home *** CD-ROM | disk | FTP | other *** search
- Path: sparky!uunet!olivea!decwrl!sgi!rhyolite!vjs
- From: vjs@rhyolite.wpd.sgi.com (Vernon Schryver)
- Newsgroups: comp.arch
- Subject: Re: Solaris (was: Sun 600MP Benchmark Anomaly)
- Message-ID: <nodh0gc@rhyolite.wpd.sgi.com>
- Date: 25 Jul 92 19:04:32 GMT
- References: <1992Jul16.210353.6464@sequent.com> <14pkdrINN5ni@uk-news.UK.Sun.COM> <1992Jul25.142435.9821@athena.mit.edu>
- Organization: Silicon Graphics, Inc. Mountain View, CA
- Lines: 46
-
- In article <1992Jul25.142435.9821@athena.mit.edu>, jfc@athena.mit.edu (John F Carr) writes:
- > In article <14pkdrINN5ni@uk-news.UK.Sun.COM>
- > adrian.cockcroft@uk.sun.com writes:
- >
- > >I have run a two processor S600MP with a Solaris 2 beta release with
- > >both CPU's at over 95% system time (to show that it can be done).
- >
- > How much of the 95% is spent in the locking code? How much is spent
- > waiting for locks?
- > --
- > John Carr (jfc@athena.mit.edu)
-
-
- Ouch!
-
- You must have see some of the some MP UNIX kernel implementations, and
- wondered how they get anything done what with all of the locking and
- P'ing and V'ing.
-
- It is an odd fact of human nature that once an organization has
- access to multiprocesses, two things invariably happen:
-
- (1) about half of the people (inside and out; all of customers,
- marketing, and implementors) get bent out of shape if anything
- in the system is not fully semaphored with tiny grains. These
- people want "pre-emption points" every dozen lines in critical
- sections. They want every instance of every data structure to
- have its own lock, and sometimes 2 or 3 per data structure to
- fix the consequent effects of having lots of tiny overlapping
- critcial sections.
-
- (2) about the other half of the same universe of people get bent
- out of shape if they cannot "dedicate" or "allocate" or
- "reserve" a CPU for a single kind of work or process or user.
- These guys are the sort who demand the ability to put reserve a
- CPU for the disk controller, the terminal serial I/O code,
- Fortran compilations, or whatever else they think most
- important.
-
-
- I'm in the third group, and have argued with both groups seemingly
- continuously since the first symmetric MP system I worked on in the
- late 70's. It gets boring after awhile
-
-
- Vernon Schryver, vjs@sgi.com
-