Olga, I introduce two variables, MTMAXCACHE and MTEXCLUSIVE. If I understand your point I could combine them into one variable such as MTMALLOC_OPTIONS in the same manner as UMEM_DEBUG or the undocumented UMEM_OPTIONS. While we are at it we could also accept other mallocctl options such as MTCHUNKSIZE, MTINITBUFFER, MTDOUBLEFREE, and MTDEBUGPATTERN. It is a bit of scope creep but I will poll my advisers. I personally like opening up libmtmalloc a bit. It would make performance analysis without source code easier.
I have to have at least one environment variable because changing the maximum cache allocation value (MTMAXCACHE) must happen at init. rick On Tue, Jun 15, 2010 at 07:58:24PM +0200, ольга крыжановская wrote: > DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; > d=gmail.com; s=gamma; > h=domainkey-signature:mime-version:received:received:in-reply-to > :references:date:message-id:subject:from:to:cc:content-type; > bh=W3ykwvI/K+Sj5wLS3sSOMYgiRu404HybSwlP6XAuSmU=; > b=UqHjBIJv8oWMrjOJRE6SLHVhluGA1iFQAddEOTnuWqiNJmzDZ9c+zDJvZyu494DSgq > 7FYvVHcpiakob86zSEjgfnqmnQ+m8i8+5o53N7rDVd7djhIjLwdV3hP5XR+DPfyeFP0A > 6G2FlXSp+10etF0dJ77pVHiz98gGnoPyv1Nb4= > DomainKey-Signature: a=rsa-sha1; c=nofws; > d=gmail.com; s=gamma; > h=mime-version:in-reply-to:references:date:message-id:subject:from:to > :cc:content-type; > b=iM5KmsDQnRV2bQQwfvZHqY9zsjh+13Fl8B1KspgI4JQXT9lgnQnzVbEg74aoLHq5dR > IqbKmMQy1UBWIasvxNOehuw5cywibXlWkHF5ARxtYsr125Abg3ElB+Dn+QmG49OZOjJ0 > nbexQuLVPvaYTcVKPQ3c4qlfhOtKFPBqwqJxc= > In-Reply-To: <[email protected]> > Date: Tue, 15 Jun 2010 19:58:24 +0200 > Subject: Re: Performance Improvements for libmtmalloc [PSARC/2010/212 > FastTrack, timeout 06/15/2010] > From: ольга крыжановская <[email protected]> > To: Peter Dennis <[email protected]> > Cc: [email protected], rick weisner <[email protected]> > X-Source-IP: mail-iw0-f172.google.com [209.85.214.172] > X-CT-RefId: str=0001.0A090206.4C17BF42.0109,ss=1,fgs=0 > > I have a comment: Why do you use 3 environment variables? Environment > variables are very expensive to use, they significantly increase start > up time of an application and slow down all applications which use > getenv() or putenv() if the size of the environment is large. libast > had the same problem, with bitter and measurable impact on start up > time, and replaced all the allocator tunable variables into 1 variable > to improve start up time. > > IMO one environment variable to control the libmtmalloc behaviour > should be enough. > > Olga > > On Tue, Jun 15, 2010 at 7:36 PM, Peter Dennis <[email protected]> wrote: > > This case times out today but there have been no comments from > > any PSARC members - any reviewers ? > > > > thanks > > pete > > _______________________________________________ > > opensolaris-arc mailing list > > [email protected] > > > > > > -- > , _ _ , > { \/`o;====- Olga Kryzhanovska -====;o`\/ } > .----'-/`-/ [email protected] \-`\-'----. > `'-..-| / http://twitter.com/fleyta \ |-..-'` > /\/\ Solaris/BSD//C/C++ programmer /\/\ > `--` `--` -- Rickey C. Weisner Software Development and Performance Specialist Principal Software Engineer Systems Quality Office Oracle Corporation cell phone: 615-308-1147 email: [email protected] _______________________________________________ opensolaris-arc mailing list [email protected]
