[prev in list] [next in list] [prev in thread] [next in thread] 

List:       busybox
Subject:    Re: [RFC/PATCH v3 0/8] readahead: daemon mode
From:       Bartosz_Gołaszewski <bartekgola () gmail ! com>
Date:       2015-08-25 13:04:28
Message-ID: CAFdkumg0qct-TADLNPq17H-yEQnA-9sszepR4aMMe+XAiZ0Zmg () mail ! gmail ! com
[Download RAW message or body]

2015-08-25 14:42 GMT+02:00 Denys Vlasenko <vda.linux@googlemail.com>:
> On Tue, Aug 25, 2015 at 1:09 PM, Bartosz Golaszewski
> <bartekgola@gmail.com> wrote:
>> While working on an embedded system running several big services, X-org,
>> fluxbox, Qt5 etc. where the boot-time was awful, it turned out that the
>> readahead implementation from systemd (the one that was nuked in 2014)
>> improves the time needed to start all the programs by a few seconds.
>>
>> This series introduces a small (LOC < 500) readahead daemon implementation
>> based on fanotify and readahead syscalls.
>
> I am agonizing on this.
>
> I do believe that you see some improvement in your setup.
> Possibly a substantial one.
>
> However.
>
> There is no "standard" readahead tool which is doing this, right?
>
> I usually accept patches to busybox tools which make them
> more compatible with "standard" tools: I want to reduce
> the frequency of cases where someone replaces a "standard"
> tool with busybox equivalent on a working system and something
> breaks.
>
> But if everything still works, it is just slow
> (example: "find -exec ... {} +" does work, it is not optimized
> compared to "find -exec ... {} ;"), this does not count as breakage.

I'm afraid I don't understand what you mean. Are you talking about
standard tools being slower?

>
> Your readahead addition does not match either of these conditions.
> It is not a copy of some standard tool, and if it is missing, things
> don't break, they are just slower.

The problem is: there is no standard readahead tool that could be
copied. In Debian there are two separate readahead packages available:
preload and readahead-fedora. Then there was the systemd one. They are
all much more complicated than what I wrote. I tried creating
something that would require zero maintenance. There's no standard to
conform to.

> How about this?
> Let's accept (some of) the changes you need for your applet
> to work, but you keep the applet per se out of the tree?
> When a "standard" readahead tool appears in "big" distros,
> we can return to this discussion.

With today's usage of SSDs and systemd adoption I don't expect that to
happen. Distros are not aimed at small boards with eMMCs, but at
servers and desktops. I can't force you to accept the patch, but there
are a lot of applets in busybox that are just random tools useful in
embedded systems - inotifyd for example which I find very useful, has
no standard counterpart. There's no bloat added by default since the
daemon mode is disabled in Config.

-- 
Best regards,
Bartosz Golaszewski
_______________________________________________
busybox mailing list
busybox@busybox.net
http://lists.busybox.net/mailman/listinfo/busybox
[prev in list] [next in list] [prev in thread] [next in thread]