[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: SCSI Disk/Controller advice please - fun

hi ya 

On Sat, 30 Oct 2004, Joao Clemente wrote:

> ?!? What if a drive fails while on those 12h/day where people are 
> actually using them? This will be a fileserver where documents are 
> constantly changed/added/removed during the work hours!

have 2 file servers ...

- your fans is more likely to die than the disk
- the  ethernet cable will be wiggled loose while you're on vacation
 	- when a fan dies.. you're dead ... everything will die too
	- when a power supply fan dies, its a matter of months before your
	cpu, memory, disks, power spply( days ) to all start to die
	( its fun to see people wonder why their system dies when the ps
	( fan is dead .. 

	always put 3-4 extra fans in the regular midtower box 

- lots of ways for things to fail ... disks being the least of the
  problems if you bought good stuff from known-good reliable stores

> Well, I never used it before, I tought it was somewhat smarter... like 
> "mirror UNTIL a drive goes bad. At that point stop mirroring, ignore 
> failed drive, alert someone"

data is written to disk1 ... and mirror to disk2/..

but if disk2's disk sector arrives under the head first, it is written
first and mirrored to disk1 later when disk1 is not bz

- it's a 2 way mirror
	- how does the systme know that the file is bad/corrupted, vs
	the file you did mean to erase on both disks
	( tricky stuff ... though it can be done when one is careful )

> I see your point. Very good point!

sw monitoring is trivial .. monitor anything and everything till you're 
blue and tired of monitoring it 
> How diferent is a "every minute" cronjob rsync'ing content in both 
> drives to RAID1 (regarding the problem you stated above): "bad info" 
> gets sync'ed to the good disk anyway... doesn't it?

manually doing 2 way mirroring is too whacky ... and non-trivial

murphy's law says your primary disk will die ... if you do 1-way mirror
> A drive failure may lock the whole disk array?

no ... it's not supposed to

when you first build the raid system ..

	- pull each disk out one at a time while its rwritting a 2GB file
	and see if survives and does finish mirroring the file its
	supposed to be saving

	- you'd be doing these test to simulate disk failures
	and if you spent too much time to fix it... the others in the
	office will say why bother with raid, it didn't work
	and hopefully, the bean counters willg ive you another $100 to get
	a new disk to put on a different pc to backup and save to a 2nd
	disk that probably will not fail for the same reason that the
	other disk died
		- disks doesn't die nicely, it dies abruptly and randomly

	- there's lots of under-utilized pc in the office to save data

c ya

Reply to: