Thread: Speed...

Speed...

From
"Svenne Krap"
Date:
Hi again,

I have been following this list for approx. 6600 messages (i think
that is 2 months or so.. .) - an yes.. Like the Cardassian Order I
keep everything :)

A couple of time, ppl has been talking about speed..

What about a couple of benchmarks from your systems, so that us
newbies (or almost) can se, how much we can expect to be able to
tune up to..

I can "go first" :)
An comments for further tuning is apreciated... btw.. the box gets
its memory doubled sometimes next week.

My setup:
Duron 750 MHz.
256 Megs memory
IBM Deskstar IDE-harddrive (no raid, no nothing)
Running 7.1.2

### BTW: this is not the box mentioned in another question this
evening.. this I my private "toybox"..

some content of postgresql.conf :
max_connections = 512
hostname_lookup = false
sort_mem = 4096
shared_buffers = 1024 # min 16
sync = false
#wal_buffers = 8
#wal_files = 0
#wal_sync_method = fsync
#wal_debug = 0
#commit_delay = 0
#commit_siblings = 5
#checkpoint_segments = 3
#checkpoint_timeout = 300

(a side question ... does this mean, that I even haven't enabled the
"magic" WAL ?)

using pgbench from /contrib i get :
sk@kitten:/usr/src/postgresql-7.1.2/contrib/pgbench > ./pgbench
pgbench -c 40
starting vacuum...end.
transaction type: TPC-B (sort of)
scaling factor: 1
number of clients: 40
number of transactions per client: 10
number of transactions actually processed: 400/400
tps = 106.943436(including connections establishing)
tps = 125.524811(excluding connections establishing)

sk@kitten:/usr/src/postgresql-7.1.2/contrib/pgbench > ./pgbench
pgbench -c 100
starting vacuum...end.
transaction type: TPC-B (sort of)
scaling factor: 1
number of clients: 100
number of transactions per client: 10
number of transactions actually processed: 1000/1000
tps = 88.924498(including connections establishing)
tps = 101.360871(excluding connections establishing)

sk@kitten:/usr/src/postgresql-7.1.2/contrib/pgbench > ./pgbench
pgbench -c 200
starting vacuum...end.
transaction type: TPC-B (sort of)
scaling factor: 1
number of clients: 200
number of transactions per client: 10
number of transactions actually processed: 2000/2000
tps = 67.320263(including connections establishing)
tps = 74.500643(excluding connections establishing)

sk@kitten:/usr/src/postgresql-7.1.2/contrib/pgbench > ./pgbench
pgbench -c 300
starting vacuum...end.
transaction type: TPC-B (sort of)
scaling factor: 1
number of clients: 300
number of transactions per client: 10
number of transactions actually processed: 3000/3000
tps = 48.291555(including connections establishing)
tps = 51.892547(excluding connections establishing)

Tia

Svenne
--
svenne@krap.dk
http://www.krap.dk
ICQ 5434480
PGP-key http://keys.pgp.dk:11371/pks/lookup?op=get&search=0xDF484022
PGP @ http://www.pgp.com / http://www.phpi.com

RE: Speed...

From
"Tim Mickol"
Date:
some quick (and I might add disappointing) benchmark results...

must tune..
.
will be performing -and publishing results- of more comprehensive and
rigorous suite on:
 - 2x700MHz cpu
 - 4GB RAM
 - 6x35GB 10k rpm RAID 5 system
 - linux 2.4.4
 - postgres 7.1.2-2

I've perf-tuned many an oracle db - look forward to the assistance of all in
tuning postgres


dev db system profile
----------------------------------
dual 600MHz PII w/256kb L2 cache
1 GB Memory
PERC RAID5 (5 x 9.1 GB - 7200rpm)
Linux kernel 2.4.4 #1 SMP
postgresql 7.1.2-2
max_connections = 512
hostname_lookup = false
sort_mem = 2048
shared_buffers = 256 # min 16
effective_Cache=16384
fsync = false



pgbench results:
----------------------------------
[postgres@sanddb /tmp]$ ./pgbench -c40 pgbench
starting vacuum...end.
transaction type: TPC-B (sort of)
scaling factor: 1
number of clients: 40
number of transactions per client: 10
number of transactions actually processed: 400/400
tps = 29.194674(including connections establishing)
tps = 30.592961(excluding connections establishing)

[postgres@sanddb /tmp]$ ./pgbench -c 100 pgbench
starting vacuum...end.
transaction type: TPC-B (sort of)
scaling factor: 1
number of clients: 100
number of transactions per client: 10
number of transactions actually processed: 1000/1000
tps = 21.655443(including connections establishing)
tps = 22.375797(excluding connections establishing)



regards,

tjm

-----Original Message-----
From: pgsql-general-owner@postgresql.org
[mailto:pgsql-general-owner@postgresql.org]On Behalf Of Svenne Krap
Sent: Friday, June 22, 2001 2:23 PM
To: pgsql-general@postgresql.org
Subject: [GENERAL] Speed...


Hi again,

I have been following this list for approx. 6600 messages (i think
that is 2 months or so.. .) - an yes.. Like the Cardassian Order I
keep everything :)

A couple of time, ppl has been talking about speed..

What about a couple of benchmarks from your systems, so that us
newbies (or almost) can se, how much we can expect to be able to
tune up to..

I can "go first" :)
An comments for further tuning is apreciated... btw.. the box gets
its memory doubled sometimes next week.

My setup:
Duron 750 MHz.
256 Megs memory
IBM Deskstar IDE-harddrive (no raid, no nothing)
Running 7.1.2

### BTW: this is not the box mentioned in another question this
evening.. this I my private "toybox"..

some content of postgresql.conf :
max_connections = 512
hostname_lookup = false
sort_mem = 4096
shared_buffers = 1024 # min 16
sync = false
#wal_buffers = 8
#wal_files = 0
#wal_sync_method = fsync
#wal_debug = 0
#commit_delay = 0
#commit_siblings = 5
#checkpoint_segments = 3
#checkpoint_timeout = 300

(a side question ... does this mean, that I even haven't enabled the
"magic" WAL ?)

using pgbench from /contrib i get :
sk@kitten:/usr/src/postgresql-7.1.2/contrib/pgbench > ./pgbench
pgbench -c 40
starting vacuum...end.
transaction type: TPC-B (sort of)
scaling factor: 1
number of clients: 40
number of transactions per client: 10
number of transactions actually processed: 400/400
tps = 106.943436(including connections establishing)
tps = 125.524811(excluding connections establishing)

sk@kitten:/usr/src/postgresql-7.1.2/contrib/pgbench > ./pgbench
pgbench -c 100
starting vacuum...end.
transaction type: TPC-B (sort of)
scaling factor: 1
number of clients: 100
number of transactions per client: 10
number of transactions actually processed: 1000/1000
tps = 88.924498(including connections establishing)
tps = 101.360871(excluding connections establishing)

sk@kitten:/usr/src/postgresql-7.1.2/contrib/pgbench > ./pgbench
pgbench -c 200
starting vacuum...end.
transaction type: TPC-B (sort of)
scaling factor: 1
number of clients: 200
number of transactions per client: 10
number of transactions actually processed: 2000/2000
tps = 67.320263(including connections establishing)
tps = 74.500643(excluding connections establishing)

sk@kitten:/usr/src/postgresql-7.1.2/contrib/pgbench > ./pgbench
pgbench -c 300
starting vacuum...end.
transaction type: TPC-B (sort of)
scaling factor: 1
number of clients: 300
number of transactions per client: 10
number of transactions actually processed: 3000/3000
tps = 48.291555(including connections establishing)
tps = 51.892547(excluding connections establishing)

Tia

Svenne
--
svenne@krap.dk
http://www.krap.dk
ICQ 5434480
PGP-key http://keys.pgp.dk:11371/pks/lookup?op=get&search=0xDF484022
PGP @ http://www.pgp.com / http://www.phpi.com

---------------------------(end of broadcast)---------------------------
TIP 3: if posting/reading through Usenet, please send an appropriate
subscribe-nomail command to majordomo@postgresql.org so that your
message can get through to the mailing list cleanly


Re: Speed...

From
"Svenne Krap"
Date:
To:                 tmickol@combimatrix.com
Copies to:          svenne@krap.dk, pgsql-general@postgresql.org
Subject:            Re: [GENERAL] Speed...
Date sent:          Sat, 23 Jun 2001 11:56:03 -0400
From:               Tom Lane <tgl@sss.pgh.pa.us>

> "Tim Mickol" <tmickol@combimatrix.com> writes:
> > some quick (and I might add disappointing) benchmark results...
>
> As has been pointed out before, you must use a pgbench scale factor
> higher than one if you are interested in making meaningful measurements
> for more than one concurrent client.  At scale one, there is only one
> "branch", so *every* transaction needs to update the same branch
> balance, so there's effectively no concurrency.
>
>             regards, tom lane
>


Hi Tom,

I must admit, that I somewhat hoped you and/or Bruce would "jump
in" with your knowledge.

How high shall the scaling factor be (or in other words, what
pgbench-parameters would you prefer for our little tuning session),
and what would you think is "target" to tune upto (using pgbench
with your parameters) for a box like mine (Duron 750, 512Megs of
memory, Normal IDE-disks) ?

Others may give their advice too :)

TIA

Svenne

--
svenne@krap.dk
http://www.krap.dk
ICQ 5434480
PGP-key http://keys.pgp.dk:11371/pks/lookup?op=get&search=0xDF484022
PGP @ http://www.pgp.com / http://www.phpi.com

Re: Speed...

From
Tom Lane
Date:
"Tim Mickol" <tmickol@combimatrix.com> writes:
> some quick (and I might add disappointing) benchmark results...

As has been pointed out before, you must use a pgbench scale factor
higher than one if you are interested in making meaningful measurements
for more than one concurrent client.  At scale one, there is only one
"branch", so *every* transaction needs to update the same branch
balance, so there's effectively no concurrency.

            regards, tom lane