Wednesday Oct 31, 2007
Yesterday I posted several Sun Fire X4100 world record performance & $/performance results
on the TPC-H benchmarks. Sun leads the pack with Paraccel's innovative software and
even the software pricing model. I wanted to try to give even more insights.
ParAccel has a very efficient data compression techniques so using the
storage associated with Sun's hardware design, the Sun TPC-H cluster submissions did not need ANY external storage.
The data could be entirely contained on the internal drives of each
node. Once again, Sun is the ONLY vendor that has been able to do so.
Other vendors, using traditional RDBMS, are forced to submit results
with a disproportionate number of spindles, in order to achieve
competitive performance, even at small scale factors.
As a consequence, the Sun ParAccel cluster solution greatly reduces
costs and requirements for space, power/cooling, maintenance and
complex hardware administration.
Sun and ParAccel are in the remarkably unique position of
having simultaneously the best performance and price/performance
by far at multiple TPC-H scale factors. As I said yesterday, innovators
seek out and find innovators.
The ParAccel Analytic Database is an innovative RDBMS designed
from the ground up as Decision Support and Business Intelligence
database.
ParAccel has a parallel cluster architecture, "shared-nothing",
with node redundancy, which relies on a column-based table
structure and heavy data compression. Compression provides for
a significantly smaller data footprint, higher speed in data
processing and transmission. The significant reduction in the
size of the data allows for even large databases to be fully
contained in the memory of a cluster of multiple systems.
The database can also operate by accessing data on permanent
storage, although with a tradeoff in performance.
ParAccel doesn't make use of complex index structures, further
reducing database size, administration complexity and simplifying
data updates.
Data partitioning and load balancing is provided automatically
by the ParAccel product and doesn't require manual DB admin
intervention, either to optimize the data layout or to create
additional indexes and structures, traditionally needed
to speedup query performance.
Results detailed at:
http://blogs.sun.com/bmseer/entry/innovation_setting_world_records_in
http://blogs.sun.com/bmseer/entry/tpc_h_100gb_sun_fire
http://blogs.sun.com/bmseer/entry/sun_paraccel_in_tpc_h
Disclosure Statement:
TPC-H @1000GB Sun Fire X4100 cluster 315,842.9 QphH@1000GB, $4.57/QphH@1000GB, avail 10/29/07; TPC-H @300GB Sun Fire X4100 cluster 198,578.1 QphH@300GB, $3.15/QphH@300GB, avail 10/29/07; TPC-H @100GB Sun Fire X4100 cluster 98,857.0 QphH@100GB, $2.65/QphH@100GB, avail 10/29/07; TPC-H, QphH, $/QphH tm of Transaction Processing Performance Council (TPC). More info http://www.tpc.org.
Tuesday Oct 30, 2007
This benchmark result demonstrates that a shared-nothing cluster of Sun Fire X4100s achieves the best performance and, simulatenously, the best price/performance among all TPC-H@300GB submissions. Each of the 30 node in the cluster was powered by 2 dual-core 2.8GHz Opteron processors.
Specifically, the Sun Fire X4100 cluster achieved a QphH@300GB of 198,578 together with $/QphH@300GB of $3.15. The performance is almost 5 times the previously reported best performance and the price/performance is more than twice as good as the previously reported best price/performance.
For other Other Innovations see:
http://blogs.sun.com/bmseer/entry/innovation_setting_world_records_in
TPC-H @300GB Performance:
$/QphH = TPC-H Price/Performance metric (smaller is better)
QphH = TPC-H Composite Metric (bigger is better)
| System |
Socket
/Core/
Thread |
CPU Type GHz |
QphH |
$/
QphH
|
total
price |
DBMS
|
Avail |
# Clust Nodes |
Sun Fire X4100
|
60/ 120/ 120
|
Opt 2.8GHz |
198,578 |
3.15
|
624,770
|
Paraccel
|
10/29/07 |
30 |
HP BL480c
|
16/ 32/ 32
|
Opt 3.0GHz |
40411
|
18.67
|
754,232
|
Oracle
|
12/18/06 |
8 |
HP BL460c
|
16/ 32/ 32 |
Xeon 3.0GHz |
39614
|
12.57
|
497,869
|
Oracle
|
9/15/07
|
8 |
Dell PE6800
|
8/ 16/ 32 |
Xeon 3.0GHz |
18881
|
24.37
|
460,004
|
Oracle
|
4/24/06
|
2 |
IBM eSvr 325
|
16/ 16/ 16 |
Opt 2.0GHz |
13195
|
65.44
|
863,410
|
DB2
|
11/08/03
|
8 |
TPC-H @300GB Price-Performance:
$/QphH = TPC-H Price/Performance metric (smaller is better)
QphH = TPC-H Composite Metric (bigger is better)
| System |
Socket/
Core/
Thread |
CPU Type GHz |
QphH |
Price/
QphH
|
Price
in
US dollars
|
DBMS
|
Available |
#
Clust
Nodes
|
Sun Fire X4100
|
60/120/120
|
Opteron 2.8 GHz |
198578
|
3.15
|
624,770
|
Paracell
|
10/29/07 |
30 |
Sun Fire X4100M2
|
2/ 4/ 4 |
Opt 3.0 GHz |
7641
|
5.89
|
45,001
|
SybIQ
|
5/25/07
|
1 |
Sun Fire X4200
|
2/ 2/ 2
|
Opt 3.0GHz |
4936
|
6.29
|
31,033
|
SybIQ
|
6/23/06 |
1 |
HP DL585 G1
|
4/ 4 /4 |
Opt 2.4GHz |
12226
|
11.71
|
143,041
|
SQLS
|
1/26/06 |
1 |
HP BL460c
|
16/ 32/ 32 |
xeon 3.0GHz |
39614
|
12.57
|
497,869
|
Oracle
|
9/15/07 |
8 |
HP DL585 G2
|
4/ 8/ 8 |
Opt 2.8GHz |
18299
|
13.67
|
250,057
|
SQLS
|
4/19/07
|
1 |
IBM x3650
|
2/ 4/ 4 |
Xeon 3.0GHz |
10165
|
15.40
|
156,535
|
DB2
|
10/06/06
|
1 |
Dell 6800
|
8/ 8/ 8 |
Xeon 3.33GHz |
11743
|
21.84
|
256,383
|
Oracle
|
1/08/06 |
2 |
Benchmark Description
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 (100GB, 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.
Disclosure Statement:
TPC-H @300GB
Sun Fire X4100 cluster 198,578.1 QphH@300GB, $3.15/QphH@300GB, avail 10/29/07;
TPC-H, QphH, $/QphH tm of
Transaction Processing Performance Council (TPC).
More info http://www.tpc.org.
See Also:
Results Summary TPC-H SF300 - 300GB benchmark
| Audited Results |
|
Database Size: |
300 GB (Scale Factor 300) |
|
TPC-H Composite: |
198,578 QphH@300GB |
|
Price/performance: |
$3.15 / QphH@300GB |
|
Available |
Oct 29, 2007 |
| Number of Systems: |
30 x Sun Fire X4100
each X4100:
16GB memory each,
2 x 146GB (10k RPM) internal SAS each
|
| Total Number Processors: |
60
|
| Total Number Cores: |
120
|
| Total Number Threads: |
120
|
| Processor/GHz of Server: |
Opteron/2.8 GHz Dual-core
|
| Storage: |
8158 Gigabytes of disk |
| Database: |
Paraccel Analystic Database |
| Operating System: |
RedHat Enterprise Linux 4.4 |
| Total 3 year Cost: |
$624,770.12 |
| Other Performance Metrics |
|
TPC-H Power: |
145,219.3
|
|
TPC-H Throughput: |
271,542.9
|
|
Database Load Time |
15 minutes 5 seconds
|
|
Storage Ratio: |
27.19
|
|
Monday Jun 04, 2007
The Sun Fire X4100 M2 has 50% better price performance than the HP DL585.
This benchmark result demonstrates that the Sun Fire X4100 M2, powered by 2 dual-core 3.0GHz Opteron, improves upon Sun's previously published world-record
$/performance result at 300GB. The Sun Fire X4100 M2 is the only 1U system ever submitted for a TPC benchmark at the 300GB scale-factor.
The Sun Fire X4100 M2 achieved the best price-performance among all systems
at 300GB. It improved upon Sun's previous world-record price-performance, achieved by the Sun Fire X4200, by 6%.
Note all of this detail, and the very different ways in which results are marketed with the IBM POWER6 post.
The Sun Fire X4100 M2 achieved a 55% QphH@300GB improvement upon previously
published 2-socket Single-core RevE Sun Fire X4200 result, i.e., 7641 QphH@300GB versus 4936 QphH@300GB.
Specifically, Sun, using its Sun Fire X4100 M2 server achieved a
$/QphH@300GB of $5.89, whereas the Sun Fire X4200 achieved a $/QphH@300GB
of $6.29. The latter result was submitted on June 23, 2006.
TPC-H @300GB Performance Results (sorted by $/QphH
for single (non-clustered) systems:
$/QphH = $/QphH@300GB TPC-H Price/Performance metric (smaller is better)
QphH = QphH@300GB TPC-H Composite Metric (bigger is better)
Disk Data Ratio is the ratio of the total number of gigabytes of
configured storage to the scale factor
number of gigabytes (smaller is better)
| System |
Sockets/
Cores/
Threads CPU GHz |
QphH |
Price/
QphH
|
Price
in
currency
|
DBMS
|
Available |
Disk
Data
Ratio
|
Cluster |
Sun X4100 M2
|
2/4/4 Opteron 3.0
|
7641
|
$5.89
|
45,001 $US
|
SybIQ
|
6/23/06 |
3.17 |
N
|
Sun X4100
|
2/2/2 3.0 Opteron |
4936
|
$6.29
|
31,033 $US
|
SybIQ
|
6/23/06 |
2.9 |
N
|
HP DL585 G1
|
4/8/8 Opteron 2.4
|
12225 |
$11.71
|
143,041 $US
|
SQLS
|
01/26/06
|
19.9 |
N
|
HP DL585 G2
|
4/8/8 Opteron 2.8
|
18298 |
$13.67
|
250,057 $US
|
SQLS
|
04/19/07
|
24.96 |
N
|
| IBM x3650 |
2/4/4 WoodC 3.0
|
10165
|
$15.40
|
156,535 $US
|
DB2
|
10/06/06
|
12.8
|
N
|
Sun V440
|
4/4/4 US IIIi 1.6
|
2501
|
$22.09
|
55,245 $US
|
SybIQ
|
05/09/05
|
1.81
|
N
|
| HP DL585 G1 |
4/8/8 Opteron 2.4
|
11915
|
$22.78
|
271,379 $US
|
DB2
|
10/05/05 |
19.7
|
N
|
| HP DL585 G1 |
4/4/4 Opteron 2.6
|
8434
|
$30.18
|
255,586 $US
|
DB2
|
05/17/05 |
13.8
|
N
|
IBM eServer 366
|
4/4/8 Xeon 3.6
|
7762
|
$32.94
|
255,702 $US
|
DB2
|
05/02/05
|
18.5
|
N
|
The results reported here were performed on a Sun Fire X4100 M2 system
running the Sybase IQ 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).
Sun achieved this result using only 14 disks. Other vendors used anywhere
from 104 disks (the IBM x3650 result) to 208 disks (the HP DL585 G2 result).
The
significance of being able to house a data warehouse with
fewer disks provides numerous advantages far beyond the
scope of the TPC-H metrics. These include, ease of management, lower
probability of admin errors, a
much lower probability of disk failures and a true
reduction in the total cost of ownership over the life of a
system.
All Sun/SybaseIQ submissions, including this one, RAID protect their storage.
Only a few, of the almost 30 existing non-Sun submissions, at 300GB RAID
protect their storage. The lack of RAID protection results in artificially
cheaper configurations, which no production shop would ever deploy.
Benchmark Description
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 (100GB, 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.
Disclosure Statement:
TPC-H @300GB Sun Fire X4100 M2 7641 QphH@300GB, $5.89/QphH@300GB, avail 5/25/07;
TPC-H, QphH, $/QphH tm of
Transaction Processing Performance Council (TPC).
More info www.tpc.org.
| Audited Results |
|
Database Size: |
300 GB (Scale Factor 300) |
|
TPC-H Composite: |
7641 QphH@300GB |
|
Price/performance: |
$5.89 / QphH@300GB |
|
Available |
May 25, 2007 |
| Number of Systems: |
Sun Fire X4100 M2
|
| Total Processors, cores, Threads: |
2,2,2
|
| Processor |
Dual-core Opteron 3.0GHz
|
| Storage: |
951 Gigabytes of disk |
| Database: |
Sybase IQ 12.6 |
| Operating System: |
Solaris 10 |
| Total 3 year Cost: |
$45,001.35 |
| Other Performance Metrics |
|
TPC-H Power: |
7847
|
|
TPC-H Throughput: |
7440.5
|
|
Database Load Time |
4 hours 22 minutes 53 seconds
|
|
Storage Ratio/type: |
3.17 ratio/ two STK3320 SCSI JBOD array
|
|
See Also
Sun Fire X4100 M2 TPC-H Executive Summary Report Acrobat PDF (68K)
Complete Sun Fire X4100 M2 TPC-H Full Disclosure Report Acrobat PDF (590K)
Transaction
Processing Performance Council (TPC) Home Page
Ideas International Benchmark page
I'll even show my math, I challenge other vendors to show it too!
6% claim from: (6.29-5.89)/6.29 = 0.0635
50% claim from: (11.71-5.89)/11.71 = 0.4970 (49.7 rounds to 50)
55% claim from: (7641-4936)/4936 = 0.5480 (54.8 rounds to 55)
Tuesday Mar 20, 2007
World Record SPECompL2001: Solaris beats RedHat Linux and
The Sun Fire X4600 M2 delivers the best performance on the SPEC OMPL2001
benchmark suite of all x86 systems.
-
Solaris 10 and Studio 11 duo help X4600 M2 perform 8% better than Red Hat
Linux (RHEL4) and PathScale compiler on SPECompL2001.
-
The Sun Fire X4600 M2 server in 4-socket configuration using dual-core AMD
Opteron Model 8220 processors, produced best SPECompL2001 result of 111,893.
-
The Sun Fire X4600 M2 beats the HP DL585 G2 (AMD Opteron 8220 4chips/8cores) using RedHat Linux
by 9%
-
The results show that the combination of Solaris 10 using Sun
Studio 11 is unmatched by the competition for assisting users in writing parallel code.
SPECompL2001 (bigger is better, ordered by peak)
| Result |
Cores |
Chips |
Thrds |
System |
| Peak |
Base |
| 111,893 |
105,465 |
8 |
8 |
2 |
Sun X4600M2 Opteron 8220 S10/SS11 |
| 103,466 |
100,610 |
8 |
8 |
2 |
Sun X4600M2 Opteron 8220 RHEL4u4, PathScale v2.5 |
| 102,283 |
99,907 |
8 |
8 |
2 |
HP DL585 G2 Opteron 8220SE, RHEL4u4, PathScale v2.5 |
| 92,725 |
92,418 |
16 |
16 |
1 |
HP Superdome 1.5GHz Itanium 2 |
| 79,627 |
68,051 |
24 |
24 |
1 |
Sun Fire 6800 |
| 44,376 |
42,400 |
16 |
16 |
1 |
HP Superdome 875MHz PA-8700+ |
| 42,864 |
41,056 |
16 |
16 |
1 |
HP server rp8400 (875MHz PA-8700+) |
Benchmark Description
The SPEC OMPL2001 Benchmark Suite was released in June 2001 and
tests HPC performance using OpenMP for parallelism.
9 programs (2 in C and 7 in Fortran) parallelized using OpenMP API.
Goals of suite are: first, target to Large-range (8-128 processor)
parallel systems, 2nd have run rules, tools and reporting similar
to SPEC CPU2000 and 3rd to have programs representative of HPC and
Scientific Applications.
Results Summary
| Result |
|
X8420 8-threads: |
|
111893 SPECompL2001 |
| Reference Date: |
|
Mar 19, 2007 |
| System: |
|
Sun Fire X4600 M2 |
| Total Number Processors: |
|
4 |
| Total Memory : |
|
32 GB (16x4GB DIMMs), DDR667 |
| Processor/GHz of Server: |
|
Opteron 8220, 2.8 GHz |
| Operating System: |
|
Solaris 10 |
| Compiler: |
|
Sun Studio 11 |
Disclosure Statement:
SPEC, SPEComp reg tm of Standard Performance Evaluation Corporation.
Results from www.spec.org as of Mar 19, 2007, Sun result submitted to SPEC.
Sun Fire X4600 M2 (S10/SS11, Opteron 8220, 8 cores, 4 chips, 8 threads), 111893 SPECompL2001.
Sun Fire X4600 M2 (RHEL4u4, Opteron 8220, 8 cores, 4 chips, 8 threads), 103466 SPECompL2001.
HP DL585 G2 Opteron 8220SE (RHEL4u4, Opteron 8220, 8 cores, 4 chips, 8 threads), 103466 SPECompL2001. Sockets refers to chips.
See Also
SPEC OMP2001 Page
SPEC Home Page
sun.com X4600 M2 Benchmark Page
Wednesday Nov 15, 2006
As mentioned in the posting earlier today, scalability is important factor in system performance. Woodcrest's poor scaling may not bode well for
Cloverton. Sure you can package for threads onto a module, but unless you design for them you'll just have more threads not delivering performance but just burning more watts.
Wattage: I'll get detailed wattage results posted soon, but it looks like
as we mentioned Opteron performance is about 20% more than Woodcrest. The
wattage for both configurations looks the same. Therefore expect Sun's
Opteron to have about 20% perf/watt advantage.
Sun's Fluent results will be posted shortly on the website, it is a busy
week with Supercomputing conference and lots of busy people. So keep
checking back. A few of the smaller gave Woodcrest a small percent advantage, but most were significantly faster on Sun's Opteron.
...maybe Woodcrest will have better idle power, but why in the world would
you buy the latest server and leave it idle?
Wednesday Oct 25, 2006
The Sun Fire X4100/X4200 M2 using Solaris 10 and Sun Studio 11 delivers the best performance on the SPEC OMPM2001
benchmark suite of all 2-socket systems running 4-threads.
The Sun Fire X4100/X4200 M2 beat the IBM p5 520 POWER5+ 1.9GHz AIX5L V5.3 result by 61%
The results show that the combination of Solaris 10 using Sun
Studio 11 is unmatched by the competition for
assisting users in writing parallel code.
SPECompM2001 Performance Comparison (bigger is better):
| Result |
Cores |
Chips |
Thrds |
System |
| Peak |
Base |
| 13222 |
12763 |
4 |
2 |
4 |
Sun Fire X4100/X4200 M2, Opteron 2220SE, 2.8GHz |
| 12574 |
12127 |
4 |
2 |
4 |
Sun Fire X2200 M2, Opteron 2218, 2.6GHz |
| 10964 |
10424 |
4 |
2 |
4 |
Sun Fire X4100, Opteron 285, 2.6GHz |
| 8174 |
8141 |
2 |
1 |
4 |
IBM System p5 520 (1.9 GHz, 2 CPU) |
See Also
Benchmark Description
The SPEC OMPM2001 Benchmark Suite was released in June 2001 and
tests HPC performance on a variety of scientific applications using OpenMP for parallelism.
It consists of
11 programs (8 fp and 3 int intensive) in C and Fortran parallelized using OpenMP API
System Configuration:
- Sun Fire X4100/X4200
- 2 x 2.8 GHz Opteron 2220SE processors
- 16GB memory (4x2GB per chip), DDR667
- Solaris 10
- Sun Studio 11
Disclosure Statement:
SPEC, SPEComp reg tm of Standard Performance Evaluation Corporation.
Results from www.spec.org as of Oct 16, 2006, Sun result submitted to SPEC.
Sun Fire X4100/X4200 M2 (4 cores, 2 chips, 4 threads), 13,222 SPECompM2001.
IBM System p5 520 (2 cores, 1 chips, 4 threads), 8,174 SPECompM2001.
Sockets refers to chips.
|
|
|
|
