[FFmpeg-devel] cache: read ahead to avoid excessive small requests

Michael Niedermayer michael at niedermayer.cc
Mon Sep 24 23:01:17 EEST 2018


On Sun, Sep 23, 2018 at 08:52:06PM +0200, Robert Nagy wrote:
> On Sun, Sep 23, 2018 at 1:40 PM Michael Niedermayer
> <michael at niedermayer.cc> wrote:
> >
> > On Sat, Sep 22, 2018 at 07:32:28PM +0200, Robert Nagy wrote:
> > > diff --git a/libavformat/cache.c b/libavformat/cache.c
> > > index 66bbbf54c9..48ff5ab363 100644
> > > --- a/libavformat/cache.c
> > > +++ b/libavformat/cache.c
> > > @@ -153,6 +153,38 @@ fail:
> > >      return ret;
> > >  }
> > >
> > > +static int cache_read_ahead(URLContext *h)
> > > +{
> > > +    Context *c= h->priv_data;
> > > +    int64_t r, read_ahead, pos;
> > > +    uint8_t buf[32768];
> > > +
> > > +    pos = c->logical_pos;
> > > +    read_ahead = c->read_ahead_limit < 0
> > > +        ? 512 * 512
> > > +        : FFMIN(512 * 512, c->read_ahead_limit);
> > > +
> > > +    while (read_ahead > 0) {
> > > +        r = ffurl_read(c->inner, buf, FFMIN(read_ahead, sizeof(buf)));
> > > +        if (r == AVERROR_EOF) {
> > > +            c->is_true_eof = 1;
> > > +            av_assert0(c->end >= c->logical_pos);
> > > +        }
> > > +        if (r<=0)
> > > +            break;
> > > +        c->inner_pos += r;
> > > +
> > > +        add_entry(h, buf, r);
> > > +        c->logical_pos += r;
> > > +        c->end = FFMAX(c->end, c->logical_pos);
> > > +        read_ahead -= r;
> > > +    }
> > > +
> > > +    c->logical_pos = pos;
> > > +
> > > +    return r < 0 ? r : 0;
> > > +}
> > > +
> > >  static int cache_read(URLContext *h, unsigned char *buf, int size)
> > >  {
> > >      Context *c= h->priv_data;
> >
> > > @@ -215,6 +247,10 @@ static int cache_read(URLContext *h, unsigned
> > > char *buf, int size)
> >
> > still not cleanly applying (due to new lines)
> >
> > Applying: cache: read ahead to avoid excessive small requests
> > error: corrupt patch at line 45
> > error: could not build fake ancestor
> > Patch failed at 0001 cache: read ahead to avoid excessive small requests
> > hint: Use 'git am --show-current-patch' to see the failed patch
> > When you have resolved this problem, run "git am --continue".
> > If you prefer to skip this patch, run "git am --skip" instead.
> > To restore the original branch and stop patching, run "git am --abort".
> >
> >
> > [...]
> >
> > --
> > Michael     GnuPG fingerprint: 9FF2128B147EF6730BADF133611EC787040B0FAB
> >
> > The greatest way to live with honor in this world is to be what we pretend
> > to be. -- Socrates
> > _______________________________________________
> > ffmpeg-devel mailing list
> > ffmpeg-devel at ffmpeg.org
> > http://ffmpeg.org/mailman/listinfo/ffmpeg-devel

>  cache.c |  111 ++++++++++++++++++++++++++++++++++++++++++++++++----------------
>  1 file changed, 84 insertions(+), 27 deletions(-)
> 5a1f791cd8dd72085fc27c9cdbf2d41a87f24fee  0001-cache-read-ahead-to-avoid-excessive-small-requests.patch
> From 17be70d9ffbfd1f55770e81958b597994abf2c99 Mon Sep 17 00:00:00 2001
> From: Robert Nagy <ronagy at icloud.com>
> Date: Sat, 22 Sep 2018 19:18:54 +0200
> Subject: [PATCH] cache: read ahead to avoid excessive small requests
> 

> This "fakes" a filler thread for reading ahead.

applies cleanly now, but why fake ?
reading alot more than requested would cause extra latency even if the user
application runs the format code in a seperate thread
It would be better to avoid this and not block a request for "block 1" any
longer than it takes to read "block 1"


[...]

-- 
Michael     GnuPG fingerprint: 9FF2128B147EF6730BADF133611EC787040B0FAB

Observe your enemies, for they first find out your faults. -- Antisthenes
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 181 bytes
Desc: not available
URL: <http://ffmpeg.org/pipermail/ffmpeg-devel/attachments/20180924/762559b1/attachment.sig>


More information about the ffmpeg-devel mailing list