> Subject: RE: Re: Re: Re: [RFC PATCH] RDMA/siw: Experimental e2e negotiation > of GSO usage. > > > -----"Krishnamraju Eraparaju" <krishna2@xxxxxxxxxxx> wrote: ----- > > >To: "Bernard Metzler" <BMT@xxxxxxxxxxxxxx> > >From: "Krishnamraju Eraparaju" <krishna2@xxxxxxxxxxx> > >Date: 05/15/2020 03:58PM > >Cc: faisal.latif@xxxxxxxxx, shiraz.saleem@xxxxxxxxx, > >mkalderon@xxxxxxxxxxx, aelior@xxxxxxxxxxx, dledford@xxxxxxxxxx, > >jgg@xxxxxxxx, linux-rdma@xxxxxxxxxxxxxxx, bharat@xxxxxxxxxxx, > >nirranjan@xxxxxxxxxxx > >Subject: [EXTERNAL] Re: Re: Re: Re: [RFC PATCH] RDMA/siw: > >Experimental e2e negotiation of GSO usage. > > > >Here is the rough prototype of iwpmd approach(only kernel part). > >Please take a look. > > > This indeed looks like a possible solution, which would not affect the wire > protocol. > > Before we move ahead with that story in any direction, I would really really > appreciate to hear what other iWarp vendors have to say. > > 0) would other vendors care about better performance > in a mixed hardware/software iwarp setting? > > 1) what are the capabilities of other adapters in that > respect, e.g. what is the maximum FPDU length it > can process? > > 2) would other adapters be able to send larger FPDUs > than MTU size? > > 3) what would be the preferred solution - using spare > MPA protocol bits to signal capabilities or > extending the proprietary iwarp port mapper with that > functionality? > > Thanks very much! > Bernard. > Hi Bernard - If we receive larger FPDU than MTU its handled in software and therefore is a hit on perf. We do support jumbo packets but we do not transmit FPDUs greater than MTU size. I recommend we do not add unspec'd bits into the MPA protocol for gso negotiation. netlink based approach or iwpmd sounds more reasonable. Hope this helps. Shiraz
