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

Re: [cobalt-users] Re: Need to buy a new Raq



on 7/19/00 3:50 PM, Will DeHaan at will@xxxxxxxxxx wrote:

> Kris Dahl wrote:
> 
>> This is very true.  RAID is designed to save you in the case of a single
>> disk failure.  However, I am not sure if the software RAID employed on the
>> Raq4 will accomplish this or not.  Sure your *data* will still be there but
>> the box will be down, right?
> 
> Maybe on RedHat, but not on Cobalt Linux.
> 
> We've worked on the Linux IDE drivers to better tolerate drive
> failures.  If a drive fails, the server will continue to operate so long
> as the drive failure doesn't also break another component such as the
> power supply.  I've never heard of the latter case happening so I think
> that would be rather unlikely.

True that it would be unlikely that a failed hard drive would cause
additional problems.

And having a hot-swap is not as important on a RAID 1 as it is on a Raid 3.

When you are building data from parity, there is a definate performance hit.
Theoretically a mirrored system could run for quite some time.

However, I am curious: how does the duplication/replication process work?
Do you just down the server, replace the hard drive (are they labeled so you
know which to replace), turn it on and it replicates?  Or is there something
more than that?

>>> As a side note, since Cobalt uses standard IDE drives, the
>>> reliability on these drives is not too good - Mirroring the drives in
>>> them is a good idea...
> 
> The hardware for "server" SCSI and "workstation" IDE drives of the same
> spindle speeds, capacities and manufacturer are identical besides the
> circuit board controllers which rarely fail.  This is a common
> reliability misconception of IDE vs. SCSI.  I sure appreciate replacing
> $140 ATA drives over $500 SCSI's.

I personally enjoy the fact that SCSI doesn't hit the CPU as hard.

Plus it is much easier to expand--not an issue so much in this discussion,
but in general its important on server applications.

>> And this is not hardware RAID, nor is it hot-swappable.  I am firmly of the
>> opinion that if you have a single production server (that is not a load
>> balanced cluster) you need to have hot swap hardware RAID on the server.
>> 
>> -k
> 
> Hot swap is nice, and Cobalt understands the value of that feature even
> though it's not present on the RaQ 4.  The RaQ 4r will allow operation
> _with_ a failed drive allowing you to replace a disk during off-hours,
> at your leisure, off-line.  You'll get email notification of a drive
> failure and detailed RAID web status reporting.

Hey its better than nothing, don't get me wrong.  I have long be a proponent
of putting RAID in these things.  What I am having a hard time with is that
at the price you are essentially paying for hardware raid, just not getting
it.  I understand completely the cost advantages of software RAID and IDE,
etc.  I run Ultra ATA 66 at home and here at work on my workstations.  My
dev. server at home and the mac here run SCSI.  We're running IDE on the web
servers we just ordered (although we went with RAID 5 on the database
server).  THe thing is that it looks like that money is saved by *Cobal(t,
not the person buying the Raq.

Cobalt RaQ 4r
AMD K6-2 450
512MB DRAM (ecc?)
Dual 30GB RAID 1 HD IDE (7200RPM? Cache?
$4,799.00

VA Linux 2200
Pentium III 650 Dual Capable
512MB ECC SDRAM upgradable to 2 GB
3 18.2 GB 10K RPM UW SCSI Hard Drive
Hotswap RAID, 32GB Usable
~$5700
WIth 7200RPM drives
~$5000

Where is the cost savings of IDE and software RAID, the AMD chip, etc?

I honestly don't see how Chilisoft ASP (even at retail of $799) and the GUI
can possible add enough value to a much faster, hot swap, upgradable, etc.
machine from VA.  VA also isn't the bottom of the barrel cost-wise--they
tend to be a little bit pricy.

> While the RAID 1 configuration is not done in hardware, it is optimized
> and poses very, very little performance impact and in some cases a
> performance increase.

I'm sure it works well, and kudos for implementing additional availability
features.  I's just like to see some of those cost savings passed on.

-k