LVM, usb drives, Active Directory

[gaf-mNDKBlG2WHs at public.gmane.org (Jerry Feldman)]

Again, I fully agree. Both SCSI and FC raise the cost significantly. As
an example, I've got 6 computers all with SCSI slots. The maximum size
SCSI is 300GB at about $300/drive. In contrast eSATA drives are roughly
$100/1TB (prices are slightly higher for the faster drives). After a
close look at our systems, 4 of our systems actually have eSATA on the
MB, so our future solution would be to pull out the SCSI backplane and
simply wire the SATA drives to the 4 SATA ports. Note that the 6 x86_64
systems + 2 IA64 systems were free for being good HP and Intel partners.
But, I have a much different situation than the OP does. My SCSIs are
all LVM NFS exported over 1Gb Ethernet, but our products are all very
memory intensive so I have 3 of the systems maxed out at 64GB. Our issue
is having local servers rather than using servers in NY or Toronto.

The OP's proposed solution was for 10 1TB USB drives. Depending on some
specific requirements based on this discussion, he could do 10TB eSATA
very cheaply.  Assuming he already owns the USB drives, he could buy 1
or more low cost enclosures. and strictly use LVM or RAID0 to sripe the
drives or RAID5, 6, or 1+0. If he does not already own the drives, he
could possibly get 5 2TB drives + enclosure.

The basic underlying issue for the OP is NOT to use USB, but to use
eSATA instead. Again, the fault tolerance is going to cost more in terms
of more drive hardware for redundancy as well as possible dedicated RAID
enclosures.

On 12/16/2009 11:21 AM, Richard Pieri wrote:
> On Dec 16, 2009, at 7:42 AM, Jerry Feldman wrote:
>  =20
>> Basically, I think that Dave Ritter's point was that USB itself is too=

>> slow and unreliable.  I personally would not place a production drive =
on
>> USB.
>>    =20
> Neither would I[1].  The other side of fault tolerance is minimizing re=
covery times when there is a fault and USB flash utterly fails to deliver=
 on that.  So, the answer to 1 is definitely eSATA (or SCSI, or FC) and p=
referably with dedicated RAID controllers (possibly a non-issue depending=
 on the enclosure).
>
> The answer to 2 depends on what you are doing.  Specifically performanc=
e needs vs. fault tolerance.  If fault tolerance is at all necessary than=
 RAID0 is right out.  Barring cost concerns I would use RAID10 (1+0), and=
 RAID6 when cost is an issue and performance is not.
>
> [1] The exception being production data that is infrequently modified a=
nd needs to be stored off-line most of the time.  Data like cryptographic=
 keys used for code signings and configuration backups.  A thumb drive st=
ored in a firebox is a safe way to store your firewall and router configu=
ration backups.
>
>  =20


--=20
Jerry Feldman <gaf-mNDKBlG2WHs at public.gmane.org>
Boston Linux and Unix
PGP key id: 537C5846
PGP Key fingerprint: 3D1B 8377 A3C0 A5F2 ECBB  CA3B 4607 4319 537C 5846