On 6/8/16, 10:54, "linux-nfs-owner@xxxxxxxxxxxxxxx on behalf of Tom Talpey" <linux-nfs-owner@xxxxxxxxxxxxxxx on behalf of tom@xxxxxxxxxx> wrote: >On 6/7/2016 6:01 PM, Jason Gunthorpe wrote: >> On Tue, Jun 07, 2016 at 05:51:04PM -0400, Chuck Lever wrote: >> >>> There is a practical number of MRs that can be allocated >>> per device, I thought. And each MR consumes some amount >>> of host memory. >> >>> xprtrdma is happy to allocate thousands of MRs per QP/PD >>> pair, but that isn't practical as you start adding more >>> transports/connections. >> >> Yes, this is all sane and valid. >> >> Typically you'd target a concurrency limit per QP and allocate >> everything (# MRs, sq,rq,cq depth, etc) in accordance with that goal, >> I'd also suggest that target concurrency is probably a user tunable.. > >This is already present - it's the RPC slot table depth. It would be >great, IMO, to make this a mount parameter since it is transport- and >also server-dependent, but currently I believe it is just a kernel >RPC module parameter. That would be a mount parameter which would only apply to RDMA since everything else uses dynamic slot allocation these days. I’d prefer to avoid that. Trond ��.n��������+%������w��{.n�����{���fk��ܨ}���Ơz�j:+v�����w����ޙ��&�)ߡ�a����z�ޗ���ݢj��w�f