I compared 2.6 with elevator=deadline. It did bring some improvement in
performance. But still it does not beat 2.4.
Attached are three files for details.
I also ran a simple insert benchmark to insert a million record in a simple
table with a small int and a varchar(30).
Here are the results
10K/xact 277 sec
100K/xact 271 sec
10K/xact Not done
100K/xact 257 sec
10K/xact 243 sec
100K/xact 246 sec
It seems that I noted a test result wrongly. I need to do it again.
Overall 2.6 needs some real IO improvements. Of course it could do better on
I guess there is no point bothering this with kernel hackers. They know this
stuff already, right.
Looking forward to next release of kernel and hope it improves things...
pgsql-performance by date
|Next:||From: William Yu||Date: 2003-08-29 16:33:51|
|Subject: Re: Hardware recommendations to scale to silly load|
|Previous:||From: Tom Lane||Date: 2003-08-29 16:18:35|
|Subject: Re: PL/pgSQL functions - text / varchar - havy performance |