On 9/1/2016 4:21 PM, Jason Gunthorpe wrote: > On Thu, Sep 01, 2016 at 02:52:55PM -0500, Steve Wise wrote: >>>> >>>> So we have nominations for >>>> >>>> srptools http://git.openfabrics.org/?p=~bvanassche/srptools.git/.git. >>>> perftest git://git.openfabrics.org/~grockah/perftest.git master >>> >>> I would drop perftest out. Here's why. I like the idea that this >>> package is for software that interacts with the kernel directly. It >>> needs to, in some way, involve itself with the kernel char interface or >>> with the kernel netlink interface. So far, everything meets that >>> critera. Perftest does not. So, in the same fashion as libfabrics and >>> fabtests, I would leave perftest out, but that isn't to say a separate >>> package that pulled together various testing apps for RDMA could be >>> created too. >>> >> >> I see your point. That is fine with me. > > I'm happy either way, but I will point out that libibverbs and > librdmacm sources contains various things similar to perftest (eg the > pingpong stuff, _bw, etc). > > If we were to merge perftest I'd probably aim to consolidate all of > those similar examples/tests/etc under one directory in the source > tree. > > I'd cast perftest and related as diagnostic tools to help users > validate verbs hardware is working as expected, so still under the > plumbing umbrella. > > Jason > I can see your point. Maybe start with things as they are, then in some subsequent work we consolidate all of the example programs from libibverbs/librdmacm into one place, then we look at what programs could possibly be folded in too (perftest would probably work, but for instance, qperf works on tcp and other things not-rdma and is generally useful even without rdma, so I don't know that I would fold it in). -- Doug Ledford <dledford@xxxxxxxxxx> GPG Key ID: 0E572FDD
Attachment:
signature.asc
Description: OpenPGP digital signature
