BM Seer Unofficial thoughts from an anonymous Sun employee

MySQL Consolidation on Sun SPARC Enterprise T5220

Wednesday Nov 05, 2008

Sun's CMT provides huge savings in Cost, Energy, and Datacenter Space!

Twenty-four Sun SPARC Enterprise T5220 servers (UltraSPARC T2 processors) were able to consolidate a customer's workload of MySQL databases which were running on over 250 Dell 2950 servers. For this medium-weight OLTP customer workload, the solution was architected to handle query distribution at the application layer. In each of the over 700 instances of MySQL, up to a 4 GB database was used.

I've inserted some updates & corrections after the initial posting.

Now this isn't an official benchmark (SPEC, TPC,...) but it was a workload comparisons that those benchmark consortiums do not cover. All vendors sometimes do this to show capabilities. (sidenote: it is funny to see IBM bloggers smear Sun for doing this and then you see IBM do this same thing as well - duplicitous behavior?). Anyway if you want to see official CMT results on standard benchmarks go to: T5220 benchmarks on sun.com

This following test was run as a customer request to do a fair an complete comparison to judge the effect of a possible upgrade. They upgraded!

Only 24 Sun SPARC Enterprise T5220's are required to consolidate the same workload that required 251 Dell 2950 servers running Linux. Sun's CMT solution required 10.5x times fewer servers.

CMT servers can easily consolidate many MySQL instances into a single server running the Solaris Operating System. No additional virtualization software was needed for this consolidation.

The Dell 2950 solution requires 10.5x more rack units than the Sun SPARC Enterprise T5220.

The Sun SPARC Enterprise T5220 also uses 8.4x less power than the Dell 2950 solution, which amounts to a yearly savings of $115,000 in electrical costs (assume $0.13/kWh).

Both configuration solutions produced the same level of performance and response time.

Customers are interested in consolidating workloads that were originally created on X64 platforms. This customer workload which consisted of a heavy MySQL database & light-weight Java application was used to compare a Dell X64-based solution to Sun's CMT-based servers.

Performance Landscape as of 10/22/2008.

Systems required to reach same performance level with same response time characteristics.

Server MySQL
DB Instances		 # Servers Total RUs Total Watts Sqft Needed
(200w / sqft)		 Annual
Power Cost
@.13/kWh
Dell 2950 700 251 502RU 114,707w 574 sqft $130,638/yr
Sun T5220 700 24 48RU 13,680w 68 sqft $15,579/yr
Sun Advantage
10.5x 10.5x 8.4x 8.4x 8.4x

This benchmark is based on actual customer workload. Each server configuration was driven to meet identical use, throughput, and response time characteristic.

Benchmark Description

The test simulated real-world requirements of a large organization's use of hundreds of MySQL instances. For this workload the customer solution is architected to handle query distribution at the application layer. Up to 4GB database per instance are used in the 700 MySQL instances.

Disclosure Statement:

MySQL consolidation comparison between Sun SPARC Enterprise T5220 (1 x UltraSPARC T2 1.2 GHz, 32GB) and Dell PowerEdge 2950 (2 x Xeon 5148, 16GB) October 15, 2008. Consolidation of 700 MySQL database instances from 251 Dell 2950 systems to 24 Sun T5220 systems, delivering same performance and response time.

after note: The watts above are measured watts that were validated by the customer who has implemented the Dell config and wants to upgrade it now to Sun in order to save money, Below are references to the power calculators so you can explore other options:

SPEC is a reg tm of Standard Performance Evaluation Corporation, for more info see www.spec.org. TPC is a registered trademarks of the Transaction Processing Performance Council (TPC). More info at http://www.tpc.org/

Results Summary

Results:
700 MySQL Instances
Reference Date:
October 15, 2008
Systems:
24 x Sun SPARC Enterprise T5220
Total Number Processors:
24
Processor/GHz of Server:
UltraSPARC T2 1.2 GHz
Operating System:
Solaris 10 8/07
Software:
MySQL 5.1.26 InnoDB


Java SE Development Kit (JDK) 6 Update 1
Note this workload primarily focuses on MySQL performance which mirrored the customer's actual workload. The Java version used is the actual version in use right now by the customer. Of course newer versions of the JVM may improve performance but the customer confirms it wouldn't have a huge effect on these results.

[11] Comments
Like this post? del.icio.us | furl | slashdot | technorati | digg
Posted at 08:17AM Nov 05, 2008 in Benchmarks
Tags:
Comments:

Because of slow threads T2 and locking Sun can't show great result in TPC-E ? Any 32 core x86 server is adequate competitor for T5220 ...

Posted by Triffids on November 05, 2008 at 09:07 AM PST #

The T5220 is 2U yes? That being the case, wouldn't 24 of them be 48 rack units not 47? Unless the Dell units are fractional rack units in height, how do 251 of them end-up as 501 rack units? Shouldn't that be 502 rack units rather than 501?

What of the storage?

Posted by rick jones on November 05, 2008 at 09:11 AM PST #

Rick, I think that other rack unit is storage, I'll try to find out and update it.

OK Triffids your seem to focus on threads not the sum total of what a server does, so if you think threads & cores are such a focus and IBM's cores are faster, then why doesn't IBM publish on TPC-E. The only results on TPC-E are SQLserver on X64 so little pickup at this point. I think Oracle & DB2 results should be published on this benchmark.

You've seen lots of results showing 2-chip CMT beating 4-socket QC and 4-socket 6C X64 so instead of ... click on the link on the posting and see real results.

Posted by BM on November 05, 2008 at 09:16 AM PST #

Slow threads = longer locks in OLTP. Maybe what's why Sun provide such strange OLTP with 700 instanses.
8-socket x86 from IBM (48 Cores) much faster on SAP-SD than CMT.

>click on the link on the posting and see real results.

Can't find "a payroll process in Oracle's E-Business Suite Payroll 11i with Oracle 10g database running on an 8-core Sun SPARC Enterprise T5220 server" at http://www.oracle.com/apps_benchmark/html/results.html
could you help ?

Posted by Triffids on November 05, 2008 at 09:56 AM PST #

There isn't an "extra" rack unit, both calculations are _short_ a rack unit. 24 * 2 is 48 not 46, 251 * 2 is 502 not 500, so 47 and 501 rack units are short, not long.

Posted by rick jones on November 06, 2008 at 08:06 AM PST #

I see that the JVM was 6u1, that's from 03/2007. (Even older performance-wise, considering that 6u1 was a pure bugfix release with zero optimization work.) You could certainly gain some extra perf with the latest updates, preferably the Performance Releases. Of course this is a "real" test on a large client's datacenter and production people are typically slow to endorse bleeding edge runtimes... still, other components like Solaris and MySQL are more recent updates. Somebody didn't do their job right ;-) one could perhaps skip one Update release or two; but there's NO excuse for using JVM technology almost two years old, remarkably in a first-tier Java platform like Solaris.

Posted by Osvaldo Pinali Doederlein on November 06, 2008 at 08:16 AM PST #

rack units updated, typo fixed in text.

Also included note about JVM version used.

Posted by BM Seer on November 06, 2008 at 09:29 AM PST #

So, what _was_ the storage for all that?

Posted by rick jones on November 06, 2008 at 06:56 PM PST #

I know the storage was specified by the customer's actual workload. I've been told that detail is coming, I will post soon.

Posted by BM Seer on November 07, 2008 at 07:42 AM PST #

Wow, you must have some naive customers that trust you to do a "fair an complete comparison" of your own hardware to someone else's!

There is very little actual detail; as usual, the details will be "coming soon". Was this actually measured with 24 T5220's, or extrapolated from measurements of a single server? Why use old dual-core processors on the Dell system that aren't even shipped any more, and half the memory? I'd love to see all the details including the storage backends, software versions for both systems, along with all the extra tuning that was done on the Sun systems.

I like how you included the SPEC boilerplate even though no use was made of SPEC benchmark results, and threw it close to a SPEC-like disclosure in bold so it looks like you are actually quoting a SPEC benchmark rather than the usual Sun bogus-mark.

Still waiting for Sun's SPECpower_ssj2008 results... coming up on a year now since the Java-based benchmark was released. Isn't this processor supposed to be wonderful for Java applications?

Posted by Nonion on November 11, 2008 at 06:39 PM PST #

Nonion: Do NOT be insulting. That customer is very knowledgeable.

The customer is running their Dell configuration now, using their intelligence and help from that other vendor. That configuration of 251 servers is one small part of one of their datacentres. They were choosing to upgrade.

It seems you are trying to create FUD about the info here.

Please look at the tag cloud and the VAST AMOUNT of information I post on benchmarks.

The spec boilerplate is included because I made references to:
"Now this isn't an official benchmark (SPEC, TPC,...)"

The customer held us to very precise standards on this test but as with most customer they don't want all of their info revealed.

So who do you work for?

Can you compile a list for me of servers that use a more realistic 32GB per server in SPECpower_ssj2008? Or a server that doesn't use exotic LV-DIMMS? or a SPECpower_ssj2008 configuration that was used in another other benchmark?

Why doesn't any other vendor show the the actual power used on a wide variety of SPEC benchmarks on realistic configurations? Sun show this.
see the tag cloud.

Posted by BM Seer on November 12, 2008 at 06:22 AM PST #

Post a Comment:
Comments are closed for this entry.