home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: comp.lang.c++
- Path: sparky!uunet!mcsun!sunic!sics.se!eua.ericsson.se!euas62c36!euamts
- From: euamts@eua.ericsson.se (Mats Henricson)
- Subject: Re: Special-purpose-heap-managers - refere
- Message-ID: <1992Dec21.102620.21845@eua.ericsson.se>
- Sender: news@eua.ericsson.se
- Nntp-Posting-Host: euas62c36.eua.ericsson.se
- Reply-To: euamts@eua.ericsson.se
- Organization: Ellemtel Telecom Systems Labs, Stockholm, Sweden
- References: <1992Dec20.153218.4338@daimi.aau.dk>
- Date: Mon, 21 Dec 1992 10:26:20 GMT
- Lines: 54
-
- In article 4338@daimi.aau.dk, sabroe@daimi.aau.dk (Morten Sabroe Mortensen) writes:
- #euamts@eua.ericsson.se (Mats Henricson) writes:
- #
- #>In article 11510@daimi.aau.dk, sabroe@daimi.aau.dk (Morten Sabroe Mortensen) writes:
- #>#Could anyone, -who have tried-, tell, what kind of experiences they have with
- #>#special-purpose-heap-managers like 'Pool', in what kind of program, with what
- #>#kind of result/speed-up?
- #
- #>We did a test of several class libraries, of which one was using a Pool class.
- #>In one test (putting and removing integers from a Set class) the Pool-based
- #>solution was ~20 times faster. That was the extreme case. In another test the
- #>gain was "only" 2.3 times.
- #
- #Nice, nice, - sounds interesting!!!
- #Are some of these classes, you tested, availably on the net -
- #do you remember where?
-
- None of them can be fetched for free, I'm afriad - you have to pay for them!
- I don't want to hang them out on the net, but I can tell you the Pool-class
- library: USL Standard Components. Just to avarage out this free ego-boosting
- for the USL crowd, I can tell you that you can find some rather messy code
- in the USL SC (just an example):
- -->grep -c goto getbuf.c
- 5
- i.e. 5 goto:s in one single file!
-
- #>BUT REMEMBER: all this performance difference is NOT because of a Pool class.
- #>The libraries differed in many other respects, which affected performance.
- #
- #>And then again: size and performance isn't everything...
- #
- #You're right! -size and performance isn't everything!
- #
- #BUT: If you want high performance, doing SOMETHING to dynamic allocation
- # seems to be one of the better ways to gain speed, i think!
- # _IF_ allocation could be handled transparently, -like 'Pool' _nearly_
- # can do, then.... But arrays...-one should be careful with arrays,
- # since they can't be managed transparently...YET!
- #
- #It ought to be possible to let a special manager manage ALL kinds of
- #allocation involving a certain class, -that's what I want, I think!
- #TRUE TRANSPARENCY - ABSTRACTION, ABSTRACTION, ABSTRACTION - HIGHLY WANTED!
- #
- #I'm thinking: When constructing a special-purpose-heap-manager, one should
- #be careful not to ramble into some kind of ALIGNMENT-PROBLEM (it could
- #easily happen!!!) - that is, it should, -of course(!)-, be portable;
- #size and performance isn't everything!
-
- I don't know how portable the USL SC are.
-
- Mats Henricson
- Ellemtel Telecom Systems Labs
- Stockholm
- Sweden
-