Обсуждение: perfromance world records
I just ran into an article about Oracle setting a world record in some kind of test: http://www.oracle.com/corporate/press/2007_feb/TPC-H_300GB_Benchmark_wHP.html?rssid=rss_ocom_pr ...which made me think: postgresql aims at the same (or very similar) clients and use cases as Oracle, DB2 and MSSQL. I pose the question from an advocacy standpoint: why doesn't postgresql hold a world record of some sort (except performance/price)? Is it because the tests (time, expertise, hardware) are too expensive? Are the other RDBMSes simply faster? Something else? I'd like to know, because it'd be a hell of an argument to use when advocating the use of pgsql on a project: "well, we *could* go with MSSQL, but it's going to tie us up...when using multiple CPUs (licences), when deploying a failover solution (licences), when you want to work with spatial information or something else: but pgsql, on the other hand...it doesn't have that kind of licencing volatility, gives you everything it's got and achieves world record performance doing so..." That's the kind of leverage I'd like to have when talking about using pgsql with my colleagues. Anyone care to comment? Cheers, Tomislav
Tomi N/A wrote: > I just ran into an article about Oracle setting a world record in some > kind of test: > http://www.oracle.com/corporate/press/2007_feb/TPC-H_300GB_Benchmark_wHP.html?rssid=rss_ocom_pr > > > ...which made me think: postgresql aims at the same (or very similar) > clients and use cases as Oracle, DB2 and MSSQL. I pose the question > from an advocacy standpoint: why doesn't postgresql hold a world > record of some sort (except performance/price)? Cost. Joshua D. Drake > Is it because the tests (time, expertise, hardware) are too expensive? > Are the other RDBMSes simply faster? Something else? > I'd like to know, because it'd be a hell of an argument to use when > advocating the use of pgsql on a project: "well, we *could* go with > MSSQL, but it's going to tie us up...when using multiple CPUs > (licences), when deploying a failover solution (licences), when you > want to work with spatial information or something else: but pgsql, on > the other hand...it doesn't have that kind of licencing volatility, > gives you everything it's got and achieves world record performance > doing so..." > > That's the kind of leverage I'd like to have when talking about using > pgsql with my colleagues. > Anyone care to comment? > > Cheers, > Tomislav > > ---------------------------(end of broadcast)--------------------------- > TIP 2: Don't 'kill -9' the postmaster > -- === The PostgreSQL Company: Command Prompt, Inc. === Sales/Support: +1.503.667.4564 || 24x7/Emergency: +1.800.492.2240 Providing the most comprehensive PostgreSQL solutions since 1997 http://www.commandprompt.com/ Donate to the PostgreSQL Project: http://www.postgresql.org/about/donate PostgreSQL Replication: http://www.commandprompt.com/products/
"Tomi N/A" <hefest@gmail.com> writes: > ...which made me think: postgresql aims at the same (or very similar) > clients and use cases as Oracle, DB2 and MSSQL. I pose the question > from an advocacy standpoint: why doesn't postgresql hold a world > record of some sort (except performance/price)? Certified TPC tests are *expensive* to run. If you search the PG archives for "TPC" you will probably find some relevant prior discussions. FWIW, Josh has given the impression that Sun is working on producing certified TPC-E results with PG. regards, tom lane
-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1 On 02/24/07 11:00, Tom Lane wrote: > "Tomi N/A" <hefest@gmail.com> writes: >> ...which made me think: postgresql aims at the same (or very similar) >> clients and use cases as Oracle, DB2 and MSSQL. I pose the question >> from an advocacy standpoint: why doesn't postgresql hold a world >> record of some sort (except performance/price)? > > Certified TPC tests are *expensive* to run. If you search the PG > archives for "TPC" you will probably find some relevant prior > discussions. What about non-certified tests? Or has the TPC copyrighted/licensed/whatever the tests, so that you can only publish certified results? > FWIW, Josh has given the impression that Sun is working on producing > certified TPC-E results with PG. -----BEGIN PGP SIGNATURE----- Version: GnuPG v1.4.6 (GNU/Linux) iD8DBQFF4HM5S9HxQb37XmcRArtIAJ0avsmTuu5QxLW3KrGEpdm2zcB5UACgoDzN X+yxEw0miUXDjMkKeMkRt5E= =6OBF -----END PGP SIGNATURE-----
Ron Johnson wrote: > On 02/24/07 11:00, Tom Lane wrote: >>> "Tomi N/A" <hefest@gmail.com> writes: >>>> ...which made me think: postgresql aims at the same (or very similar) >>>> clients and use cases as Oracle, DB2 and MSSQL. I pose the question >>>> from an advocacy standpoint: why doesn't postgresql hold a world >>>> record of some sort (except performance/price)? >>> Certified TPC tests are *expensive* to run. If you search the PG >>> archives for "TPC" you will probably find some relevant prior >>> discussions. > > What about non-certified tests? > > Or has the TPC copyrighted/licensed/whatever the tests, so that you > can only publish certified results? You can not publish TPC tests without a TPC fee :). However there are plenty of other tests such as dbt2 and odbcbench that can give you comparable and free results. Joshua D. Drake > >>> FWIW, Josh has given the impression that Sun is working on producing >>> certified TPC-E results with PG. > > ---------------------------(end of broadcast)--------------------------- TIP 9: In versions below 8.0, the planner will ignore your desire to choose an index scan if your joining column's datatypes do not match -- === The PostgreSQL Company: Command Prompt, Inc. === Sales/Support: +1.503.667.4564 || 24x7/Emergency: +1.800.492.2240 Providing the most comprehensive PostgreSQL solutions since 1997 http://www.commandprompt.com/ Donate to the PostgreSQL Project: http://www.postgresql.org/about/donate PostgreSQL Replication: http://www.commandprompt.com/products/
2007/2/24, Joshua D. Drake <jd@commandprompt.com>: > Ron Johnson wrote: > > On 02/24/07 11:00, Tom Lane wrote: > >>> "Tomi N/A" <hefest@gmail.com> writes: > >>>> ...which made me think: postgresql aims at the same (or very similar) > >>>> clients and use cases as Oracle, DB2 and MSSQL. I pose the question > >>>> from an advocacy standpoint: why doesn't postgresql hold a world > >>>> record of some sort (except performance/price)? > >>> Certified TPC tests are *expensive* to run. If you search the PG > >>> archives for "TPC" you will probably find some relevant prior > >>> discussions. > > > > What about non-certified tests? > > > > Or has the TPC copyrighted/licensed/whatever the tests, so that you > > can only publish certified results? > > You can not publish TPC tests without a TPC fee :). However there are > plenty of other tests such as dbt2 and odbcbench that can give you > comparable and free results. I mentioned a TPC test as an example: any kind of (well known) "standard" test would do. I guess it goes without saying anyone running such a test would do well to send word to the mailing list with a URL to the results. :) t.n.a.
"Tomi N/A" <hefest@gmail.com> writes: > 2007/2/24, Joshua D. Drake <jd@commandprompt.com>: >> You can not publish TPC tests without a TPC fee :). However there are >> plenty of other tests such as dbt2 and odbcbench that can give you >> comparable and free results. > I mentioned a TPC test as an example: any kind of (well known) > "standard" test would do. If the objective is to claim a world record, we'd look pretty silly trying to do so with a nonstandard, non-certified test. The point of certification in this context is that you have someone else attesting to the validity of your results. Without that, your claim isn't going to be believed. regards, tom lane
2007/2/25, Tom Lane <tgl@sss.pgh.pa.us>: > If the objective is to claim a world record, we'd look pretty silly > trying to do so with a nonstandard, non-certified test. The point > of certification in this context is that you have someone else > attesting to the validity of your results. Without that, your claim > isn't going to be believed. Makes sense. I got carried away a bit. I guess I'll have to stick to the available case studies...maybe even contribute one or two of those myself. Cheers, t.n.a.
On 2/24/07, Tomi N/A <hefest@gmail.com> wrote: > > That's the kind of leverage I'd like to have when talking about using > pgsql with my colleagues. > Anyone care to comment? The name brand test are basically paid pr for the big databases. Basically, the tests are in environments controlled completely by the vendor. Like almost everything else in the commercial database world, the major point is to distract and confuse people and not provide any substantive information. Some people who follow graphics card developments might remember how both ATI and nVidia caught a lot of heat by optimizing their drivers for specific benchmarks to make themselves look good. In the database world, this kind of behavior is encouraged and publication of user run benchmarks is prohibited. A much more fair benchmark would be to publish the hardware/OS platform in advance but not the SQL in a benchmark. Only standard SQL would be used and if any unexpected results come back the test is considered 'failed' by the database. I don't think such a thing would ever happen though, but I would expect PostgreSQL to do extremely well in such a test. merlin