| From: | Andrey Borodin <x4mmm(at)yandex-team(dot)ru> |
|---|---|
| To: | Andres Freund <andres(at)anarazel(dot)de> |
| Cc: | Юрий Соколов <funny(dot)falcon(at)gmail(dot)com>, Thomas Munro <thomas(dot)munro(at)enterprisedb(dot)com>, Konstantin Knizhnik <k(dot)knizhnik(at)postgrespro(dot)ru>, Pg Hackers <pgsql-hackers(at)postgresql(dot)org> |
| Subject: | Re: [HACKERS] Clock with Adaptive Replacement |
| Date: | 2018-05-01 04:30:38 |
| Message-ID: | 65626298-9176-460A-8D64-06CD417D79E4@yandex-team.ru |
| Views: | Whole Thread | Raw Message | Download mbox | Resend email |
| Thread: | |
| Lists: | pgsql-hackers |
Hi!
> 30 апр. 2018 г., в 23:15, Andres Freund <andres(at)anarazel(dot)de> написал(а):
> On 2018-04-30 15:39:08 +0500, Andrey Borodin wrote:
>> I suspect that performance benefits can be not that big or even
>> marginal, if we do not extend comprehensive eviction strategy with
>> bgwriter fixes and O_DIRECT.
>
> If so, then the improvements aren't real. Bgwriter doesn't matter for
> read-only workloads. O_DIRECT doesn't matter much if shared_buffers are
> 60+% of OS memory. And even disregarding that, you can just compute
> cache hit ratios to see whether things are improving.
Even considering simply changing eviction strategy - it is not just about hit ratio. It is also about eviction complexity less than O(N).
But I think you are right. If we compare performance effect of half-measures in the real system, probably it is more accurate than comparing isolated algorithms in a sand box.
Best regards, Andrey Borodin.
| From | Date | Subject | |
|---|---|---|---|
| Next Message | Andres Freund | 2018-05-01 05:08:46 | Re: wal_consistency_checking reports an inconsistency on master branch |
| Previous Message | Tom Lane | 2018-05-01 04:13:30 | Re: "could not reattach to shared memory" on buildfarm member dory |