Owen, ----- Original Message ----- From: "Owen Scott Medd" <[EMAIL PROTECTED]> To: "Heikki Tuuri" <[EMAIL PROTECTED]> Cc: <[EMAIL PROTECTED]> Sent: Wednesday, June 04, 2003 3:13 PM Subject: Re: malloc'ing 2GB+ of memory in mysql
> I know we are facing this same question right now (I have 8 way servers > with 16GB of memory running MySQL, with 5 GB sitting unused while the poor > innodb buffer pool sits starved for memory). Do we replace these servers > with 4 way Opterons (are there 8 ways promised yet?) or is there another > answer? best to test first with a borrowed Opteron. I want to see the stability and that fsync() and other kernel calls work fast. A customer tested recently a big Itanium II box and its performance looked ok. > In particular, what is involved with being a "feature sponsor"? > (lol... show me the money, eh?). And, more importantly, have you done any > estimates about how long it would take to implement AWE in the RedHat AS > environment (which happens to be our environment too)? The sponsorship of multiple tablespaces will take all available time till September 15th, 2003. On that day you will be able to put every InnoDB table to its own file :). After that it might be a 2 week job to get AWE to Linux. The idea is to attach shared memory areas to the mysqld process on demand. Since you already have the appropriate hardware, testing is no problem. > Curious, > Owen Best regards, Heikki > On Wed, 4 Jun 2003, Heikki Tuuri wrote: > > Per, > > > > I remember someone also reporting a problem that glibc or Linux does not > > allow creation of new threads if one has allocated >= 2 GB user memory. I > > think there are problems in where the OS places the excutable, thread > > stacks, etc. > > > > So it is uncharted territory. Oracle seems to have an option to use AWE > > memory on the Red Hat Advanced Server. Then the limit is 64 GB on a 32-bit > > Intel processor. InnoDB-4.1 has the same AWE option, but only on certain > > Windows versions. > > > > If Itanium and Opteron fail to take off, or a feature sponsor appears, I may > > consider implementing AWE also on Linux. The memory crunch is getting so > > severe that I believe some 64-bit processor must become common by 2005. > > > > Best regards, > > > > Heikki Tuuri > > Innobase Oy > > http://www.innodb.com > > Transactions, foreign keys, and a hot backup tool for MySQL > > Order MySQL technical support from https://order.mysql.com/ > > > > > > ----- Original Message ----- > > From: "Per Andreas Buer" <[EMAIL PROTECTED]> > > Newsgroups: mailing.database.mysql > > Sent: Wednesday, June 04, 2003 1:21 PM > > Subject: malloc'ing 2GB+ of memory in mysql > > > > > > > Hi > > > > > > The Mysql binary distribution for IA32-linux is statically linked with > > > glibc. glibc malloc limits memory allocations to 2GB, which means that a > > > buffer in mysql can't grow beyond 2GB. This is due to some paranoia in > > > glibc malloc - they don't rely on the size to be an unsigned int - which > > > limits the size to 2^31 on any 32-bit platform. > > > > > > Has anyone tried to remove this limit in glibc malloc or linking Mysql > > > with another malloc implementation? > > > > > > -- > > > Per Andreas Buer > > > > > > -- > > > MySQL General Mailing List > > > For list archives: http://lists.mysql.com/mysql > > > To unsubscribe: > > http://lists.mysql.com/[EMAIL PROTECTED] > > > > > > > > > > > > > -- > USMail: InterGuide Communications, 230 Lyn Anne Court, Ann Arbor, MI 48103 > phone: +1 734 997-0922 fax: +1 734 661-0324 > mailto:[EMAIL PROTECTED] http://www.interguide.com/~osm/ > > [ Sometimes wrong. Never in doubt. ] > -- MySQL General Mailing List For list archives: http://lists.mysql.com/mysql To unsubscribe: http://lists.mysql.com/[EMAIL PROTECTED]
