From: ludo@gnu.org (Ludovic Courtès)
To: Stefan Israelsson Tampe <stefan.itampe@gmail.com>
Cc: guile-devel <guile-devel@gnu.org>
Subject: Re: vhash speed thread safeness
Date: Tue, 29 Oct 2013 18:54:56 +0100 [thread overview]
Message-ID: <87zjpsnerz.fsf@gnu.org> (raw)
In-Reply-To: <CAGua6m3R1Y4i34-CekAhN5safW3R13wvP_HL5k93-cBpQ27zDw@mail.gmail.com> (Stefan Israelsson Tampe's message of "Tue, 29 Oct 2013 15:21:34 +0100")
Stefan Israelsson Tampe <stefan.itampe@gmail.com> skribis:
> On Tue, Oct 29, 2013 at 1:34 PM, Ludovic Courtès <ludo@gnu.org> wrote:
>
>> Hi, Stefan,
>>
>> Stefan Israelsson Tampe <stefan.itampe@gmail.com> skribis:
>>
>> > I did some tests witha C-based vhash implementation, it's possible to
>> > increse the speed by 30x compared to current vlist imlpementation in
>> > guile. It would be possible to get this speed today if one implemented
>> > vhash-assoc as a VM op. Anyway using the source as a standard lib will
>> > get you about 10x speed increase.
>>
>> As we discussed, I don’t really like the idea of implementing that much
>> in C.
>>
> My neither, but it is good to see that in cache friendly cases we can
> improve the situation 30x. Gives us a goal to strive for. Also the main
> intention is to use vashses for guile-log. For this we can note.
OK.
[...]
>> > Another pressing need is that vhashes are not thread safe,
>>
>> Section 2.8 of Bagwell’s paper proposes a simple solution. All that is
>> missing AFAICS is an atomic test-and-set VM op to implement it (which
>> may also be useful in other places.)
>>
>> What do you think of this approach?
>
>
> For vlists it's probably a good idea, I don't know if it's enough for
> vhashes though.
Oooh, right, sorry for overlooking that.
> Maybe you need a mutex. But lock overhead will be significant
Surely, especially if it’s a fat mutex.
Hmm hmm. Of course that could be an argument for doing some C
(primitives, not VM ops), but looking at ‘%vhash-assoc’, that would
clearly mean reimplementing pretty much all of the vlist + vhash in C,
which sucks.
I wonder if there’s some other data structure with similar properties
that doesn’t have the thread-safety issue. Maybe Ian has an idea?
The weight-balanced trees in MIT/GNU Scheme look interesting:
http://www.gnu.org/software/mit-scheme/documentation/mit-scheme-ref/Weight_002dBalanced-Trees.html
Thoughts?
Thanks,
Ludo’.
next prev parent reply other threads:[~2013-10-29 17:54 UTC|newest]
Thread overview: 6+ messages / expand[flat|nested] mbox.gz Atom feed top
2013-10-18 16:21 vhash speed thread safeness Stefan Israelsson Tampe
2013-10-29 12:34 ` Ludovic Courtès
2013-10-29 14:21 ` Stefan Israelsson Tampe
2013-10-29 17:54 ` Ludovic Courtès [this message]
2013-10-29 18:50 ` Ian Price
2014-03-24 20:59 ` Andy Wingo
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
List information: https://www.gnu.org/software/guile/
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=87zjpsnerz.fsf@gnu.org \
--to=ludo@gnu.org \
--cc=guile-devel@gnu.org \
--cc=stefan.itampe@gmail.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).