Re: memory strangeness (fwd) - Mailing list pgsql-admin
From | Gregor Mosheh |
---|---|
Subject | Re: memory strangeness (fwd) |
Date | |
Msg-id | 20020705101359.V48217-100000@osiris.deathkeep.com Whole thread Raw |
In response to | Re: memory strangeness (fwd) (Curt Sampson <cjs@cynic.net>) |
Responses |
Re: memory strangeness (fwd)
|
List | pgsql-admin |
First off, thanks for your help, guys. I'm aware of the problems of over-allocating RAM, and I surely wouldn't want to force the buffers into swap. (thanks, Curt, for kern.ipc.shm_use_phys) On this particular system, though, it's doing nothing except PG. 384 MB of RAM, I can give PG 160 of it, which leaves me with some 170 MB of idle RAM. Here are the limits pulled from vmparam.h, by Curt's suggestion: #define MAXTSIZ (128UL*1024*1024) /* max text size */ #define DFLDSIZ (128UL*1024*1024) /* initial data size limit */ #define MAXDSIZ (512UL*1024*1024) /* max data size */ I'd read somewhere (obiously outdated) that the MAXDSIZ was 128 MB. I've since rebooted with the kernel.GENERIC... What's the sysctl setting I use to set/check the data size limits? Tom: You said that max_connections shouldn't be set as low as 5. I only intend to use 1 for the application, and I'll only need 1-2 for my own admin use, so a setting of 3 should work for my needs. Is there a technical reason it should be higher? On Fri, 5 Jul 2002, Curt Sampson wrote: > On Thu, 4 Jul 2002, Gregor Mosheh wrote: > > > Hiya. I've installed Postgres 7.2 on a dedicated FreeBSD system with 384 > > MB RAM. Because the system will be doing nothing except PG, I'd like to > > dump as much memory as possible into PG's shared memory. > > Well, see Tom's comments on this, 'cause that's what I always say > too. (I'm a NetBSD developer, but FreeBSD's internals aren't so > different, really.) > > However, if you wanted to do it both ways, and benchmark your > application, I'd be really interested in hearing about the results. > > > I rebuilt the kernel with very large limits: 330 MB on the MAXDSIZ and > > DFLDSIZ, and 330 MB for SHMMAXPAGES. This gives me: > > You don't need to rebuild for DFLDSIZ; you can always bump that > up to MAXDSIZ with sysctl (assuming you're root--do it before > you "su pgsql -c nadanada" in your startup script). And moving > MAXDSIZ to 330 MB looks like a reduction to me; FreeBSD's default in > 4.6-RELEASE is 512 MB. (NetBSD's is 1GB.) You probably want to check > /usr/include/machine/vmparam.h for default settings before changing > stuff like this, lest you accidently lower it instead. > > > I still cannot set PG's shared_buffers higher than 20000 (160 MB): > > Shared memory pages, IIRC, are locked, meaning that they cannot be > swapped. There's always a limit on the number of locked pages you > may have in the system in total, if only becuase the system has only > so much physical RAM. But for some reasont he RLIMIT_MEMLOCK parameter > on my nearest FreeBSD 4.6 system is a bit bogus: > > server2 $ ulimit -Ha | grep locked > lockedmem(kbytes) unlimited > > (I certainly cannot lock an unlimited number of pages in this 1GB > machine! And for some reason it's also saying that my RLIMIT_RSS > is unlimited; again, this should return no more than the physical > RAM available in the machine.) > > So I suspect you're running into some other, more secret, limit on > the number of pages that the system or a process may lock into RAM. > So you'll want to dig around the kernel to find out where this is, > and tweak it. > > cjs > -- > Curt Sampson <cjs@cynic.net> +81 90 7737 2974 http://www.netbsd.org > Don't you know, in this new Dark Age, we're all light. --XTC > > -- Gregor Mosheh, B.S. http://www.blackangel.net/ COMPUTER: An electronic entity which performs sequences of useful steps in a totally understandable, rigorously logical manner. If you believe this, see me about a bridge I have for sale in Manhattan.
pgsql-admin by date: