musl 14x slower?

musl 14x slower?

[John Mudd]

[-- Attachment #1: Type: text/plain, Size: 503 bytes --]

Please cc me on replies.

I built Postgres with musl. I used version 1.1.0. That works well and I've
been distributing musl Postgres in production.

I still have an install running a previous build using standard libc from
and old 2.4 kernel build. The "native" build runs 14x faster. And this is
not trying to do anything fancy, just sequentially reading records from a
400 MB Postgres table.

native: 0:25
musl: 5:42

I know my musl version is dated but is this slow performance to be expected?

John

[-- Attachment #2: Type: text/html, Size: 680 bytes --]

Re: musl 14x slower?

[Rich Felker]

On Fri, Mar 13, 2015 at 03:53:20PM -0400, John Mudd wrote:
> Please cc me on replies.
> 
> I built Postgres with musl. I used version 1.1.0. That works well and I've
> been distributing musl Postgres in production.
> 
> I still have an install running a previous build using standard libc from
> and old 2.4 kernel build. The "native" build runs 14x faster. And this is
> not trying to do anything fancy, just sequentially reading records from a
> 400 MB Postgres table.
> 
> native: 0:25
> musl: 5:42
> 
> I know my musl version is dated but is this slow performance to be expected?

No. Is it possible that you built musl and/or the musl-linked
postgresql with -O0? I would suggest using perf(1) to measure where
all the additional time is being spent.

Rich

Re: musl 14x slower?

[John Mudd]

[-- Attachment #1: Type: text/plain, Size: 733 bytes --]

I think I found the cause. I added --enable-cassert Postgres option when
building Postgres with musl. This makes more sense.

On Fri, Mar 13, 2015 at 3:53 PM, John Mudd <johnbmudd@gmail.com> wrote:

> Please cc me on replies.
>
> I built Postgres with musl. I used version 1.1.0. That works well and I've
> been distributing musl Postgres in production.
>
> I still have an install running a previous build using standard libc from
> and old 2.4 kernel build. The "native" build runs 14x faster. And this is
> not trying to do anything fancy, just sequentially reading records from a
> 400 MB Postgres table.
>
> native: 0:25
> musl: 5:42
>
> I know my musl version is dated but is this slow performance to be
> expected?
>
> John
>

[-- Attachment #2: Type: text/html, Size: 1231 bytes --]

Re: Re: musl 14x slower?

[Rich Felker]

On Fri, Mar 13, 2015 at 04:26:10PM -0400, John Mudd wrote:
> I think I found the cause. I added --enable-cassert Postgres option when
> building Postgres with musl. This makes more sense.

Have you tried rebuilding without that option? How does the speed of
the resulting build compare? If it's still significantly slower (or
faster? :) than the "native" (presumably glibc?) build it would be
nice to know.

Rich