Server Update

[inaki]

Soo another day done on the server. The consistancy check completed, but the data volumes wouldn't mount. We're going to have to do some data recovery measures. For now, we're bringing the server back up on a new drive system. We're going to be using a Netapp drive shelf instead of an internal array. It provides much more security, better performance, support for multiple machines, and most importantly, e-mail and paging notifications of drive failures. x.x

This will also get past the space issues we had earlier. It'll eat a bit of extra rack space, but the storage scalability is well worth it. The system is much better organized now, as well. All in all, once done this'll be a pretty bitchin system.

Hey Julia! You up for recovering VMDKs off a 5 disk RAID5 volume that had two disk failures? I'm certianly open to discussions of compensation!

Comments

[thraxarious.livejournal.com]

This is why I don't recommend Raid 5 anymore. Mirroring is much easier to come back from in data recovery. I only recommend raid 5 if you plan on duplicating the entire storage system, hopefully in a different location.

Let me know if you want to see about getting tighter pricing on some RAID edition drives. desktop drives often can't deal with the heat and vibration of being stacked together tightly.

[inaki.livejournal.com]

Eh we're just putting a Netapp in and calling it done. =P

[tixen.livejournal.com]

Raid 5 is a perfectly respectable system- if you're running it on hardware that can notify you before catastrophe.

In the previous system, there were two drives dedicated for Parity. If a drive failed, they could pop another in and get everything back to normal, no problems. Two drives failed- one a while ago, and no one at the data center told them anything was wrong.

The NetApp will have two drives for parity, and two hot spares in a Raid 5. Add email notifications on drive failure, and you have a nearly failsafe system. :3

[inaki.livejournal.com]

The Netapp also has much more storage capacity, up to 3TB (after raid and spares) in one shelf, and more shelves can be added easilly. This was a big problem with the old array, since it needed very specific drives.

[higginsdragon.livejournal.com]

Google proved that heat has nothing to do with drive failure rates, as well as "enterprise" drives not being any more reliable than cheap drives.

[squnq.livejournal.com]

Well, at least it's an excuse to create a new library for failover.

[inaki.livejournal.com]

hah. True!