RE: ENRP Server Hunt Timer
Johnson Walter-CWJ002 <[email protected]>
| Newsgroups | gmane.ietf.rserpool |
|---|---|
| Message-ID | <6F8DFFA2C996D711945800065BFC9E4A119D542C@il02exm11> |
Thomas Dreibholz <mailto:[email protected]> wrote: > -----BEGIN PGP SIGNED MESSAGE----- > Hash: SHA1 > > On Friday 29 October 2004 00:25, Johnson Walter-CWJ002 wrote: >> Currently the draft states in ENRP server hunt procedure of section >> 3.6 bullet item SH5 on page 21 that >> >> SH5) If the T5-Serverhunt timer expires the following should be >> performed: >> >> SH5.1) The endpoint MUST double the value of the T5-Serverhunt >> timer. SH5.2) The endpoint SHOULD stop the establishment of >> associations and connections. SH5.2) The endpoint >> SHOULD repeat trying to establish an association or >> connection by proceeding to step SH1. It SHOULD attempt to >> select a different set of transport addresses to connect >> to. >> >> Currently the T5-Serverhunt timer keeps on doubling and there is no >> upper limit on this so it appears it could go on forever. It would >> seem reasonable to add a ceiling on the max value that the >> T5-Serverhunt can reach (e.g. a variable max.severhunt.value to >> section 5.1) or place to place a threshold of attempts to make called >> MAX-SERVER-HUNTS in section 5.3. If either max is reached, a signal >> could be sent to an ASAP application of the failure to find a home >> ENRP server. > > An upper limit for doubling the timeout is useful. > > But it may be useful to keep on trying to connect to a NS even when > the maximum timeout has been reached, because it may be better for the > application to wait instead of failing. For this case, the number of > trials must also be limited (this could be a configuration option, to > be set by the application - I have implemented this behavior in the > rsplib prototype). > > Example: > TimeoutLimit=4s > MaxTrials=100 > > Trail #1: timeout=1s > Trail #2: timeout=2s > Trail #3: timeout=4s (no more doubling) > Trail #4: timeout=4s > ... > Trail #100: timeout=4s => After this trial, signal failure to the > user. > I agree with your proposal above.