Date: Thu, 1 Sep 2005 09:58:51 +0200 From: Holger Kipp <hk@alogis.com> To: Norberto Meijome <freebsd@meijome.net> Cc: freebsd-amd64@freebsd.org Subject: Re: Which SCHED_ for DB server Message-ID: <20050901075851.GA19900@intserv.int1.b.intern> In-Reply-To: <4316A5BC.1000405@meijome.net> References: <4316A5BC.1000405@meijome.net>
next in thread | previous in thread | raw e-mail | index | archive | help
Hello, On Thu, Sep 01, 2005 at 04:54:52PM +1000, Norberto Meijome wrote: > I'm building a server that will run PostgreSQL with a database > containing several 10s of million records. The only things happening on > this box will be the SQL processes and other processes to parse raw data > and load into the DB. Users = a few connections via HTTP from an > intranet server (not more than 5 concurrently). > I was wondering what is the best SCHED_ to set in the kernel. > I currently have SCHED_4BSD but was wondering if _ULE would be better > for this I would not care that much about the scheduler you want to use. In worst case, recompile the kernel and look at performance / userinteraction. What looks worse (imho) is this: > 4 x SATA 150 drives, 90% of it RAID5 with gvinum. I do not know your applications, but if writing to disk is an issue: - WAL-files should be written to physically separated disks. - Write performance to RAID5 is not very good (tm). This is both with softraid and with hardware raid, although those raid controllers with 64MB or higher (+batterybackup) might increase the visible performance if only a few processes generate disk-i/o-load. - ATA is not good for production - better use SCSI. Best regards, Holger Kipp
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20050901075851.GA19900>