home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: comp.sys.isis
- Path: sparky!uunet!gatech!rpi!batcomputer!cornell!ken
- From: ken@cs.cornell.edu (Ken Birman)
- Subject: Re: 2.2.5: spooler problems
- Message-ID: <1993Jan3.142806.12077@cs.cornell.edu>
- Organization: Cornell Univ. CS Dept, Ithaca NY 14853
- References: <QfEQ_ve00h5K13C2cS@cs.cmu.edu>
- Date: Sun, 3 Jan 1993 14:28:06 GMT
- Lines: 49
-
- In article <QfEQ_ve00h5K13C2cS@cs.cmu.edu> Sean.Levy@cs.cmu.edu writes:
- >Hey all,
- >
- >I cannot get the spooler to function properly under 2.2.5. If this is
- >all better in 2.2.7, I appologize for the waste in bandwidth, but I
- >haven't had a chance to install the new version, so...
- >
-
- Well, I think that it works in 2.2.5 and 2.2.7, but we have run into
- a number of problems with one of our commercial users. Tim Clark is the
- spooler expert now and will be back from a vacation/business trip around
- January 15. Meanwhile, here's my understanding of the situation.
-
- - The spooler works on V2.2.5 except when the destination WAN is
- specified as "all". "all" causes many problems and can hang the
- failure-recovery code -- bad news.
-
- For V2.2.7, the spooler should work ("all", too) but is annoyingly
- slow. Moreover, it lacks flow control mechanisms and hence if you
- try even moderately hard, you can get the spooler itself to consume a lot
- of memory and eventually shut down.
-
- For WAN message rates of up to about 5 per second, average, the V2.2.7/
- V3.0.7 spooler should be fine. (1k messages).
-
- - Tim recoded the problematic part of the wide-area code and got a big
- speedup plus better handling of large messages and better flow control.
- His version is available now for beta-test users (now means "when he
- gets back from his trip"). It gets about 30 messages per second to a
- remote node and by the time of the real release we may be able to push
- this up by an additional 20-30%.
-
- The main issues are that "all" was implemented incorrectly in V2.2.5/V3.0.7
- and just didn't work, and that the disk logging mechanism used was unfortunate
- in those versions of the spooler. V3.0.7/V2.2.7 fixes the bug in all, but
- we needed to recode the way that the wan spool files themselves are managed
- to get acceptable performance -- thats where the 30/second numbers come from;
- the limit is related to the bandwidth of a flushed write to a local file
- system... we may introduce a less safe mode in which the disk IO is not]
- flushed on every WAN write; this would let us push to about 100/second...
-
- I hope this is helpful. Tim will send more info when he gets back. For
- CMU, I guess I would get Tim to prove the new version and work with that
- directly -- no point in reinstalling it after just 2 weeks!
-
- --
- Kenneth P. Birman E-mail: ken@cs.cornell.edu
- 4105 Upson Hall, Dept. of Computer Science TEL: 607 255-9199 (office)
- Cornell University Ithaca, NY 14853 (USA) FAX: 607 255-4428
-