BM Seer Unofficial thoughts from an anonymous Sun employee

Sun Fire X4275 Sybase IQ TPC-H 1000GB World Record Price Performance, Non-Clustered

Tuesday Apr 14, 2009

This TPC-H result demonstrates that the Sun Fire X4275 server, powered by 2 Quad-core 2.93 GHz Intel Nehalem X5570 processors, using only 12 internal disks (SAS 300GB 15K RPM), achieved a QphH@1000GB of 23,365 with a price performance of $2.41. This is the best price performance among all non-clustered server results at 1000GB.

Best price/performance among all TPC-H results at 1000GB, 70% better than the previous best (Sun Fire X4500) and 75% better than the previous second best ie. the HP DL585.

It is the Best 2-chip or 2-socket server result, even better than many 4-sockets servers.

To put this result in perspective, the best non Sun single server submission at 1000GB was the HP Superdome. The Superdome achieved  a QphH of 69,999 (about 3 times the Sun Fire X4275 performance) BUT:  it required almost 100 times the number of disks, more than 35 times the price and 8 times the number of cores when compared  to the Sun Fire X4275 configuration!

Once again, the Sun/SybaseIQ combination has produced a self-contained (i.e. a server without any external storage or external processing engines) data warehousing solution. Only Sun has the hardware and expertise to produce such TPC-H results. To date, Sun has published self-contained TPC-H results for each of the 100GB, 300GB, 1000GB and 3000GB scale-factors.

This is a extremely compact solution requiring only 2 rack units in total. Again contrast the Sun result with the HP Superdome, using 97 storage arrays at 3 RU each plus a 48 inch cabinet for the server.

Extremely efficient power consumption; peak power consumption throughout the entire benchmark run was 825 Watts with an average consumption of 750 Watts.

{humor: Any comments from HP or Dell or IBM why they never publish watts on any standard benchmarks with real size memory (i.e. anything above 16GB) ? } I'll take comments from incognito HP, IBM, or Dell employees below, as always. :)

Performance Results

In order to put the Sun Fire X4275 TPC-H result in perspective, the table below shows the top non-clustered TPC-H@1000 results from Sun, Bull and HP in ascending order of  $/QphH as of April 14, 2009.

System
CPU

so/
co/
th

DB

QphH

$/QphH

Price
$USD

# Disks

Avail-
able

Data
Ratio

Sun Fire X4275, 72GB
Intel X5540, 2.93GHz

2/8/16

Sybase IQ

23,365

2.41

56,263.91

12

4/14/09

3.5

Sun Fire X4500, 64GB
AMD Opteron 2.8GHz

2/4/4

Sybase IQ

5,604

8.11

45,439

48

10/15/07

11.2

HP DL585 G2, 32GB
AMD Opteron 2.8GHz

4/8/8

SQL Server

14,773

9.73

143,736

206

4/25/07

7.8

Bull Novascale 3045, 64GB
Itanium 1.6GHz

4/8/16

SQL Server

12,087

12.56

151,870

160

3/6/07

5.7

HP DL585 G1, 64GB
AMD Opteron 2.4GHz

4/4/4

SQL Server

10,493

13.83

145,264

164

3/2/06

6.4

HP Superdome

32/
64/
64

SQL Server

69,999

28.69

2,008,168

1198

6/18/07

40.63

Legend:

so/co/th = sockets, cores, threads
QphH  = Overall TPC-H Composite Metric (bigger is better).
$/QphH  = TPC-H Price/Performance metric (smaller is better)
Data Ratio = Total disk to actual data ratio

Complete benchmark results may be found at http://www.tpc.org.

Benchmark Description

The results reported here were performed on a Sun Fire X4275 system and used Sybase IQ as the database manager. Sybase IQ is a special product designed specifically for data warehousing applications. Sybase IQ was developed as a totally separate product from the more widely known Sybase database management system (Sybase Adaptive Server).

The TPC-H benchmark is a performance benchmark established by the Transaction Processing Council (TPC) to demonstrate Data Warehousing/Decision Support Systems (DSS). TPC-H measurements are produced for customers to evaluate the performance of various DSS systems. These queries and updates are executed against a standard database under controlled conditions. Performance projections and comparisons between different TPC-H Database sizes (300GB, 300GB, 1000GB, 3000GB and 10000GB) are not allowed by the TPC.

TPC-H is a data warehousing-oriented, non-industry-specific benchmark that consists of a large number of complex queries typical of decision support applications. It also includes some insert and delete activity that is intended to simulate loading and purging data from a warehouse. TPC-H measures the combined performance of a particular database manager on a specific computer system.

The main performance metric reported by TPC-H is called the TPC-H Composite Query-per-Hour Performance Metric (QphH@SF, where SF is the number of GB of raw data, referred to as the scale factor). QphH@SF is intended to summarize the ability of the system to process queries in both single and multi user modes. The benchmark requires reporting of price/performance, which is the ratio of QphH to total HW/SW cost plus 3 years maintenance. A secondary metric is the storage efficiency, which is the ratio of total configured disk space in GB to the scale factor.

The QphH composite metric is the Geometric Mean of 2 components: (1) a single user component, called Power, and a (2) multi-user component, called Throughput.  Power is a performance measurement of a single user stream of 22 queries, one batch insert and one batch delete, all run serially. The Throughput metric, instead, consists of essentially N concurrent Power streams (or “users” submitting queries), where N is a minimum number of required streams dependent upon the database size. For example, at 300GB, N must be at least 5 and at 300GB N must be at least 6. Both Power and Throughput are calculated metrics and each is inversely proportional to the queries elapsed time: thus the faster the queries finish, the larger the metric becomes and the better the result.

Disclosure Statement:

TPC-H, QphH, $/QphH are registered trademarks of the Transaction Processing Performance Council (TPC). More info at http://www.tpc.org/. Sun Fire X4275 23,365@1000GB, $2.41/QphH@1000GB, available 4/14/09.

Results Summary

Audited Results
  Database Size:   1000 GB (Scale Factor 1000)  
  TPC-H Composite:   23,365.3  
  Price/performance:   $2.41  
  Available   4/14/09  
Number of Systems:   1  
Total Number Processors:   2  
Total Number of Cores   8  
Total Number of Threads   16  
Processor/MHz of Server:   Intel Nehalem 2.93 GHz X5570 Quad Core  
Storage:   12 x 15K SAS drives (all internal)  
Database:   Sybase IQ 15  
Operating System:   Solaris 10  
Total 3 year Cost:   $56,263.91  
Other Performance Metrics      
  TPC-H Power:   29,824.6  
  TPC-H Throughput:   18,304.9  
  Database Load Time:   5 Hr 39 Min  
  Storage Ratio:   3.35  

[1] Comments
Like this post? del.icio.us | furl | slashdot | technorati | digg
Comments:

I suppose that even with Sun publishing their first SPECpower_ssj2008 result you still don't consider it a standard benchmark for there are three SPECpower_ssj2008 results published thusfar with > 16 GB of RAM. One from Apple, and two from HP.

FWIW I went to the power calculator for the X4275 and plugged-in the config information from the FDR, and unless the internal PCIe SAS controller consumes a boatload of power or the calculator has been fed some bogus data for component power consumption, it came-up quite short - 613 Watts to your stated average of 750 and peak of 825 or 22% and 34% respectively. Probably aught to add "database" to the HPC disclaimer in the description of the power calculator.

In the past you have been quite vocal about how RAM is a huge component of power consumption. Assuming it can be trusted, taking the delta of 18 DIMMs vs 12 DIMMs in the two processor configuration, the power calculator indicates that the DIMMs here are 4.33W per compared to 104W per for the processors (ISTR a 105 W TDP from Intel? so that seems reasonable). It also suggests the discs were 16W each which seems a little high. So that would be ~78W for the DIMMs to 208W for the Processors to 256 W for the Discs. If we take your stated 750 W average power consumption, it means the RAM was a whopping 10.4% of the power consumption, 9.5% if we went with peak. If we sum the processors RAM and discs we get 642 Watts, leaving 108 (38% more than RAM) watts unaccounted against average and 183 (134% more than RAM) against peak. Was that all fans?

Posted by rick jones on April 16, 2009 at 09:08 AM PDT #

Post a Comment:
Comments are closed for this entry.