From: system@SendSpamHere.ORG Sent: Friday, January 12, 2001 6:22 AM To: Info-VAX@Mvb.Saic.Com Subject: Re: Faster gzip, thank you Paul Repacholi! In article <3A5E32A5.43FA904C@GCE.com>, "Glenn C. Everhart" writes: >This is an area that's been a bit of a pet peeve of mine. >It should be possible to make a process default to using >readahead/writebehind and have the result basically invisible >to programs; the RMS semantics otherwise are not altered. > >Brian Schenkenberger did a few experiments with an RMS >call intercept to set the RAB bits needed forcibly though, >and found that the overhead of the intercept he had was >enough to make speed advantage often small. Just let me clarify the way the interection was accomplished. The first trial had each RMS service intercepted such that the ROPs were set on the way in and reset on the way out. This, it was completely transparent to the applicationt that it was being done. The overhead of saving the ROP bits going in and restoring going out was excessive. Later, intercerpting just the $CONNECT and having it set the ROP bits for all time did show remarkable performance increases for tests performed. Glenn and I left it at that. -- VAXman- OpenVMS APE certification number: AAA-0001 VAXman(at)TMESIS(dot)COM city, n., 1. a place where trees are cut down and streets are named after them.