Thread: Suggestions for benchmarking 7.4RC2 against 7.3
Hi,
I plan to put 7.4-RC2 in our production servers in next few hours.
Since the hardware config & the performance related GUCs parameter
are going to remain the same i am interested in seeing the performance
improvements in 7.4 as compared 7.3 .
For this i plan to use the OSDB 0.14 and compare the results for both the
cases.
Does any one has suggestions for comparing 7.4 against 7.3 ?
Since i am using OSDB for second time only any tips/guidance
on usage of that is also appreciated.
H/W config:
CPU: 4 X Intel(R) Xeon(TM) CPU 2.00GHz
MEM : 2 GB
I/O config : PGDATA on 10000 RPM Ultra160 scsi , pg_xlog on a similar
seperate SCSI
GUC:
shared_buffers = 10000
max_fsm_relations = 5000
max_fsm_pages = 55099264
sort_mem = 16384
vacuum_mem = 8192
All other performance related parameter have default
value eg:
#effective_cache_size = 1000 # typically 8KB each
#random_page_cost = 4 # units are one sequential page fetch cost
#cpu_tuple_cost = 0.01 # (same)
#cpu_index_tuple_cost = 0.001 # (same)
#cpu_operator_cost = 0.0025 # (same)
BTW i get following error at the moment:
-----------------------------------------
/usr/local/bin/osdb-pg-ui --postgresql=no_hash_index
"osdb"
"Invoked: /usr/local/bin/osdb-pg-ui --postgresql=no_hash_index"
create_tables() 0.78 seconds return value = 0
load() 1.02 seconds return value = 0
create_idx_uniques_key_bt() 0.64 seconds return value = 0
create_idx_updates_key_bt() 0.61 seconds return value = 0
create_idx_hundred_key_bt() 0.61 seconds return value = 0
create_idx_tenpct_key_bt() 0.62 seconds return value = 0
create_idx_tenpct_key_code_bt() 0.45 seconds return value = 0
create_idx_tiny_key_bt() 0.46 seconds return value = 0
create_idx_tenpct_int_bt() 0.46 seconds return value = 0
create_idx_tenpct_signed_bt() 0.45 seconds return value = 0
create_idx_uniques_code_h() 0.46 seconds return value = 0
create_idx_tenpct_double_bt() 0.46 seconds return value = 0
create_idx_updates_decim_bt() 0.45 seconds return value = 0
create_idx_tenpct_float_bt() 0.46 seconds return value = 0
create_idx_updates_int_bt() 0.46 seconds return value = 0
create_idx_tenpct_decim_bt() 0.46 seconds return value = 0
create_idx_hundred_code_h() 0.45 seconds return value = 0
create_idx_tenpct_name_h() 0.46 seconds return value = 0
create_idx_updates_code_h() 0.46 seconds return value = 0
create_idx_tenpct_code_h() 0.45 seconds return value = 0
create_idx_updates_double_bt() 0.46 seconds return value = 0
create_idx_hundred_foreign() 0.41 seconds return value = 0
populateDataBase() 11.54 seconds return value = 0
Error in test Counting tuples at (6746)osdb.c:294:
... empty database -- empty results
perror() reports: Resource temporarily unavailable
someone sighup'd the parent
Any clue?
------------------------------------------
Regards
Mallah.
the error mentioned in first email has been overcome by running osdb on the same machine hosting the DB server. regds mallah. Rajesh Kumar Mallah wrote: > > Hi, > > I plan to put 7.4-RC2 in our production servers in next few hours. > > Since the hardware config & the performance related GUCs parameter > are going to remain the same i am interested in seeing the performance > improvements in 7.4 as compared 7.3 . > > For this i plan to use the OSDB 0.14 and compare the results for both > the > cases. > > Does any one has suggestions for comparing 7.4 against 7.3 ? > Since i am using OSDB for second time only any tips/guidance > on usage of that is also appreciated. > > > > H/W config: > > CPU: 4 X Intel(R) Xeon(TM) CPU 2.00GHz > MEM : 2 GB > I/O config : PGDATA on 10000 RPM Ultra160 scsi , pg_xlog on a similar > seperate SCSI > > GUC: > shared_buffers = 10000 > max_fsm_relations = 5000 > max_fsm_pages = 55099264 > sort_mem = 16384 > vacuum_mem = 8192 > > All other performance related parameter have default > value eg: > > #effective_cache_size = 1000 # typically 8KB each > #random_page_cost = 4 # units are one sequential page fetch > cost > #cpu_tuple_cost = 0.01 # (same) > #cpu_index_tuple_cost = 0.001 # (same) > #cpu_operator_cost = 0.0025 # (same) > > > > BTW i get following error at the moment: > ----------------------------------------- > /usr/local/bin/osdb-pg-ui --postgresql=no_hash_index > "osdb" > "Invoked: /usr/local/bin/osdb-pg-ui --postgresql=no_hash_index" > > create_tables() 0.78 seconds return value = 0 > load() 1.02 seconds return value = 0 > create_idx_uniques_key_bt() 0.64 seconds return value = 0 > create_idx_updates_key_bt() 0.61 seconds return value = 0 > create_idx_hundred_key_bt() 0.61 seconds return value = 0 > create_idx_tenpct_key_bt() 0.62 seconds return value = 0 > create_idx_tenpct_key_code_bt() 0.45 seconds return value = 0 > create_idx_tiny_key_bt() 0.46 seconds return value = 0 > create_idx_tenpct_int_bt() 0.46 seconds return value = 0 > create_idx_tenpct_signed_bt() 0.45 seconds return value = 0 > create_idx_uniques_code_h() 0.46 seconds return value = 0 > create_idx_tenpct_double_bt() 0.46 seconds return value = 0 > create_idx_updates_decim_bt() 0.45 seconds return value = 0 > create_idx_tenpct_float_bt() 0.46 seconds return value = 0 > create_idx_updates_int_bt() 0.46 seconds return value = 0 > create_idx_tenpct_decim_bt() 0.46 seconds return value = 0 > create_idx_hundred_code_h() 0.45 seconds return value = 0 > create_idx_tenpct_name_h() 0.46 seconds return value = 0 > create_idx_updates_code_h() 0.46 seconds return value = 0 > create_idx_tenpct_code_h() 0.45 seconds return value = 0 > create_idx_updates_double_bt() 0.46 seconds return value = 0 > create_idx_hundred_foreign() 0.41 seconds return value = 0 > populateDataBase() 11.54 seconds return value = 0 > > Error in test Counting tuples at (6746)osdb.c:294: > ... empty database -- empty results > perror() reports: Resource temporarily unavailable > > someone sighup'd the parent > > Any clue? > > ------------------------------------------ > > > Regards > Mallah. > > > > > > > ---------------------------(end of broadcast)--------------------------- > TIP 5: Have you checked our extensive FAQ? > > http://www.postgresql.org/docs/faqs/FAQ.html
After a long battle with technology,mallah@trade-india.com (Rajesh Kumar Mallah), an earthling, wrote: > the error mentioned in first email has been overcome > by running osdb on the same machine hosting the DB server. Yes, it seems unrealistic to try to run the "client" on a separate host from the database. I got the osdb benchmark running last week, and had to separate client from server. I had to jump through a fair number of hoops including copying data files over to the server. The benchmark software needs a bit more work... -- let name="cbbrowne" and tld="cbbrowne.com" in String.concat "@" [name;tld];; http://cbbrowne.com/info/lsf.html Nobody can fix the economy. Nobody can be trusted with their finger on the button. Nobody's perfect. VOTE FOR NOBODY.
Rajesh, Chris, > I got the osdb benchmark running last week, and had to separate client > from server. I had to jump through a fair number of hoops including > copying data files over to the server. The benchmark software needs a > bit more work... What about the OSDL's TPC-derivative benchmarks? That's a much more respected database test, and probably less buggy than OSDB. -- Josh Berkus Aglio Database Solutions San Francisco
Josh Berkus wrote:
pg_dump | psql is almost finishing in next 20 mins.
creating indexes at the moment :)
Really sorry can't rollback and delay anymore becoz my
website is *unavailable* for past 30 mins.
I ran OSDB .15 version and pg_bench .
Regds
Mallah.
Hmm... really sorry! myRajesh, Chris,I got the osdb benchmark running last week, and had to separate client from server. I had to jump through a fair number of hoops including copying data files over to the server. The benchmark software needs a bit more work...What about the OSDL's TPC-derivative benchmarks? That's a much more respected database test, and probably less buggy than OSDB.
pg_dump | psql is almost finishing in next 20 mins.
creating indexes at the moment :)
Really sorry can't rollback and delay anymore becoz my
website is *unavailable* for past 30 mins.
I ran OSDB .15 version and pg_bench .
Regds
Mallah.
<br /><br /> RC2 is running in production without any apparent problems<br /> till now. Well its difficult to say at themoment how much speed<br /> gain is there unless the heavy duty batch SQL scripts are run by<br /> cron. <br /><br />Count(*) and group by on large tables are significantly (5x) faster<br /> and better error reporting has made it easierto spot the faulty data.<br /> eg in fkey violation.<br /><br /> Will post the OSDB .15 versions' results on 7.3 &7.4 soon.<br /><br /> Regds<br /> Mallah.<br /><br /> Christopher Browne wrote:<br /><blockquote cite="midm34qxau6bb.fsf@wolfe.cbbrowne.com"type="cite"><pre wrap="">After a long battle with technology,<a class="moz-txt-link-abbreviated"href="mailto:mallah@trade-india.com">mallah@trade-india.com</a> (Rajesh Kumar Mallah), anearthling, wrote: </pre><blockquote type="cite"><pre wrap="">the error mentioned in first email has been overcome by running osdb on the same machine hosting the DB server. </pre></blockquote><pre wrap=""> Yes, it seems unrealistic to try to run the "client" on a separate host from the database. I got the osdb benchmark running last week, and had to separate client from server. I had to jump through a fair number of hoops including copying data files over to the server. The benchmark software needs a bit more work... </pre></blockquote><br />