On Wed, 30 Mar 2005, Digby Tarvin wrote:

> The oldest of these systems were installed
> with 5.25" Seagate 1 GB Wren drives, which have been running 24/7
> for over a decade, and 95% of them are either still running, or
> were retired after 7-8 years of flawless operation.

I have U160 SCSI Seagates in an old server that's been running 24x7
for just over 5 years (not as long as you!). Mind you, a good quality
"server-grade" mobo (this is an Intel ISP2150 platform) goes a long way
too. We still have older Intel P-II and P-III server boards running
reliably. This is what you get when you go for name brand (Intel, Sun,
etc) hardware.

Unfortunately, you do pay more for quality and reliability - usually
there's a mid-point where you get maximum bang for your buck without
sacrificing quality.

Another approach is to plan for constant failures. Last night I went to
hear a presentation given by a Google engineer which was quite amusing.

Google has thousands of servers all over the world. They use super cheap
PCs with consumer-grade drives and PSUs. Data is replicated across many
servers. They assume that everything will fail at some point and just plan
around it. If a server goes down, traffic is not directed to it until it
comes back up. The servers themselves are designed to have everything
easily replaced and therefore are very barebones. For example, PSUs held
in with velcro instead of screws (!) so its quick 'n' easy to rip out the
old and slap in a new one. In some data centers, a guy walks around
literally with huge cartloads of drives and PSUs :-)


-- 


--
[email protected] mailing list

Reply via email to