On Mon, Jul 20, 2009 at 6:17 AM, Sasha Khapyorsky<[email protected]> wrote: > On 08:39 Mon 20 Jul , Nicolas Morey-Chaisemartin wrote: >> So what is best in your mind? >> Bring everything path into the perf counter struct (meaning all plugin >> will have to change to fit the new API) or keeping the approach I proposed. > > We can add new counter to existing structure at the end and to not break > API. > > The better option is to unify API (counter structures) of course, but it > is different change. Basically using current (modified) plugin API it is > not required to put a counters to plugin, some sort of "counters ready" > event would be enough.
There's also a counter validity issue (in terms of PortXmitWait) if the difference between 0 and not supported is to be indicated. -- Hal >> I'd rather choose the second solution not only because it's already >> done, but in the future I'd like to have as many information as possible >> on contention related counters (xmit_wait per VL for examples) and >> probably some day vendor specific info will appear in this struct. > > Yes, but it is different counters and you will need to add something > anyway. > > Sasha > _______________________________________________ > general mailing list > [email protected] > http://lists.openfabrics.org/cgi-bin/mailman/listinfo/general > > To unsubscribe, please visit http://openib.org/mailman/listinfo/openib-general > _______________________________________________ general mailing list [email protected] http://lists.openfabrics.org/cgi-bin/mailman/listinfo/general To unsubscribe, please visit http://openib.org/mailman/listinfo/openib-general
