[openib-general] Lustre over OpenIB Gen2
Eric Barton
eeb at bartonsoftware.com
Thu Nov 10 11:20:04 PST 2005
Yes, of course; I meant the QP.
Regarding the total number of outstanding RDMA work requests,
I can keep a separate cap on that, so if relatively few peers
are active, I push the maximum number of RDMAs at them, but if
many peers are active the number of active RDMAs per peer reduces.
However I guess this still means that CQ resources sufficient for
the maximum number of RDMAs I _could_ queue have to be allocated...
> -----Original Message-----
> From: Sean Hefty [mailto:mshefty at ichips.intel.com]
> Sent: Thursday, November 10, 2005 7:12 PM
> To: Eric Barton
> Cc: openib-general at openib.org
> Subject: Re: [openib-general] Lustre over OpenIB Gen2
>
>
> Eric Barton wrote:
> > 5. Should I pre-map all physical memory and do RDMA in
> page-sized fragments?
> > This avoids any mapping overhead at the expense of
> having much larger
> > numbers of queued RDMAs. Since I try to keep up to 8
> (by default) 1MByte
> > RDMAs active concurrently to any individual peer, with
> 4k pages I can have
> > up to 2048 RDMA work items queued at a time per peer.
>
> This is 20 million outstanding RDMA work requests per node.
>
> > And if I pre-map, can I be guaranteed that if I put the
> CQ into the error
> > state, all remote access to my memory is revoked (e.g.
> could a CQ I create
> > after I destroy the one I just shut down somehow alias
> with it such that a
> > pathalogically delayed RDMA could write my memory)?
>
> I think that you mean QP into the error state. If the QP is
> in the error state,
> then further access from a remote system should be impossible.
>
> - Sean
>
More information about the general
mailing list