On Mon, 2006-06-05 at 08:00 +0300, Marc A. Volovic wrote:
> Quoth Shachar Shemesh:
> 
> > Marc A. Volovic wrote:
> > > 2. in case of drive failure, recovery process is a pain
> > >   
> > Well, doing "sfdisk -l /dev/sda > partitions" in advance, and then doing
> > "sfdisk /dev/sdb < partitions" isn't all that hard, really.
> 
> Which - especially in the case of complex raid volumes and doubly
> especially when also running LVM - makes for managing a whole pile of small
> and very important files which tend to *pooof* when the need is greatest.

Create a single partition, mark it as FB, turn on GUID in mkraid, and
you're done.

I've switched controllers, mixed drive types (SCSI 68/80, IDE/SATA),
disconnected IDE drivers by mistake (IDE doe not support hot-plug),
killed the kernel, killed the md drivers... I've managed to screw it
all, and never saw an MD array die.

On the other hand, I've had -very- bad experience with older 3ware
(SATA) and Adaptec (SCSI) raid controller.

> 
> > > 3. in case of device move, reintegration of volume is also a pain and may
> > >    lead to data loss
> > >   
> > Not if you configure MD properly. Properly configured, it finds the
> > partitions based on GUID, which means that a move is a no-brainer.
> 
> I have yet to see a case when GUID-based mounting helps rather than
> hinders. Especially - again, on post-failure - when re-integrating two
> devices with two.... errr... I mixed this up with disk-labels. Doh. Ok,
> GUID is possible. But same counter as above - managing GUID labels in a
> crisis is error-prone and each mistake increases crisis.

Why should -I- manage the GUID by hand?

> 
> > > 4. lower performance than any hardware raid
> > >   
> > ANY hardware raid?
> > You obviously have not seen some of the shitty stuff that floats around.
> 
> Yes - ANY hardware raid. and I do NOT mean that crap BIOS-based raids. I
> mean normal raids - Mylex, Vortex, Raidcore, LSI, etc.

A year ago I compared 3ware 9500 with 6 250GB drives to MD5 raid using
two el-cheapo  SIL3114 (?) SATA controllers.
In most cases, non-static benchmarking (using the in-house data
streaming application) showed the 3ware to be ~5-15% faster. In several
tests it was actually slower.
Upgrading the machine to a faster dual Opteron (instead of the original
dual Xeon machine) seem to indicate that the software RAID is scaling
better then the hardware one.

> 
> > A RAID controller that will have almost not buffers.
> > A controller that will restart resync in the middle if the drive is
> > being accessed too much.
> 
> That is flaky or borken (sic) hardware.

You call 3ware 8500/9500  flaky?

..

I do... but that's me ;)

> 
> > better performer than MD. If you take a GOOD raid controller, MD will
> > have poorer performance, but then it's really a question of budget,
> > isn't it?
> 
> A raid controller (4 ports, SATA-I) will cost some US$350. Hardly a budget
> breaker and well worth it.
> 
> > http://oss.metaparadigm.com/safte-monitor/
> 
> Ooooh - never saw that. Nice.
> 
> > I do believe you are either biased or always buying from someone else's
> > pocket. Either way, all the above do not relate to my situation.
> 
> I am indeed biased and I am not buying from someone elses pocket in all
> cases (in many, but not all)... For my personal and professional use, I buy
> from my pocket. It is too expensive to buy poor shit or rely on
> labour-intensive stuff - my labour (and, in fact, almost everyone's, too)
> is too expensive to expend on coddling weird stuff. I do not do system
> admin as a hobby, only as a necessary evil.

In FC/RHEL you can build the MD array (including the required partition
label) from within Anaconda. 
I doubt that LSI MegaRAID BIOS is easier to operate.

> 
> > On the other hand, it also has some nice things about it, the most
> > obvious being that you can use different partitions on the disk at
> > different RAID levels.
> 
> Reminds you of my point above on managing little clitical files, no?
> Surely, you cavil...
> 
> > As I stated above, it's all a question of budget and trade offs.
> 
> Assume a SATA RAID controller costs US$400. Assume you cost US$80/h (I am
> being a cheap bugger). Assume life-cycle at 36 months. The math is NOT
> complex.
> 
> 

My old workspace ate bundles of crap from 3ware... I wouldn't touch a
3ware controller if it was free.
Last time I tested LSI's SATA MegaRAID it was dog slow and don't get me
started about Adaptec's SATA RAID controller.

Gilboa


=================================================================
To unsubscribe, send mail to [EMAIL PROTECTED] with
the word "unsubscribe" in the message body, e.g., run the command
echo unsubscribe | mail [EMAIL PROTECTED]

Reply via email to