website stat

Software RAID in GNU/Linux

Checkout my brief guidelines on getting you started with Software RAID in GNU/Linux [1].
Comments, suggestions, insults and sexual appraisals can be sent to the usual email.

[1] - No GUI, sorry. Plain good old CLI.


2 Responses to “Software RAID in GNU/Linux”

Feed for this Entry
  1. Carlos Rodrigues
    Published at June 8th, 2006 at 7:09 pm

    “Hardware RAID is a must for servers who have a high demand and need serious throughput. Not the case of my home server that’s acting as a fileserver!”

    If you have CPU cycles to spare (not that much), software RAID may actually be faster than hardware RAID… Having the OS know the layout of the drives may offer it a better chance to optimize I/O requests (stuff like the matching “stripe size” with ext3’s “stride” parameter). Well, with SCSI this may not be true, SCSI controllers are very smart.

    “I don’t know how the IDE controller works nor how the software RAID works so I couldn’t conclude if setting up both hard drives on the same IDE connector (1x Master, 1x Slave) would be faster than having separate IDE connectors (1x Master, 1x Mater). I picked the latter. I’ll wait for some insightful response to this question.”

    An IDE channel is a shared bus, so only one drive may be pumping/receiving data at any given time, and some commands lock the bus (try having a CD-Writer and a CD reader on the same channel, and opening and closing the tray on the reader while the writer is recording… instant buffer underrun). Plus, changing directions (write a “read” command/read data/write a “write” command/…) is expensive over IDE, so software RAID on a single IDE channel is very bad performance-wise.

    “To set up a partitition, run fdisk twice, each one for each disk”

    You can “fdisk” the first disk and then use “sfdisk” to copy the partition table onto the other disk, if both disks have the same geometry.

  2. mlopes
    Published at June 8th, 2006 at 9:01 pm

    Thanks Carlos, I’ve just updated the article :-)