home *** CD-ROM | disk | FTP | other *** search
- Xref: sparky comp.parallel:2788 comp.object:4619
- Newsgroups: comp.parallel,comp.object
- Path: sparky!uunet!gatech!hubcap!fpst
- From: erwin@trwacs.fp.trw.com (Harry Erwin)
- Subject: Re: Load balancing in object-oriented systems
- Message-ID: <1992Dec21.192942.15328@hubcap.clemson.edu>
- Apparently-To: comp-parallel@uunet.uu.net
- Sender: fpst@hubcap.clemson.edu (Steve Stevenson)
- Organization: TRW Systems Division, Fairfax VA
- References: <1992Dec18.191720.16237@prism.poly.edu>
- Date: Mon, 21 Dec 1992 18:11:11 GMT
- Approved: parallel@hubcap.clemson.edu
- Lines: 25
-
- I haven't seen much useful on computer performance engineering in
- object-oriented systems, particularly those like Smalltalk-80 or my C++
- multitasking stuff, where a lot of CPU is used up by the executive. I will
- say that even some Ada tasking environments show evidence that the
- programmers don't understand these issues. (Integrating inter-task and
- external IO tends to be a pain and a half, since both the operating system
- and the executive are messing around with the same task input queues.)
- Running a debugger can be eye-opening, particularly if there is a lot of
- inheritance and objects are being zeroed at allocation, as required by the
- TCSEC. Since load-balancing requires some sort of predictive model for the
- load generated by specific applications, the potential for unstable
- behavior and thrashing is significant. I've done some work here, but not
- enough.
-
- Checking the CMG and SIGMETRICS back issues, it doesn't look like there's
- much that has been published. If you come up with something, you might
- think about publishing. An article titled "CPE for OO Systems" would
- probably be an automatic "accept" for CMG' 93. Think about it.
-
- Cheers,
- --
- Harry Erwin
- Internet: erwin@trwacs.fp.trw.com
-
-
-