Reliability of Network/Firewire/Usb Hard Disk

[Alessandro Pilati]

Hi to all,
I would be happy to know if could be a good choice to base the storage of a mission critical environment ( 2 servers in cluster, exactly 2 proliant DL380 or DL580 ) on NETWORK/Firewire/Usb hard disks, in terms of reliability, performances and easiness to recover.
Are this kind of hard disk as reliable as the normal hard disks?

Thank you,
Alessandro

[Matti_Kurkela]

I think you'd get more answers to this kind of question on Server forum than on HP-UX forum, but here goes...

Basically, the answer is greatly dependent on how much money you're going to invest. You probably can find acceptable solutions using all of the technologies you mentioned, if you search hard enough.

USB hard disks seem to be profiled more toward entry-level needs. Often their main selling point is that they are easy to carry with you and move your data to a different location. Of course, there are exceptions.
USB disk cannot be connected to two servers at the same time (a USB bus can have only one master), so clustering solutions without a single point of failure are impossible without special equipment (which may or may not exist). I think it's hard to find mission critical qualified solutions in this group.

Firewire is the one least familiar to me. It seems to be possible to use it in clustering, but I have no idea what kind of devices there are currently on the market.

Network disks have been a standard fare for a while, so you are sure to find proven solutions there. A "network disk" can mean a NAS (dedicated disk-server hardware on a standard LAN, accessible using a variety of protocols) or a SAN (typically a FibreChannel network that is dedicated for storage use and looks like a set of huge SCSI disks to the operating system).

NAS solutions can be found for all levels from a small "workgroup" NAS box to enterprise level. You can find ones with integrated or pre-designed backup solutions with competitive prices. Typically NAS solutions have some RAID setup with a hot spare as standard, so a failure of a single disk is either completely invisible or just causes some reduction of data throughput.

SAN is at its best in enterprise-level installations where the amount of data is measured in terabytes, or where reliability and fault tolerance are critical requirements. Here you're the most likely to find solutions where data in one storage unit is replicated in real time through a fiber-optic link to another storage unit in a different location.

[Prashant (I am Back)]

Hi,

First Step Check the compatibility of hdd(USB) with the server.
http://h18004.www1.hp.com/products/servers/platforms/usb-support.html

Regards,
Prashant S.

[Benoit TOURRET_1]

In case of a Microsoft Cluster (MSCS), the
disks must be SCSI (or FCAL) attached. they cannot be USB or FireWire.
The Entry price is the DL380 packaged Cluster (With MSA500 disk array)
all you need is an existing NT4 or 2000 Domain, and two Windows 2000 Advanced Server or two Windows 2003 Enterprise Edition.

a ATA / Sata disk have a very bad MTBF, compared with a SCSI Disk, specially if the SCSI disk is connected to a Raid Controller...

A Cluster is more dificult to administrate
than a simgle server.

A mission critical application on an usb/network or FW disk is IMHO a nonsense.

you can use a stand-by server and insert the
data disks (using Raid 1, 10, 5 or ADG)
from the failed server... (Same hardware is
mandatory)

99% of reliability is a standard PC
99,5% will be a server with redondant PS, Raid and redundant Nic
99,9 will be very expensive !!! and not so easy to maintain.

Regards,
Benoit.

[Florian Heigl (new acc) · Accepted solution]

They're definitely not as reliable, because they're built for a different work-cycle, better at withstanding hundreds of power-on/-off cycles and not meant to run 7x24x365.

For a mission critical environment there is no point in even discussing this.

i.e. I once ran my laptop for >6 weeks until I noticed the disk was failing (timeouts, strange sounds, etc.). I powercycled it about 15 times with no problems to follow - the disk is still working. It simply needed an opportunity to go to landzone and wipe it's heads. I felt with the poor disks and gave it it's break, but Your mission critical users won't be as understanding. :))

we've got two NetAPP R150 with IDE drives in them, but this is a calculated risk - these filers are a means of online d/r backup under service contract, and the disks are of course attached via scsi-ide bridges to a scsi hba, via redundant paths, in raid configuration and made available using dual fibrechannel hbas. oh, and before I forget it: this system is *not* mission-critical, but only the backup of a mission-critical system.

at least follow the microsoft guidelines for the cluster, so that it's really supported.

this at least means a scsi cluster connect, personally I'd advise You to use a low-end fibrechannel config for the avoided hassle.

[Alessandro Pilati]

Thank you Florian,
I'm trying to find a solution less expensive than 2 DL580/DL380 (RHEL 4, Oracle 10g) with an MSA1000 for our customers.

We need 2 servers in fault tolerance with Oracle ( no a lot of data, infact 2 shared disks are enough, space used about 10 Gb ), and not a lot of CPU calculation power...
And I think MSA1000 is too much for only 2 disks...

Oracle runs on 1 server, if it fails, the other server takes the resources itself and the services continues up... ( so no load balancing, because works 1 server at time ).

What do you think about this article?

http://www.oracle.com/technology/pub/articles/hunter_rac10g.html

Thanks a lot,
Alessandro

[Florian Heigl (new acc)]

I know that article very well ;)
But...
"_______Learn_____ how to set up and configure an Oracle RAC 10g Release 1 development cluster for less than US$1,800."


A low-cost cluster I once designed was based on a RA3000 controller unit (dual redundant, active-active) connected via two scsi paths to two adaptec 2944 controllers in each host.
(scsi bus resets disabled and no special cluster cabling, I left explicit instructions instead). the RA3000 was connected to a 7*4.3GB JBOD, with a second channel prepared for a low-cost high capacity Arena IDEtoSCSI array).
but that was 3 years ago, prices have dropped!

try to find a reseller for a eurologic (essantially the netapp FC9) fibrechannel shelf, 5 73GB (data) and two 18GB (quorum) drives and get two copper fibrechannel hba's.

this might not cost $1850, but won't be much more than $3000 ;)