THE SQL Server Blog Spot on the Web

Welcome to - The SQL Server blog spot on the web Sign in | |
in Search

Joe Chang

Barcelona SAP SD 2-tier

Initial Barcelona performance results are out. The benchmark is SAP 2-tier, which is not surprising. I do not know this benchmark well, but from what I have seen of SAP installations, it is network round-trip intensive. Not that the network bandwidth is high, but SQL Server can only handle so many RPC stored proc calls for single row index seeks/sec. On a 4-way Xeon (NetBurst) 2.7GHz without network tuning (interrupt handling, port affinity alignment), that was something like 27,000/sec. I do not have more recent numbers based on Core 2 architecture. Opteron was very good at this because of the low memory latency (23K for 2 cores?), and TCP/IP packet handling requires this.

System            Processors                               Users               SAPS

BL685cG5       4 x 2.3GHz Opteron 8356         3,524               17,650

DL580G5        4 x 2.93GHz Xeon 7350            3,705               18,530


DL385G5        2 x 2.3GHz Opteron 2356          2,102               10,520

DL380G5        2 x 3.16GHz Xeon 5460             2,436               12,180

(HP ProLiant systems with Quad-Core Opteron or Xeon processors)


The 4-way QC Opteron is closer to Xeon than the 2-way, in part because the X7350 is 65nm like Opteron, but also because the 4-way Xeon only has 4-wide memory, while the 4x Opteron is 8-wide, two per socket. I think Opteron would have matched Xeon at 2.4-2.6GHz.

I expect the 2-way TPC-C will still favor Xeon by a wider margin. I am not happy with the 4-way Xeon TPC-C of 407K on SQL, Opteron 2.3 should match this. Intel can put out higher clock, but not without blowing the 130W TDP. Several years ago, no tier 1 vendor believed the Intel estimates for power, and always allowed headroom for higher clock, but I do not know if this is still the case. The fan noise is really annoying now. Intel does have Dunnington coming out soon with 6 cores and the big L3, which also helps high call volume apps, but I am concerned the 7300 chipset does not have the memory transaction rate, ie, channels, to realize the full potential of 24 cores.

Well there is now a 4-way TPC-C for quad-core Opteron 2.3GHz of 403K. This is actually very good compared to the 4-way Xeon 7340 2.93GHz result. I do not expect Intel to push the 65nm Core 2 architecture any further, and there is reason to believe AMD should be able to push Barcelona higher at 65nm. My expectation is Intel will base their strategy around the 45nm Dunnington 6-core. I am curious as to whether Intel might push the 45nm X5460 higher, as the dual core variant runs at 3.33GHz, and thats not the top it can do, or would Intel just sit until Nehalem is out.  

Published Monday, March 31, 2008 1:08 AM by jchang
Filed under:

Comment Notification

If you would like to receive an email when updates are made to this post, please register here

Subscribe to this post's comments using RSS



Linchi Shea said:


I think the two numbers for Opteron should the ther other way around. In addition, shouldn't the benchmark results be measured in SAPS instead of the number of users you seem to be quoting?

May 24, 2008 11:50 PM

jchang said:

you are right, I switched it. I did not quote units

July 7, 2008 4:39 PM

Leave a Comment


About jchang

Reverse engineering the SQL Server Cost Based Optimizer (Query Optimizer), NUMA System Architecture, performance tools developer - SQL ExecStats, mucking with the data distribution statistics histogram - decoding STATS_STREAM, Parallel Execution plans, microprocessors, SSD, HDD, SAN, storage performance, performance modeling and prediction, database architecture, SQL Server engine

This Blog


Privacy Statement