From: Hal Rosenstock <hal-LDSdmyG8hGV8YrgS2mwiifqBs+8SCbDb@public.gmane.org>
To: Jens Domke <domke.j.aa@m.titech.ac.jp>
Cc: linux-rdma-u79uwXL29TY76Z2rM5mHXA@public.gmane.org,
Torsten Hoefler <htor-gy3b+zu4XSAfv37vnLkPlQ@public.gmane.org>
Subject: Re: [PATCH 2/2] OpenSM: DFSSSP - workaround for better VL balancing
Date: Fri, 31 May 2013 05:48:25 -0400 [thread overview]
Message-ID: <51A871E9.50600@dev.mellanox.co.il> (raw)
In-Reply-To: <1369880550-3031-1-git-send-email-domke.j.aa@m.titech.ac.jp>
On 5/29/2013 10:22 PM, Jens Domke wrote:
> Currently, DFSSSP maps the src/dest paths statically to certain VLs.
> Especially for deadlock-free topologies this can result in an
> unfair balancing. Some VLs within one link might be overused,
> which results in slower bandwidth for some src/dest pairs.
>
> The fix changes the VL assignment in two ways: first we balance the
> number of paths per VL; and second we randomly assign the VL
> as long as this doesn't violate the deadlock-freedom.
>
> 1) The balancing splits the paths across available free VLs, so that
> the maximal number of paths per VL is minimized. We save the number
> of VLs for each deadlock-free channel dependency graph. E.g. for
> 8 VLs, paths per CDG: {14,5,1} => balanced VLs: {{3,3,3,3,2},{3,2},1}
> we have 5 VLs to choose from for CDG(0), two for CDG(1) and
> one for CDG(2).
>
> 2) get_dfsssp_sl(...) will use the information of (1) to randomly
> assign the VL for one src/dest pair within the possible number of
> VLs. E.g. for a src/dest pair of CDG(0) we have 5 VLs to choose from,
> therefore VL := baseVL + rand()%5
>
> Signed-off-by: Jens Domke <domke.j.aa@m.titech.ac.jp>
Thanks. Applied with removal of debug code.
-- Hal
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
prev parent reply other threads:[~2013-05-31 9:48 UTC|newest]
Thread overview: 2+ messages / expand[flat|nested] mbox.gz Atom feed top
2013-05-30 2:22 [PATCH 2/2] OpenSM: DFSSSP - workaround for better VL balancing Jens Domke
2013-05-31 9:48 ` Hal Rosenstock [this message]
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
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=51A871E9.50600@dev.mellanox.co.il \
--to=hal-ldsdmyg8hgv8yrgs2mwiifqbs+8scbdb@public.gmane.org \
--cc=domke.j.aa@m.titech.ac.jp \
--cc=htor-gy3b+zu4XSAfv37vnLkPlQ@public.gmane.org \
--cc=linux-rdma-u79uwXL29TY76Z2rM5mHXA@public.gmane.org \
/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