2003-01-29 01:03:09

by Cliff White

[permalink] [raw]
Subject: [OSDL][BENCHMARK] DBT2 results question


As Andrew Morton requested, we have rerun the OSDL Database Test 2 on cached
and non-cached versions comparing 2.4.21-pre3 and 2.5.49 and this time
we have vmstat data.
Actually, we have a _ton of data, which is why i'm posting this.
What data in particular are you kernel coders looking for?

We could graph each statistic for each set of runs for a side by
side comparison 2.4 versus 2.5.

We can also post the raw data, but 40 runs is a bit much to post on a mailing
list.

What would you find most useful?

We will get the runs Andrew requested on 2.5.58 when we can get past the
I/O bug Dave Olien reported last week.

cliffw



2003-01-29 01:17:17

by Andrew Morton

[permalink] [raw]
Subject: Re: [OSDL][BENCHMARK] DBT2 results question

Cliff White wrote:
>
> As Andrew Morton requested, we have rerun the OSDL Database Test 2 on cached
> and non-cached versions comparing 2.4.21-pre3 and 2.5.49 and this time
> we have vmstat data.
> Actually, we have a _ton of data, which is why i'm posting this.
> What data in particular are you kernel coders looking for?
>
> We could graph each statistic for each set of runs for a side by
> side comparison 2.4 versus 2.5.
>
> We can also post the raw data, but 40 runs is a bit much to post on a mailing
> list.
>
> What would you find most useful?

Didn't you have a standardised set of number for the dbt1 runs? They
were pretty good if I recall correctly.

It is very hard to find dbt1 results at osdl.org. I normally have to
grep the lse-tech archives to find anything at all.

Ah yes.
http://www.osdl.org/projects/dbt1prfrns/results/2proc/linux-2.4.18-1tier/index.html

That seems a reasonable summary.

> We will get the runs Andrew requested on 2.5.58 when we can get past the
> I/O bug Dave Olien reported last week.

That isn't fixed yet, and there's no workaround, alas.