running the SPECjbb2000 benchmark on HP -UX 11
20
pages
English
Documents
Le téléchargement nécessite un accès à la bibliothèque YouScribe Tout savoir sur nos offres
20
pages
English
Documents
Le téléchargement nécessite un accès à la bibliothèque YouScribe Tout savoir sur nos offres
running the SPECjbb2000
benchmark on HP-UX 11
Ingolf Buttig
AppDev ATC
October, 2001
© 2001 HEWLETT-PACKARD COMPANY
running the SPECjbb2000 benchmark
on HP-UX 11
introduction
description of the benchmark
SPECjbb2000 was announced by the Standard Performance Evaluation Corp. (SPEC) in June
2000 as a new benchmark for evaluating the performance of servers running typical Java business
applications and can be found at http://www.spec.org/osg/jbb2000/press/release.html.
In the official SPECjbb2000 documentation the benchmark is described as follows:
“SPECjbb2000 is a Java program emulating a 3-tier system with emphasis on the middle tier.
Random input selection represents the 1st tier user interface. SPECjbb2000 fully implements the
middle tier business logic. The 3rd tier database is replaced by binary trees.
SPECjbb2000 is inspired by the TPC-C benchmark and loosely follows the TPC-C specification
for its schema, input generation, and transaction profile. SPECjbb2000 replaces database tables with
Java classes and replaces data records with Java objects. The objects are held by either binary trees
(also Java objects) or other data objects.
SPECjbb2000 runs in a single JVM in which threads represent terminals in a warehouse. Each
thread independently generates random input (tier 1 emulation) before calling transaction-specific
business logic. The business logic operates on the data held in the binary trees (tier 3 ...
r u n n i n g t h e S P E C j b b 2 0 0 0 benchmark on -HUPX 11
© 2001EHWLETT-PACKARDCOMPANY
Ingolf Buttig AppDev ATC October, 2001
r u n n i n g t h e S P E C j b b 2 0 0 0 b e n c h m a r k o n H P- U X 1 1
description of the benchmark SPECjbb2000 was announced by the Standard Performa nCcoer pE. v(aSlPuEatCi)o nin June 2000 as a new benchmark for evaluating the performance of servers running typical Java b applications and can be fhp:ttww//ps.wo.ceo/grj/gsrler00p/bb02esas/..hte ml In the official SPECjbb2000 documentation the benchmark is described as follows:
“SPECjbb2000 is a Java program emu-ltaiteirn gs yas t3em with emphasis on the middle tier. Random input selection represents the 1st tier user inbte2r0f0a0c ef. ulSlyP EiCmjbplements the middle tier business logic. The 3rd tier database is replace d by binary trees. SPECjbb2000 is inspired by t-hCe bTePnCchmark and loosely follows -tCh se pTePciCfication for its schema, input generation, and transa cStPioEnC jpbrbo2fi0le0.0 replaces database tables with Java classes and replaces data records with Java objects. The objects are held by either bin (also Java objects) or other data objects.
SPECjbb2000 runs in a single JVM in which threads reprlse sien nat er.euismac tEh anawoher thread independently generates random input (tier 1 emulation) befo-rsep eccailfliion cn gastcrtna business logic. The business logic operates on the data held in the binary trees (tier 3 emulati benchmark does no diskr I/nOe towork I/O .
© 2 100EHWLETT-PACKARDCOMPANY
2
System architecture:
For additional details, please refer to the official SPECjbb2000 website http://www.spec.org/osg/jbb 20i0n0cl/uding a detailed description of this benchmark: http://www.spec.org/osg/jbb2000/docs/whitep aper. . html
purpose of the paper The aim of this paper is to enable you to perform the SPECjbb2000 benchmark on one of H server systems using VHMP.' s TJhe steps covered include successfully configuring the operating system, installing the software and running the benchmark.
Due to a customer request, the ATC team successfully performed all these steps in late Fe 2001 in the HP Performancer PH htiw i11 XU-maer(Gn nd a),nyniB etniegeolbCenand the latest JVM 1.3, were able to get higher numbers than the officially published numbers from May 2000
These values were reached only by using “standard performance tuning steps for J applicatio ns.
© 2 00 1EHWLETT-PACKARDCOMPANY
3
p r e p a r a t i o n
where to get a copy of SPECjbb2000 A copy can be ordered directly from:
Standard Performance Evaluation Corporation (SPEC) 6585 Merchant Place, Suite 100 Warrenton, VA 20187, USA Phone: 1 540 349 7878 FAX: 1 540 349 5992 E-Mail.oo@infecsp rg
or by using the online product order inquiry form atpe.sww/w:/tphtro/nib-igc/gro.cder. SPECjbb2000 is available for $400. Academic institutions may be eligible for a discounted price of $200.. latest version of the Java SDK The latest version of the SDK can be obtained from the http://www.hp.com/products1/unix/java/java2/sdkrte13 /iwnedbesxi.thet.mFlollow the links to _ the download page for the latest Enterprise Java 1.-3U XS D1K1 . fForo rH tPhe tests for this paper, JDK 1.3.00 was u sed.
patches The performance of the JVM -oUn X HcPan be improved by a number of patches that are currently listh.wphc.mott:p//wwp/a.tcudorp//xinu/1snf/ivajaryraibol hcsetlh.mt See Appendix A for a list of softwareh aens din pstaatlcled on the testing .emachin
kernel tuning TheHPjconfigtool was used to determine the kernel parameter settings for a typical Java Enterprise application available at http://www.hp.com/products1/unix/java/java2/hpjconfig/. .htmndexil Please note, that HPjconfig only gives recommendations for kernel parameters.
The real kernel parameter setting has tosbaemo d orknmet uvinae .
Please seevams.OverK/sorapCdeni/scouxhpnl/op..hm/cohttdocsp:// ifeowrAal l.thlm description of the tunable kernel parameters and http://www.hp.com/products1/unix/java/infolibrary/prog_guide/java1_3/con fiogrmtul.hontira general kernel parameter recommendations for Java- UsXu.p port on HP See Appendix B for a list of kernel parameter settingsc hoine t.he test ma
installatio n
© 10 02EHWLETT-PACKARDCOMPANY
4
Installation and testing of SPECjbb2000 is straightforward according to the “Installati Instructions in the “User's Guugdi.ehbb2000/docs/userps.wo.ceo/grj/gshp:ttww//.t ml
To install the benchmjaavrka: setu fpor a GUI based installatijoanv ao rs out pe for an installation without GUI.
To run the benchmraurnk.:sh
r u n n i n g t h e b e n c h m a r k
configuration files SPECjbb2000 takes two properties files as input:
·acontrol properties file ·a descriptive properties file. The control properties file is used to modify the operation of the benchmark, for exa altering the length of the measurement interval.
The descriptive properties file is used to docutemme nut ntdherstyest; this documentation is required for publishable results, and is reflected in the output files. The values of the des properties do not affect the workload.
Please see the “User'stthw.ww//p:/ssureg0200d/co/osg/jbbspec.org ufiodre.ah tml detailed description of these property files.
In Appendix C are the property files used in the tests used for this paper.
running the benchmark Edit the run.sh script byg atdhdei nnecessary JVM startup parameters. Start the benchmark by invokinrgu n.sh.
In Appendix D is the run.sh script used in this paper.
g e t t i n g m a x i m u m n u m b e r s
analysis of garbage collection One of the most powerful performance tuning stepsoisf the analysisioct Tn.s hiGabraC egello is done by using tXhverbosegc option.
Please see: http://www.hp.com/products1/unix/java/java2/sdkrte1_ry3/itatnemuh.no1003cod/a1kdsfoinbrli tm lfor a detailed description of all the Xverbosegc output fields. The initial and maximal heap used was 1850 MB:
© 21 00EHWLETT-PACKARDCOMPANY
5
-Xms1850 m
-Xmx1850 m
Normally, the ratio between the “New Generation area and the overabll e h1e:a4p. size should But during testing it was discovered that there were- comllaenctyi ognasr bfoarg ethe “New Generation area only. (Re1a isno nt he Xverbo-soeugtcput). So the “New generation area was increased up to 800 MB:
-Xmn800 m
These settings already t btrhoeu gehxpected performance nu mbers of
SPECjbb2000 = 41947 ops/s
Note:Interesting to see that a balance is needed between the size of the “New Generation and the overall heap size. Inc-rXemasni ntog o much resulted in problems wit-ch olglaercbtinao.g e (Reason 5 in the Xverb-osegcpuut. t)
As an example, in Appendix E lists the Xverbosegc output of a successful run and of a run garba-gceollection proble ms.
h p r e s u l t s
1)In Q2 2000, HP officially posted 2 results: ·p 2 ,0065JJaP Hs,orsscero2.2.to1 topsavH .05 SPECjbb2000 11.834 ops/sec =
See al02000--503s200s/rejbb20q2/thpt/:bj/g002ber/0tlusww/wpe.sorc.osg/010l.tm.h020
·5 1.2.2.0H topstoH PaJavorp 8 ,0,srossec00N4 SPECjbb2000 = 40.192 ops/sec
See al2000/jbb/osg.orgpsceww.w:p//htt0000-220bb/jq20002ser/stluser/-5301000. mlht1. Note that these resuletrse wpublished for Q2, 2000. For the latest HP results, see http://www.spec.org/osg/jbb2000/results/.
2)eCtnni roamcn esuc emotf dea roP HrfPeatr he tu onewertl serusasury meiallfficg inowllfoe Th Boeblingen in February 2001: Date: 28/02/2001 Hardware: N4000 (8 procs, 550 MHz, 32 GB memory)
© 12 00EHWLETT-PACKARDCOMPANY
6
