Forum OpenACS Q&A: ide controller failure

Collapse
Posted by Jonathan Ellis on
The Linux Software RAID HOWTO says

It is very important, that you only use one IDE disk per IDE bus. Not only would two disks ruin the performance, but the failure of a disk often guarantees the failure of the bus, and therefore the failure of all disks on that bus.

Has anyone else heard of this happening? Hmm. I have three disks + 2 channels = bad news for my controller if this is true. Not that I'm planning on a drive failure but still... :)
Collapse
Posted by Don Baccus on
From a safety point of view, if you split each half of your RAID 1 mirror onto a separate channel you're probably OK.

Also, I don't know if the warning is relevant for modern ATA (UDMA) mode disks, or only for disks operating in PIO mode.  Note that in my experience Linux defaults to PIO, not DMA, mode because some early implementations of DMA are screwed up.  Shouldn't be a problem with modern disks, I run all mine in DMA mode (and Win98/Win2K/NT run ATA disks in DMA).  DMA mode's a lot faster if you're not already doing it,  higher transfer rate and far lower CPU consumption.

Actually, I don't know if the warning has any statisical basis in fact at all, or  if it is just based on opinion or anecdotal evidence!

Collapse
Posted by Chris Mauritz on
My understanding from being a member of the linux-raid list for eons is that if either the master or slave disk fails on a channel with 2 or more disks, you're almost certain to lose access to the entire channel. With SCSI it's a tossup, depending on the severity of the failure and the controller/firmware interactions.

What I would recommend is that folks that want IDE RAID use something like the 3WARE IDE hardware RAID controller. They're pretty cheap (~US$125 for 2 channels and ~$200 for 4 channels) and come in 2/4/8 port models...each having a dedicated channel for each disk. With a 2 channel card and a pair of cheap quantum 7200rpm 20.5gig disks striped as a RAID 0 device, I was able to get about 52mb/sec on reads and about 45mb/sec on writes using bonnie and 4X the physical RAM for the dataset. That's on a P3-733 with 256mb RAM and Redhat 6.2.

Cheers,

Chris