Foos' 1.5TB server build thread

Discussion in 'Technology' started by Foosinho, Apr 28, 2006.

  1. Foosinho

    Foosinho New Member

    Jan 11, 1999
    New Albany, OH
    Club:
    Columbus Crew
    Nat'l Team:
    United States
    OK, I’ve already migrated some data over to the new machine–but not user accounts yet. I was trying to get the web working, and I’ve got some weird Apache configuration issue going on where I can access the server from inside the LAN, but when I flip my firewall over to forward incoming HTTP traffic to the new server, I get timeouts. Any Apache configuration wizards out there? (NB: Accessing by internal IP works. Accessing by external IP or by domain name does not.)

    The server is in the “rack” now. All the web docs are migrated, so once I get it working right it’ll just be a matter of flipping the switch on the firewall box. I’ve already moved the bittorrent tracker to the new box and recompiled it without any problems. I’m sure there will be some minor issues with getting all the web-based software (Gallery, Wordpress, etc) working again, but I’m not too worried about those.
     
  2. Foosinho

    Foosinho New Member

    Jan 11, 1999
    New Albany, OH
    Club:
    Columbus Crew
    Nat'l Team:
    United States
    Got the windows domain moved to the new machine. My linux home directory is mapped to the Z: drive, to provide access to all my documents and a roaming profile.

    [​IMG]
    [​IMG]

    Web and mail are the only things still to be migrated.
     
  3. Chicago1871

    Chicago1871 Member

    Apr 21, 2001
    Chicago
    Nat'l Team:
    United States
    What did you just call my mother?! :mad:
     
  4. Own Goal Hat-Trick

    Jul 28, 1999
    ColoRADo
    dude thats awesome to see 14gb take up such a tiny sliver.
     
  5. Foosinho

    Foosinho New Member

    Jan 11, 1999
    New Albany, OH
    Club:
    Columbus Crew
    Nat'l Team:
    United States
    I know. That made me smile. You know what else made me smile? Copying that 14GB took only a few minutes on my GigE lan. w00t!
     
  6. Foosinho

    Foosinho New Member

    Jan 11, 1999
    New Albany, OH
    Club:
    Columbus Crew
    Nat'l Team:
    United States
    OK, I've got the web server migration solved. I think the routing at the firewall is confused since michel-delving is technically on both ORANGE and GREEN. When I change the forwarding on the firewall to point at michel-delving's ORANGE IP rather than bywater, I cannot access web pages from any PC on GREEN, but it turns out I *can* from a computer outside my LAN.

    Thankfully, I can fix this by editing /etc/hosts on the firewall to use the internal IP for all the domains I use. The plus side is that I can actually access NamedVirtualHosts from GREEN and fix any software configuration issues before rolling the sites out to the "real world". Like get my database and PHP settings configured properly.

    I think having a server that is directly on ORANGE and GREEN simultaneously is a bit of a security risk, however I can't seem to find a way to both get the RAID array on the GREEN GigE network and the web/mail server capabilities available to the world without adding another computer. I'll have to play around with the routing rules on michel-delving to make sure things are set up correctly.
     
  7. Kryptonite

    Kryptonite Fancy Title Here

    Apr 10, 1999
    Columbus
    Club:
    Columbus Crew
    Nat'l Team:
    United States
    So, you're putting music files up there? Did I read FLAC? It sounds like you have the space for that, so i'd just go with that instead of a lossy format like MP3.
     
  8. Foosinho

    Foosinho New Member

    Jan 11, 1999
    New Albany, OH
    Club:
    Columbus Crew
    Nat'l Team:
    United States
    My mail configuration is here: http://genco.gen.tc/postfix_virtual.php

    Most of the software is available thru yum, with the exception of the courier packages, so I have to build those myself. Thankfully, I can compare back to my previous settings and build commands on my old box to make sure that any deviations I made from the script before are duplicated here.

    Then I'll just have to copy over the vmail user from the old computer and all the mailbox data, flip the pointer on the router, and we'll be in business.
     
  9. Foosinho

    Foosinho New Member

    Jan 11, 1999
    New Albany, OH
    Club:
    Columbus Crew
    Nat'l Team:
    United States
    Alright, so I'm not exactly sure what the problem was, but I finally managed to get Cyrus SASL to authenticate my virtual email addresses within Postfix, and got the Courier IMAP server running. So I'm now fully flipped over to the new server!

    I've downloaded the pictures I took of the build with my new camera, but for some reason I'm having problems opening the RAW files to convert them to JPEG. I'll get that sorted.

    But, on the plus side, this leg of the project is done. I'll probably take down the old server tomorrow night, and start cleaning it up to make room for the capture cards. You can follow along in the MythTV backend build thread.
     
  10. Foosinho

    Foosinho New Member

    Jan 11, 1999
    New Albany, OH
    Club:
    Columbus Crew
    Nat'l Team:
    United States
    Well, there seems to be a minor (or not-so-minor) problem that when I'm pushing a LOT of data, the network appears to die. All services on the box become unresponsive from the outside, and since it's running headless I can't access a command prompt to work on the problem.

    It seems like everything is still working, however, as I can hear the fans and it sounds like the drives are still spinning. But, without a head, the only fix is to reboot.

    So, as soon as I get the MythTV backend set up and running right, I'll yank the monitor, keyboard, and mouse and stick 'em back on the server so I can figure out this issue. I've looked in the logs and nothing has stuck out as being the obvious problem. At the very least with shell access I could pull down the network and push it back up to restart it.

    I *may* add this to the MythTV array as another tunerless backend to utilize "dead" clock cycles, if the rest of the MythTV setup can't keep up with commercial flagging and transcoding requirements. Right now my transient loads are usually well below 0.20, and the average load in the 0.01 range. I'll have to monitor that during multiple simultaneous recordings to see how well the processor keeps up with all that network and RAID traffic.
     
  11. yimmy

    yimmy Moderator

    Aug 23, 2004
    California
    Do you use copper or fibre for your GigE lan? I'm guessing copper. Does Cat-5 support GigE or did you have to use Cat6? What's the effective range?
     
  12. Foosinho

    Foosinho New Member

    Jan 11, 1999
    New Albany, OH
    Club:
    Columbus Crew
    Nat'l Team:
    United States
    Copper. Cat5e. Actually, I'm not sure of the effective range but I'm sure it's a lot longer than any typical house length run. In my case, all of my computers are currently in the basement (sans my laptop), so no run is more than about 30 ft of wire.
     
  13. Foosinho

    Foosinho New Member

    Jan 11, 1999
    New Albany, OH
    Club:
    Columbus Crew
    Nat'l Team:
    United States
    Well, unfortunately the worst possible scenario occurred - my system drive crashed. Ugh. So, I had to run out and buy a new 250GB drive, and am currently reinstalling the OS now. Thank heavens I made this thread - it's serving as a nice reminder of what I did right and wrong in the process before.
     
  14. Grouchy

    Grouchy Member+

    turkey bacon with swiss
    Apr 18, 1999
    Canal Winchester
    Club:
    Columbus Crew
    Nat'l Team:
    United States
    You don't have the install scripted yet? ;)
     
  15. Foosinho

    Foosinho New Member

    Jan 11, 1999
    New Albany, OH
    Club:
    Columbus Crew
    Nat'l Team:
    United States
    No. :(

    However, I actually probably could now. I've recovered the data in my raid array (which was actually probably the hardest step) after installing a new system drive and installing CentOS. Thank Loki that the RAID and LVM info is actually stored on the disks in question - it was just a matter of figuring out what commands to issue.

    So, while I can mount /home again, it's not doing everything in the right order - it's not actually loading the drivers until the end of the boot process (which is probably the root problem), which means /dev/md0 is not being creating, the LVM volumes are not being recognized, and the fstab mounts are failing.

    I think I'm going to print out all relevant configuration files and stick 'em in a binder for future reference. Of course, even once I get this boot order stuff licked, I've got to get all the software configurations running again, as my entire /etc directory was wiped clean. The mail configuration is going to be the worst, as I don't think I cleanly documented nearly enough of it before.

    Anybody know how to make a "restore" CD for linux? So if my root drive fails again, I can just restore an image from DVD that will get me right back to where I was when I burnt the DVD?
     
  16. Foosinho

    Foosinho New Member

    Jan 11, 1999
    New Albany, OH
    Club:
    Columbus Crew
    Nat'l Team:
    United States
    Got the RAID array up and mounted at boot time. Got the webserver back up - at least, the most important sites, as it will take a lot of time to rebuild the httpd.conf entries for all of the subdomains I run. Got the mail program *mostly* working - I can connect (TLS), but I can't get the authentication to work when sending yet.
     
  17. Grouchy

    Grouchy Member+

    turkey bacon with swiss
    Apr 18, 1999
    Canal Winchester
    Club:
    Columbus Crew
    Nat'l Team:
    United States
    Just curious, now that a few months have past, how have the results of this project been?
     
  18. Foosinho

    Foosinho New Member

    Jan 11, 1999
    New Albany, OH
    Club:
    Columbus Crew
    Nat'l Team:
    United States
    Code:
    [brian@michel-delving ~]$ df -H
    Filesystem             Size   Used  Avail Use% Mounted on
    /dev/hda2              246G   3.4G   230G   2% /
    /dev/hda1              1.1G    45M   963M   5% /boot
    none                   780M      0   780M   0% /dev/shm
    /dev/mapper/raid_disk-home
                           1.5T   751G   714G  52% /home
    /dev/mapper/raid_disk-usr--local
                            11G   3.8G   6.6G  37% /usr/local
    The good: 1.5TB home directory! My music is effortlessly shared between "My Music" on the desktop box and my MythTV server. Disk space is not even remotely a concern.*

    The bad: Don't put your swap partition on the RAID array. Just don't. It sounds like a good idea, but trust me - it's not. Under heavy load, when you need the RAM the most, you end up spending extra time computing parity bits when you most need to NOT be doing so. Extra RAM is your friend - I had to stick a 1GB stick in to fix the problem.

    The ugly: RAID5+LVM+CentOS+XFS = doubleplusungood. Frequent kernel panics. I was able to salvage things by moving to ext3 before the partition got too full, but that took a LOT of time.

    If I had it all to do over again, I'd make some changes. I'll outline those by talking about my next upgrade plan (at some undetermined point in the future).

    1) I like the redundency RAID5 provides for home directories and the like (the cost of mirroring is prohibitive for home use, IMO). However, for low-priority data (TV recordings) it's a waste. My next system will actually be 1GB drives, each shared via NFS individually (MythTV supports recording to multiple volumes, and will automatically load-balance during recording and will distribute files evenly across all recording volumes).

    2) Making the services box the same as the NAS box might have been a bad idea. Heavy loads are multiplied. The next system will be a FreeNAS box (FreeBSD, boots from CD and stores config to a floppy, allowing 100% of the hard drive space to be used for NAS).

    3) I've not had a drive failure, but I'm dreading one due to the difficulty of swapping out a drive. It won't even be easy to identify which drive is bad (unless it's making noises). The next system will also be rackmount, but it'll have hotswap drive bays to allow me to easily switch out failed drives, and expand storage.

    4) RAID5 also maximizes your power usage. Every drive is used in a write, and since on my system mail storage and the MySQL server data files are on the RAID array, none of the disks ever spin down. Efficiency is a big concern of mine going forward (it wasn't so much for this project), so that's one reason I'll be going with the highest-capacity drives I can get, and each drive will be stand-alone. That'll allow me to set the disks to spin down when idle and save some juice. Performance is already adequate under all but the most extreme situations, so low-speed drives will be preferred.

    5) My current box will in all liklihood stay in service as the main server, and the MythTV recordings / data will be moved to the new NAS (they are the big performance hogs).

    * Except with MythTV recordings. My next box will be significantly larger.
     
  19. Grouchy

    Grouchy Member+

    turkey bacon with swiss
    Apr 18, 1999
    Canal Winchester
    Club:
    Columbus Crew
    Nat'l Team:
    United States
    I had wondered because I recently came across a decent amount of storage, was researching hardware and software RAID 5 solutions and remembered this thread. For the home user RAID 5 does appears to be more trouble than it is worth and with storage getting cheaper and cheaper it almost makes sense just to mirror via hardware on the motherboard / adapter card or look into USB / eSATA attached storage then doing an occasional backup.

    I wouldn't put swap on a RAID 1 or 5 array and if possible, not even on the same physical disks and/or controller. Hardware RAID 0 for /tmp, RAID 1 for logs and maybe the OS, RAID 5 for data.
     
  20. Foosinho

    Foosinho New Member

    Jan 11, 1999
    New Albany, OH
    Club:
    Columbus Crew
    Nat'l Team:
    United States
    Concur completely. Like I said, putting the swap on a striped RAID arrangement (two RAID5 arrays striped in software) made some sense on it's surface, but the what I didn't consider was the fact that it's the SAME array as the data is on.

    For low-priority data, like TV, I don't think RAID is worth it at all. Single drives are large, fast, and cheap, and if a drive fails and I lose the data on it... TV is on again. I can always, eventually, re-record it. And it's just TV! If it's really "critical" (series finale of the Sopranos) I'll just grab it via bittorrent.
     
  21. Grouchy

    Grouchy Member+

    turkey bacon with swiss
    Apr 18, 1999
    Canal Winchester
    Club:
    Columbus Crew
    Nat'l Team:
    United States
    Or simply have a mirror of "critical" data on one or more external USB drives. Or for most current television, rent the DVDs from Netflix and re-encode.

    I'm probably going to skip the RAID 5 after experimenting with it a little bit. I already have USB enclosures for the good stuff if necessary. I was backing up movies and stuff to DVD-R; but now its cheaper to get hard disk and enclosures compared to DVD writers and DVD recordables; and much, much faster.
     
  22. yimmy

    yimmy Moderator

    Aug 23, 2004
    California
    So the striping for the Raid 5 array was done in software with some sort of logical volume manager? Are there any decent hardware solutions for RAID out there for the small home or office user? Like a Promise RAID controller perhaps? I always thought that that a hardware solution would be more efficient than a software solution :confused:
     
  23. Foosinho

    Foosinho New Member

    Jan 11, 1999
    New Albany, OH
    Club:
    Columbus Crew
    Nat'l Team:
    United States
    I have a Promise controller (I think I specify which earlier in the thread) which creates two RAID5 arrays "in hardware", each consisting of 4 250GB drives. Then, Linux does a software striped RAID to create one logical disk, called md0. Then, I use LVM on top of that, to allow dynamic partitioning.
     
  24. Grouchy

    Grouchy Member+

    turkey bacon with swiss
    Apr 18, 1999
    Canal Winchester
    Club:
    Columbus Crew
    Nat'l Team:
    United States
    So basically it's

    Hardware RAID 5: 250GB / 250GB / 250GB / 250GB (750GB)
    +
    Software RAID 0: 750GB / 750GB = 1.5 TB
    +
    Hardware RAID 5: 250GB / 250GB / 250GB / 250GB (750GB)

    One of the things I have found in discussions on the Internet is there is a growing debate over hardware vs. software RAID. Proponents of software RAID (5 for the sake of discussion) state that most modern processors and sub-systems are fast enough to handle typical home or small business RAID 5 applications (like file servers, etc.) and that the one good thing about software RAID 5 is that it is portable from hardware to hardware - if your controller dies - you need an identical or backward compatible controller to be able to recover your disk array. If that's true, that is a good point especially for home users. It's also cheap as in free. The downside, of course, is that all operations including the expensive parity calculations and verification of written data are done by the processor / system.

    There are currently 34 AMD compatible motherboards that have built-in RAID 5 via SATA (i.e. right on the motherboard) from a $40 FoxConn to a $250 ASUS; although with some budget controllers they still do the XOR parity process in the CPU.

    If you are going to do IDE RAID-5 every recommended to keep drives either all master (prefer) or all slave. I've seen a few recommend carving your array into 40-80GB slices then use LVM to attach them (to improve stability and re-sync time). Not sure if the latter helps or not.
     
  25. Foosinho

    Foosinho New Member

    Jan 11, 1999
    New Albany, OH
    Club:
    Columbus Crew
    Nat'l Team:
    United States
    That's exactly how I did it. All 8 drives are on a RocketRaid 454 controller (I'm pretty sure that was the model off the top of my head), and one RAID5 was constructed out of the "master" drives, and the other out of the "slave" drives (so that the loss of an IDE channel would NOT mean the loss of the entire array). Yeah, it's IDE.

    Like I said, it was a fun little experiment, but I'd change a lot about how I build my next system, especially given the relatively low value of recorded TV data. For home directories, I think that I've got enough storage to last me a LOOOOONG time with this system, and I will probably just mirror any future home directory storage drives.
     

Share This Page