On 8/30/07, Sasha Khapyorsky <[EMAIL PROTECTED]> wrote: > On 04:19 Thu 30 Aug , Yevgeny Kliteynik wrote: > > Sasha Khapyorsky wrote: > > > On 19:12 Wed 29 Aug , Hal Rosenstock wrote: > > >> On 8/29/07, Sasha Khapyorsky <[EMAIL PROTECTED]> wrote: > > >>> On 17:48 Wed 29 Aug , Hal Rosenstock wrote: > > >>>>> OK, there are three ways we want this thing to work: > > >>>>> 1. QoS is off > > >>>>> 2. The old QoS is on but w/o policy file > > >>>>> 3. The old QoS is on, plus reading policy file > > >>>>> > > >>>>> The first option is clear: if a user doesn't turns QoS on (-Q), QoS is > > >>>>> off as before. > > >>>>> > > >>>>> Second and third options: if QoS is on, OpenSM looks for policy file > > >>>>> in > > >>>>> the default > > >>>>> location or in other location that was provided by user. If the file > > >>>>> is > > >>>>> not found, > > >>>>> QoS works as before. > > >>>> This sounds OK to me and is my first preference. > > >>>> > > >>>>> Do we want to add additional option for "enhanced" QoS? > > >>>>> If so, we will have three QoS-ralated command line options: > > >>>>> - option for turning the QoS on (currently -Q) > > >>>>> - option to turn the new QoS on (some new letter - must get > > >>>>> one quick before they all run out... :) > > >>>>> - option for policy file location if differs from default (currently > > >>>>> -Y) > > >>>> This seems like the least preferable to me. Also, would need to deal > > >>>> with both on which seems to mean use new QoS. > > >>>> > > >>>>> Alternatively, we can turn -Q option into levels: > > >>>>> -Q 0: QoS is off (default) > > >>>>> -Q 1: old QoS is on > > >>>>> -Q 2: old QoS plus reading policy file > > >>>> This one also seems OK to me (second preference). > > >>>> > > >>>> Anyone else with an opinion on this ? Sasha ? > > >>> I like -Q and -Y as Yevgeny proposed. > > >> So is that the first option ? > > > Yes. It is simplest and provides the same functionality. > > > > This is what's implemented right now > > > > >> Actually, I think I like the third option best now that I think more > > >> on this. It seems a little odd to me to rely on the policy file not > > >> being present to determine which QoS to run. Seems a little cleaner > > >> this way to me. > > > We need file name option anyway, so things like '-Q 1 -Y ...' are > > > unclear. Also it would be nice to have "universal" (not for "two QoS") > > > user interface in order to not change it later. > > > > I would say that -Q and -Y are enough, but it poses some questions: > > Do we want to allow the case when a user has policy file in a default > > location, but he wants OpenSM to ignore this file and still have QoS on? > > -Y '' should be fine then. > > > In case of partitions file we don't have an option to ignore partition > > config file in a default location. > > > > And if OpenSM ignores policy file, what would it mean? > > Would it be the "old" QoS? > > Yes. Is there another useful options? > > > And when the setup part of the new QoS will be ready, would we still want > > the SL2VL and VLArb tables to appear in the opts file? > > Don't think. Why we need it in two places?
Are you saying that the existing QoS will get these tables from the new QoS syntax (and the existing syntax eliminated) ? If so, what about backward compatibility ? Also, how straightforward is it to specify the equivalent as the existing syntax ? Will a conversion tool be supplied ? -- Hal > 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
