oracle Exadata Storage Server cells (a.k.a., V1 Exadata) were used in this June 2009 HP BladeSystem 1-TB Scale TPC-H. However, merely involving Exadata hardware can hardly support Oracle’s marketing claim vis a vis winning benchmarks against key competitors.

There is a big difference between being involved in a benchmark and being the technology that contributes to the result.

I made it clear, in Part I, that Exadata storage was used in that 2009 HP TPC-H result but none of the Exadata features contributed to the result. I clarified that assertion by pointing out that the particular benchmark in question was an In-memory Parallel Query result. Since the result establish Oracle database performance achieved through in-memory database processing I didn’t feel compelled to shore up my assertion. I didn’t think anyone would be confused over the fact that in-memory database processing is not improved by storage technology.

I was wrong.

In the comment section of Part I a comment by a blog reader took offense at my audacious claim. Indeed, how could I assert that storage is not a relevant component in achieving good in-memory database processing benchmark results. The reader stated:

You give reference to a TPHC that used Exadata and then say no Exadata features were used. [..] You obviously don’t know what you are talking about

Having seen that I began to suspect there may be other readers confused on the matter so I let the comment through moderation and decided to address the confusion it in this post.

So now it’s time to address the reader’s comment. If Exadata is used in a benchmark, but Exadata Storage Server offload processing is disabled, would one consider that an Exadata benchmark or was Exadata merely along for the ride?

Here is a screenshot of the full disclosure report that shows Exadata storage intelligence (offload processing) features were disabled. For this reason I assert that Exadata has never won a benchmark against “competitors”, neither “key” nor otherwise.

The screenshot:

Oracle’s Timeline, Copious Benchmarks And Internal Deployments Prove Exadata Is The Worlds First (Best?) OLTP Machine – Part I

I recently took a peek at this online, interactive history of Oracle Corporation. When I got to the year 2008, I was surprised to see no mention of the production release of Exadata–the HP Oracle Database Machine. The first release of Exadata occurred in September 2008.

Once I advanced to 2009, however, I found mention of Exadata but I also found a couple of errors:

• The text says “Sun Oracle Database Machine” yet the photograph is that of an HP Oracle Database Machine (minor, if not comical, error)
• The text says Exadata “wins benchmarks against key competitors” (major error, unbelievably so)

What’s First, Bad News or Good News?

Bad News

The only benchmark Exadata has ever been used in was this 1TB scale TPC-H in 2009 with HP Blades.  Be aware, as I pointed out in this blog post, that particular TPC-H was an in-memory Parallel Query benchmark. Exadata features were not used. Exadata was a simple block storage device. The table and index scans were conducted against cached blocks in the Oracle SGAs configured in the 64 nodes of the cluster. Exadata served as nothing more than persistent storage for the benchmark. Don’t get me wrong. I’m not saying there was no physical I/O. The database was loaded as a timed test (per TPC-H spec) which took 142 minutes and the first few moments of query processing required physical I/O so the data could be pulled up into the aggregate SGAs. The benchmark also requires updates. However, these ancillary I/O operations did not lean on Exadata feature nor are they comparable to a TPC-H that centers on physical I/O.  So could using Exadata in an in-memory Parallel Query benchmark be classified as winning “benchmarks against key competitors?” Surely not but I’m willing to hear from dissenters on that. Now that the bad news is out of the way I’ll get to what I’m actually blogging about. I’m blogging about the good news.

Good News

The good news I’d like to point out from screenshot (below) of Oracle’s interactive history is that it spares us the torment of referring to the Sun Oracle Exadata Machine as the First Database Machine for OLTP as touted in this press release from that time frame.  A system that offers 60-fold more capacity for random reads than random writes cannot possibly be mistaken as a special-built OLTP machine.  I’m delighted that the screen shot below honestly represents the design center for Exadata which is DW/BI. For that reason, Exadata features have nothing at all to do with OLTP. That’s a good readon the term OLTP is not seen in that screen shot. That is good news.

OLTP does not trigger Smart Scans thus no offload processing (filtration,projection, storage index, etc). Moreover, Hybrid Columnar Compression has nothing to do with OLTP, except perhaps, in an information life cycle management hierarchy. So, there’s the good news. Exadata wasn’t an OLTP machine in Oracle’s timeline and it still is not an OLTP machine. No, Oracle was quite right for not putting the “First OLTP Machine” lark into that timeline. After all, 2009 is quite close to 40 years after the true first OLTP Machine which was CICS/IMS.  I don’t understand the compulsion to make outlandish claims.

Bad News

Yes, more bad news. Oracle has never published an Exadata benchmark result even with their own benchmarks. That’s right. Oracle has a long history of publishing Oracle Application Standard benchmarks–but no Exadata results.

I’ve gone on the record as siding with Oracle for not publishing TPC benchmarks with Exadata for many reasons.  However, I cannot think of any acceptable excuse for why Oracle would pitch Exadata to you as best for Oracle Applications when a) there are no OLTP features in Exadata*, b) Oracle Corporation does not use Exadata for their ERP and c) there is no benchmark proof for Exadata OLTP/ERP capabilities.

Closing Thoughts

Given all I’ve just said, why is it that (common knowledge) the majority of Exadata units shipping to customers are quarter-rack for non-DW/BI use cases? Has Exadata just become the modern replacement for “[...] nobody ever got fired for buying [...]?” Is that how platforms are chosen these days? How did we get to that point of lowered-expectations?

Enjoy the screen shot of memory lane, wrong photo, bad, good and all:

* I am aware of the Exadata Smart Flash Log feature.

Xeon E5-2600 OS CPU To Core / SMT Thread Mapping On Linux. It Matters.

Ages ago I blogged about the Intel topology tool and mapping Xeon 5500 (Nehalem EP) processor threads to OS CPUs on Linux. I don’t recall if I ever blogged the same about Xeon 5600 (Westmere EP) but I’ll cover that processor and Xeon E5-2600 in this short post.  Fist, Xeon 5600.

The following two screen shots are socket 0 and socket 1 from a Xeon 5600 server. Socket 0 first:

Now, socket 1:

So, based on the information above, one would have to specify OS CPUs 0,1,2,3,4,5 if they wanted thread 0 from the first 3 cores on each CPU (c0_t0). I never liked that much. That’s why I’m glad Sandy Bridge presents itself in a more logical manner. As you can see from the following two screen shots, specifying affinity for thread 0 of cores on socket 0 is as simple as 0,1,2,3,4,5,6,7. First, socket 0:

And now, socket 1:

Lest this come off as simple tomfoolery, allow me to show the 2x difference in siphoning off a fifo when the data flows socket-local versus socket-remote:

Be aware that this level of disparity will not necessarily be realized when a server is booted SUMA (nor even when BIOS NUMA is enabled but the grub boot string includes numa=off). I’d test the difference and blog that here but that would just be tomfoolery