| Lists: | pgsql-hackers |
|---|
| From: | "Zeugswetter Andreas SB SD" <ZeugswetterA(at)spardat(dot)at> |
|---|---|
| To: | "Matthew T(dot) O'Connor" <matthew(at)zeut(dot)net>, "Christopher Browne" <cbbrowne(at)acm(dot)org> |
| Cc: | <pgsql-hackers(at)postgresql(dot)org> |
| Subject: | Re: Experimental patch for inter-page delay in VACUUM |
| Date: | 2003-11-04 13:47:37 |
| Message-ID: | 46C15C39FEB2C44BA555E356FBCD6FA4962057@m0114.s-mxs.net |
| Views: | Raw Message | Whole Thread | Download mbox | Resend email |
| Lists: | pgsql-hackers |
> Or... It seems to me that we have been observing something on the order
> of 10x-20x slowdown for vacuuming a table. I think this is WAY
> overcompensating for the original problems, and would cause it's own
> problem as mentioned above. Since the granularity of delay seems to be
> the problem can we do more work between delays? Instead of sleeping
> after every page (I assume this is what it's doing) perhaps we should
> sleep every 10 pages,
I also think doing more than one page per sleep is advantageous since
it would still allow the OS to do it's readahead optimizations.
I suspect those would fall flat if only one page is fetched per sleep.
Andreas
| From: | Andreas Pflug <pgadmin(at)pse-consulting(dot)de> |
|---|---|
| To: | Zeugswetter Andreas SB SD <ZeugswetterA(at)spardat(dot)at> |
| Cc: | pgsql-hackers(at)postgresql(dot)org |
| Subject: | Re: Experimental patch for inter-page delay in VACUUM |
| Date: | 2003-11-04 13:55:34 |
| Message-ID: | 3FA7AFD6.9010604@pse-consulting.de |
| Views: | Raw Message | Whole Thread | Download mbox | Resend email |
| Lists: | pgsql-hackers |
Zeugswetter Andreas SB SD wrote:
>>Or... It seems to me that we have been observing something on the order
>>of 10x-20x slowdown for vacuuming a table. I think this is WAY
>>overcompensating for the original problems, and would cause it's own
>>problem as mentioned above. Since the granularity of delay seems to be
>>the problem can we do more work between delays? Instead of sleeping
>>after every page (I assume this is what it's doing) perhaps we should
>>sleep every 10 pages,
>>
>>
>
>I also think doing more than one page per sleep is advantageous since
>it would still allow the OS to do it's readahead optimizations.
>I suspect those would fall flat if only one page is fetched per sleep.
>
>
So maybe the setting shouldn't be "n ms wait between vacuum actions" but
"vacuum pages to handle before sleeping 10 ms".
Regards,
Andreas