[openib-general] RFC: SDP plans

[Bill Boas]

Michael,

Having a stable, performant SDP in Rel. 1.0 available from Novel and RedHat
is absolutely critical for Wall Street and others.

If anyone diagrees with this requirement please speak up!

Bill.


On 2/27/06, Michael S. Tsirkin <mst at mellanox.co.il> wrote:
>
> I started preparing a stable linux SDP implementation,
> with the eye towards mainline inclusion.
>
> The idea is to get to a drastically simple code base and get this admitted
> in
> mainline, then add enhancements.
>
> The plan (as compared to existing SDP implementation) includes:
> - Use CMA API
> - Reuse generic code from sock.c
> SO_SNDBUF/SO_RCVBUF should work properly
> - Use sock_lock_t and simple spin_lock_bh for socket locking
> - Use skbuff and standard skbuff queues (in struct sock)
> for incoming/outgoing messages
> - Implement transport-level queues by simple circular buffer,
> attach BSDH by s/g
> - Set socket bits to signal the need for control messages
> - Single CQ, perform all CQ polling from interrupt context
> - Code must be sparse-clean, keep network data in __beXX structures
> - Proper use of DMA API
> - Use sysfs for statistics, entry per socket
> - Only support the BCopy mode for starters -
> any advertisements will be answered with sendsm
> - Only support synchronous operations for starters: applications
> can use userspace AIO emulation, and hopefully generic AIO support
> in kernel will mature meanwhile.
>
> --
> Michael S. Tsirkin
> Staff Engineer, Mellanox Technologies
> _______________________________________________
> openib-general mailing list
> openib-general at openib.org
> http://openib.org/mailman/listinfo/openib-general
>
> To unsubscribe, please visit
> http://openib.org/mailman/listinfo/openib-general
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openfabrics.org/pipermail/general/attachments/20060227/03294229/attachment.html>