Follow @Openwall on Twitter for new release announcements and other news
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date: Mon, 23 Apr 2018 12:43:01 -0400
From: Rich Felker <dalias@...c.org>
To: musl@...ts.openwall.com
Subject: Re: What's wrong with musl's malloc

On Mon, Apr 23, 2018 at 01:50:47AM -0500, A. Wilcox wrote:
> On 04/22/18 14:34, Markus Wichmann wrote:
> > Each of these is constant time. I am not sure optimizing fragmentation
> > is worth reducing the performance for. Especially in today's desktop
> > and server systems, where anything below 16GB of RAM will just get
> > laughed off the stage (slight exaggeration).
> 
> 
> this is so far off the mark I am actually writing a reply on a Sunday.
> 
> FreeBSD is currently discussing optimising NFS to run better on 256 MB
> RAM i386 machines, and is considering the case where a non-zero number
> of people are running it in 64 MB virtual machines.
> 
> Adelie is running KDE Plasma 5 on a Pentium III with 256 MB RAM, with
> enough left over for a few tabs in Otter browser.  In fact, our base
> specifications for a server install are a 40 MB 486 or better (you can
> squeeze text mode much lower, but apk won't run right below 40; our
> baseline is 32 MB, however, on the PowerPC).
> 
> "Today's" systems fall in to two buckets: computers with insane specs
> bought by people in the upper classes, and used computers with lower
> specs bought by people in the lower classes.  after spending a
> significant chunk of my life (~15 years) in both, I can't see defences
> like "anything below 16GB of RAM will get laughed off the stage" as
> anything *but* internalised classism.  as engineers, our jobs are to
> make software for all people, not for the chosen few that can afford
> 16GB of RAM.
> 
> hell, my new Talos cost me so much that even though I'd technically be
> considered upper-class now, I still couldn't afford more than 8 GB RAM
> for it in the beginning (I'm planning to upgrade in the next months as
> my finances allow).
> 
> I'm sorry if this comes off as ranty or preachy.  I'm just trying to
> enlighten everyone that 1) not everyone has or *can afford* 16 GB RAM;
> 2) that's a poor excuse for not tuning software to be the best it can be.
> 
> After all, wasted memory is wasted memory, whether you have a lot or a
> little.  (Wouldn't you rather fit more photos / videos / text in there
> instead of wasting it on malloc overhead?

This. All of this. Considerations like the above were largely the
motivation for musl.

Beyond the question of whether everyone has or can afford 16GB of ram
financially -- while I believe it's important, I'm also less
idealistic than I used to be about whether we can engineer efficient
replacements for bloated stuff faster than Moore's law makes the
bloated stuff usable on affordable machines -- there are questions of
what you can do with what you can afford that immediately become
relevant once you stop thinking just about the OS as something that
runs on a physical present-day high-end box and consider uses like
containers, full virtual machines, embedded systems, etc. I recall
hearing an story I never verified that there was one PC game with
virtualized computers inside the game world running musl on them. This
is also the whole reason people are interested in Alpine on Docker.

Rich

Powered by blists - more mailing lists

Confused about mailing lists and their use? Read about mailing lists on Wikipedia and check out these guidelines on proper formatting of your messages.