[NTLUG:Discuss] RAID-1 server goes down after disk failure/SWAP -- musings..

[Michael Barnes]

Richard Geoffrion told me on 11/21/2006 12:14:
> Before asking the group 'why', I did my obligatory google search and I 
> think I came up with a rather common sense answer!
> 
> The situation I have is that servers using software RAID-1 don't seem 
> very stable when one of the hard drive fails.   When a hard drive does 
> crash, the server locks up.  I think I have finally identified 'the 
> smoking gun'[tm].  See, since the beginning of my using software raid-1 
> in linux, I have created a separate swap partition on each drive.  My 
> thoughts were that there would be a speed increase along with space 
> savings if I split my swap partitions across two physical 
> drives/controllers.  The problem with this setup (as described in the 
> Software-Raid-HOWTO FAQ) is that if you swap to a drive then lose the 
> drive---well.. you're goin down!
> 
> Reference:  
> http://www.faqs.org/docs/Linux-HOWTO/Software-RAID-HOWTO.html#ss2.5
> 
> As I am currently rebuilding a few servers, I will need to do 
> 'something' about my swap partitions.  So... What is the current wisdom 
> concerning setting up Software Raid-1?   What are the prevailing 
> recommendations concerning swap files?   I've heard  everything from 
> SWAP=2 times the system's RAM to SWAP= 1/2 the system's RAM.  It seems 
> that swap will only use up to 2 gigabytes of a swap partition, so a 4gig 
> swap partition on a 2gig of RAM server is....wasteful.  In today's 
> multi-gig RAM climate, I can't seem to agree with the ever climbing SWAP 
> space configurations.  Back in the days, 128Meg of swap on a 64Meg 
> machine was plenty. Now that we have GIGs of RAM it would seem that the 
> increase of all this RAM should obsolete the need for swap altogether?
> 
> And what about using a FILE for swap space instead of a partition??
> 
> And what about using TMPFS?   What *IF* you used tmpfs for your /tmp 
> partition.  That would seem to reduces the amount of RAM available thus 
> creating the need for more SWAP space??  Can one configure tmpfs to use 
> swap space??   That would seem to be a cool way to have a totally 
> temporary file system.
> 
> What practical applications and/or pitfalls am I missing?
> 

Regarding swap partitions.
The advice I have been given in the past was
RAM < 128 then swap = 2xRAM
RAM > 256 then swap = RAM
RAM > 1 GB then no swap

Regarding RAID1.
In my experience, in a RAID1 system, when one drive fails, the entire 
machine becomes unstable and should be shutdown and repaired ASAP. In 
fact just this last Sunday we had a client machine lose a hard drive. 
It is a Dell dual processor server with a hardware RAID running Windows 
XP.  One drive croaked. The POST data said the RAID was 'degraded' and 
reported one drive as 'error occured'. The machine would not boot.  A 
few boot attempts and it started reporting corrupted/missing files and 
XP could not start.  Since it was under a maintenance contract, I just 
left it until the Dell guy showed up today with a new hard drive. 
Replaced the drive, fired it up, and all was well again.
So, the RAID1 crumping had nothing to do with swap space.

Michael