Colleen
21-10-2004 10:18:13
So I got my file server up and running - yay!
Machine is my e250, which is a dual 250MHz UltraSPARC II with 512MB of RAM.
I put the 6 36GB SCSI drives that I wanted to put files on in a big software RAID 0 that I set up with mdadm. It was super-easy to get the RAID setup working.
I tried and tried and tried to get LVM working on top of the RAID - no such luck. I installed the LVM tools, patched the kernel & compiled in all LVM options that the patch made available. I could create a volume group with no problem, but when I tried to create logical volumes in that group it wouldn't let me. It said that my kernel didn't have support for the "striped" module (of course, I wasn't even trying to use striping in the logical volume, but it didn't care). I tried both kernel.org sources and debian-sparc sources, as well as upstream and debian packaged versions of the LVM patch, but still no luck. I suspect this is a weird sparc/2.4.27 thing since I've never encountered it before, but who knows. I finally decided to give up on it because I didn't think it would do much for me on this machine other than up the coolness factor.
Formatted it with reiserfs. The only real problem that I had with the reiserfs was that all the docs say that you can use a block size of 8192. I wanted to try it with both a block size of 4096, and a block size of 8192, to see if there was a performance difference. My motivation for that was that the memory page size on UltraSPARC is 8192, so I though that I might get a little performance increase by only having to fetch one block per page. So I set up the block size of 4096 first, and ran an iozone benchmark on it. After that finished (a couple of hours later) I attempted to do the same with the larger block size. You can indeed format your reiser partition with that block size, but you can't mount it on a machine running a 2.4 kernel. Apparently anything but 4096 causes kernel instability on 2.4 kernels (http://lkml.org/lkml/2002/10/31/69), so their solution was to patch the kernel to not allow the mount. So much for that idea!
I also wanted an FTP server on that machine, mainly to make stuff like the latest kernel source & linux isos easily fetchable by my other machines. At the recommendation of Despite and Chad, I went with vsftpd. It occurred to me that I didn't really need anything other than anon access, since I could just put the files on there with scp/sftp, so that made things rather simple and meant that I didn't have to fight with the PAM/LDAP stuff. vsftpd was really easy to configure, and has some nice configuration options like the ability to only accept certain passwords for anon access, etc. The other thing that I really appreciated was how easy it was to change the ftp user's home directory. I was a little worried that I was gonna have to get the source and give an option to ./configure for things to work right, but it just gets it out of /etc/passwd so that change was painless.
Other than the LVM stuff, my biggest struggle was getting the automounter working on the Linux clients. My only previous experience with the automounter had been on OS X, where you don't directly configure it, you use NetInfo Manager or niutil to give your mount info, and it deals with the automounter. Although I did RTFM, I didn't really understand that it created the mount point itself on the fly when you tried to access it. So I thought I had it all configured, but I kept ls'ing the directory that the mount point was supposed to be under but the mount point wasn't there. Finally I ls'd the invisible mount point, and there were my files - yay! After that I just created a symlink to the mount point to make life easier.
Future plans: Automount can get maps via LDAP, so I'm going to set up file shares for the other people on the file server, and then put their automount maps in LDAP. My goal is to get it where no matter which machine they're logged into, their files are there without me having to do much configuration. I think that's doable. I considered just having home dirs on the NFS server, but I don't think that's the greatest idea.
So that's my file server saga. Thanks for the ftpd recommendations!
Machine is my e250, which is a dual 250MHz UltraSPARC II with 512MB of RAM.
I put the 6 36GB SCSI drives that I wanted to put files on in a big software RAID 0 that I set up with mdadm. It was super-easy to get the RAID setup working.
I tried and tried and tried to get LVM working on top of the RAID - no such luck. I installed the LVM tools, patched the kernel & compiled in all LVM options that the patch made available. I could create a volume group with no problem, but when I tried to create logical volumes in that group it wouldn't let me. It said that my kernel didn't have support for the "striped" module (of course, I wasn't even trying to use striping in the logical volume, but it didn't care). I tried both kernel.org sources and debian-sparc sources, as well as upstream and debian packaged versions of the LVM patch, but still no luck. I suspect this is a weird sparc/2.4.27 thing since I've never encountered it before, but who knows. I finally decided to give up on it because I didn't think it would do much for me on this machine other than up the coolness factor.
Formatted it with reiserfs. The only real problem that I had with the reiserfs was that all the docs say that you can use a block size of 8192. I wanted to try it with both a block size of 4096, and a block size of 8192, to see if there was a performance difference. My motivation for that was that the memory page size on UltraSPARC is 8192, so I though that I might get a little performance increase by only having to fetch one block per page. So I set up the block size of 4096 first, and ran an iozone benchmark on it. After that finished (a couple of hours later) I attempted to do the same with the larger block size. You can indeed format your reiser partition with that block size, but you can't mount it on a machine running a 2.4 kernel. Apparently anything but 4096 causes kernel instability on 2.4 kernels (http://lkml.org/lkml/2002/10/31/69), so their solution was to patch the kernel to not allow the mount. So much for that idea!
I also wanted an FTP server on that machine, mainly to make stuff like the latest kernel source & linux isos easily fetchable by my other machines. At the recommendation of Despite and Chad, I went with vsftpd. It occurred to me that I didn't really need anything other than anon access, since I could just put the files on there with scp/sftp, so that made things rather simple and meant that I didn't have to fight with the PAM/LDAP stuff. vsftpd was really easy to configure, and has some nice configuration options like the ability to only accept certain passwords for anon access, etc. The other thing that I really appreciated was how easy it was to change the ftp user's home directory. I was a little worried that I was gonna have to get the source and give an option to ./configure for things to work right, but it just gets it out of /etc/passwd so that change was painless.
Other than the LVM stuff, my biggest struggle was getting the automounter working on the Linux clients. My only previous experience with the automounter had been on OS X, where you don't directly configure it, you use NetInfo Manager or niutil to give your mount info, and it deals with the automounter. Although I did RTFM, I didn't really understand that it created the mount point itself on the fly when you tried to access it. So I thought I had it all configured, but I kept ls'ing the directory that the mount point was supposed to be under but the mount point wasn't there. Finally I ls'd the invisible mount point, and there were my files - yay! After that I just created a symlink to the mount point to make life easier.
Future plans: Automount can get maps via LDAP, so I'm going to set up file shares for the other people on the file server, and then put their automount maps in LDAP. My goal is to get it where no matter which machine they're logged into, their files are there without me having to do much configuration. I think that's doable. I considered just having home dirs on the NFS server, but I don't think that's the greatest idea.
So that's my file server saga. Thanks for the ftpd recommendations!