Re: [Lksctp-developers] [PATCH] Allow room for bundled SACK / ECNE in sctp data fragmentation

[Vlad Y. <vla...@hp...>]

The problem with is that when you don't need to bundle chunks, you
send less data then you can really send.

Running iperf and netperf with you patch gives a reduction of about 10%
in throughput (this is on top of already really dismal numbers).

-vlad

Mark Butler wrote:
> This patch reduces the sctp data fragmentation point to allow room for a
> small SACK chunk and an ECNE chunk to be bundled with DATA chunks from
> fragmented messages. This keeps SACK and ECNE from being delivered in
> standalone control-chunk only packets when there is bidirectional
> traffic.  It also keeps the fragmentation point a multiple of four bytes.
> Signed-off-by: Mark Butler <bu...@mi...>
> 
> ===
> diff -Nru  a/include/net/sctp/sctp.h b/include/net/sctp/sctp.h
> --- a/include/net/sctp/sctp.h    2005-12-14 16:50:41.000000000 -0700
> +++ b/include/net/sctp/sctp.h    2005-12-21 20:14:05.000000000 -0700
> @@ -442,13 +442,18 @@
> {
>     int frag = pmtu;
> 
> -    frag -= sp->pf->af->net_header_len;
> -    frag -= sizeof(struct sctphdr) + sizeof(struct sctp_data_chunk);
> +    frag = min_t(int, frag, SCTP_MAX_CHUNK_LEN);
> +
> +    frag -= sp->pf->af->net_header_len;  +    frag -= sizeof(struct
> sctphdr) + sizeof(struct sctp_data_chunk);
> +
> +    frag -= 24;  /* allow room for small SACK */
> +    frag -= 8;   /* allow room for ECNE or CWR */
> 
>     if (sp->user_frag)
>         frag = min_t(int, frag, sp->user_frag);
> 
> -    frag = min_t(int, frag, SCTP_MAX_CHUNK_LEN);
> +    frag &= ~0x3;   /* multiple of 4 bytes */
> 
>     return frag;
> }
> 
> 
> 
> -------------------------------------------------------
> This SF.net email is sponsored by: Splunk Inc. Do you grep through log
> files
> for problems?  Stop!  Download the new AJAX search engine that makes
> searching your log files as easy as surfing the  web.  DOWNLOAD SPLUNK!
> http://ads.osdn.com/?ad_id=7637&alloc_id=16865&op=click
> _______________________________________________
> Lksctp-developers mailing list
> Lks...@li...
> https://lists.sourceforge.net/lists/listinfo/lksctp-developers
>