From: Tom Lane <tgl@sss.pgh.pa.us>
To: Alvaro Herrera <alvherre@2ndquadrant.com>
cc: Hugh Ranalli <hugh@whtc.ca>, Justin Pryzby <pryzby@telsasoft.com>,
        pgsql-performance@lists.postgresql.org
Subject: Re: Perplexing, regular decline in performance
In-reply-to: <20190626190743.GA15026@alvherre.pgsql>
References: <20190626190743.GA15026@alvherre.pgsql>
Comments: In-reply-to Alvaro Herrera <alvherre@2ndquadrant.com>
	message dated "Wed, 26 Jun 2019 15:07:43 -0400"
MIME-Version: 1.0
Content-Type: text/plain; charset="us-ascii"
Content-ID: <8278.1561576694.1@sss.pgh.pa.us>
Date: Wed, 26 Jun 2019 15:18:14 -0400
Message-ID: <8279.1561576694@sss.pgh.pa.us>
Precedence: bulk

Alvaro Herrera <alvherre@2ndquadrant.com> writes:
> On 2019-Jun-26, Hugh Ranalli wrote:
>> From my research in preparing for the upgrade, I understood transparent
>> huge pages were a good thing, and should be enabled. Is this not correct?

> It is not.

Yeah ... they would be a good thing perhaps if the quality of the kernel
implementation were better.  But there are way too many nasty corner
cases, at least with the kernel versions people around here have
experimented with.  You're best off to disable THP and instead manually
arrange for Postgres' shared memory to use huge pages.  I forget where
to look for docs about doing that, but I think we have some.

			regards, tom lane