Re: [Nbd] NBD using SDP
- To: Wouter Verhelst <w@...112...>
- Cc: email@example.com
- Subject: Re: [Nbd] NBD using SDP
- From: John Leidel <john.leidel@...17...>
- Date: Tue, 16 Oct 2007 09:31:32 -0500
- Message-id: <1192545092.4558.40.camel@...140...>
- In-reply-to: <20071016141928.GB4218@...39...>
- References: <1192490510.4558.2.camel@...140...> <20071016105002.GA4218@...39...> <1192538339.4558.24.camel@...140...> <20071016141928.GB4218@...39...>
Yeah, my fingers hit `ctrl-s` instead of `ctrl-v`... :-)
You'll need to make sure that you either have the openfabrics stack
installed or the kernel has been built with Infiniband support
On Tue, 2007-10-16 at 16:19 +0200, Wouter Verhelst wrote:
> On Tue, Oct 16, 2007 at 07:38:59AM -0500, John Leidel wrote:
> > The patches essentially allow one to take advantage of the transport
> > layer RDMA capabilities of the fabric medium using the OpenFabrics
> > software stack. The most common transport fabric is Infiniband,
> > although Chelsio 10GbE cards also support the protocol.
> > You'll probably see the most benefit when using fairly large block sizes
> > as the transfer mechanisms in OpenFabrics tend to favor large messages
> > [although, they are building in better support for smaller messages].
> > I'm not sure what to expect if you mount swap in this manner. It would
> > either give you quite a performance boost, or completely hose up the
> > swap.
> > A more complete discussion on SDP is here:
> > http://en.wikipedia.org/wiki/Sockets_Direct_Protocol
> > and here
> I think you forgot to paste in a link here.
> Anyway, yeah, that sounds like something useful; I'll probably integrate
> this into nbd, provided I can (at the very least) get it to compile and
> make it a runtime configurable option or so.