Date: Thu, 29 Jan 1998 13:21:47 -0600 From: "Jeffrey J. Mountin" <mountin.man@mixcom.com> To: "Troy Settle" <rewt@i-plus.net>, "(ML) FreeBSD ISP" <isp@FreeBSD.ORG> Subject: Re: filesystems Message-ID: <3.0.3.32.19980129132147.007041dc@198.137.186.100> In-Reply-To: <018401bd2cda$3a4d2970$3a4318d0@abyss.b.nu>
next in thread | previous in thread | raw e-mail | index | archive | help
At 12:20 PM 1/29/98 -0500, Troy Settle wrote: >In light of the recent 'Sendmail - low on space' thread, here's a few >thoughts on partitioning disk space... > >A year ago, when faced with the prospect of setting up an ISP, I wasn't >sure how to go about allocating disk space. I had a single server, and 2 >HDs totaling about 3.5 gig. I ended up creating a number of smallish >(250-500 meg) partitions, and playing musical filesystems with them. > >Last month, I realized that I had a totally crappy filesystem, so I got a >4.3 gig SCSI drive to replace the 2 IDE drives, and started fresh. Personally I would have went for 2 - 2 GB drives or 4 - 1 GB... >For a 1000 user system, here's what I came up with: >/dev/sd0a 95359 38469 49262 44% / >/dev/sd0s1d 1813583 364656 1303841 22% /home >/dev/sd0s1h 508655 247143 220820 53% /play >/dev/sd0s1g 127151 685 116294 1% /tmp >/dev/sd0s1f 289231 152536 113557 57% /usr >/dev/sd0s1e 1017327 482766 453175 52% /var > >/ ~100 megs (plenty of breathing room) >swap ~250 megs >/usr ~300 megs Might be small, but /usr/ports could be a mount point if needed in the future. >/tmp ~128 megs More than enough. >/var ~1 gig >/home ~1.8 gig >/play ~500 megs > > >The /play partition is temporary, it will soon be /var/local >(/usr/local/var) for my mysql databases > >The /var partition holds mail and my squid cache until I get a seperate >server for squid. > >I originally wanted to mount /usr as read-only, and have /usr/local on a >seperate partition, but found myself running out of room. > >Eventually, I'll have /home moved over to a second HD, then move /var to >where /home was, and /var/local to where /var was, freeing up ~500 megs >for /usr/local so I can mount /usr as readonly. All adjustments for what your system's needs are, but periodically it would be a good idea to run 'iostat -c 60' and see if the drives are reaching their transfer limits and if the load is spread over the drives relatively evenly. One web server I maintain has long peaks where the drive's transfer is max'd out and needs 2 faster drives to replace the older slower one. I'd consider moving mail to it's own server at some point, depending on the amount of traffic you have, but first getting a drive for just the mail spool. With only 1 drive, disk IO will most likely kill the system. What does this server handle and what hardware does it have? For a similarly setup server I ran: 1 - 540 IDE / swap /usr 1 - 540 IDE /tmp /var 1 - 1GB SCSI /u (/usr/home) /var/spool/mqueue 1 - 1GB SCSI /misc (local code, RADIUS, most logging and hardest hit) It should have had 2 more SCSI's sharing the logging and breaking things down a bit more, but this worked very well. The server did DNS, mail, shell, RADIUS, and full FTP (most sites were virtual FTP on web servers) on a P133 64MB. Preferably RADIUS and mail should have had their own server, since they were causing DNS to fallback to secondary. Most would also agree that SCSI should be used for all drives, but for infrequently accessed filesystems, IDE is a less expensive start if the bean counter is tight. >comments? Perhaps we could combine our experiences, and add a section to >the handbook, "Being a FreeBSD ISP" It would be very complex and probably should be the best of the methodologies presented. To me size and free space are always a 2nd to disk IO. Plenty of free space may be nice, but if it's slow... Jeff Mountin - Unix Systems TCP/IP networking mountin.man@mixcom.com
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?3.0.3.32.19980129132147.007041dc>