[YLUG] Server purchase

[Richard G. Clegg]

Arthur Clune wrote:
> 
> On 19 May 2006, at 11:14, Richard G. Clegg wrote:
> 
>> 1) Some kind and lovely person from a different dept is going to  buy 
>> us a shiny new server and I've no real idea what I am looking for.
> 
> 
> We use the new Sun Opteron servers. A single opteron 1U box is very  
> cheap (about 6-700 quid). Goes like stink as well.

Excellent -- that's the kind of numbers I like to hear.

>>  I've had a brief scout and the hardware RAID I could see involved  
>> stupidly expensive SCSI disks -- is this unavoidable?
> 
> You could go SATA instead.

Great, a couple of people have said this now.  The SCSI was the bit that 
was making the price shoot up.

>> 2) The idea at the moment seems to be to have two machines, a main  
>> machine and a backup and to use a virtual IP address, IP chains and  
>> heartbeat to get the machines to switch over on failure.
> 
> UltraMonkey is what you want for this bit of it. http:// 
> www.ultramonkey.org

I'll have a dig around -- thanks.

> Thing is, what are you covering yourself against here? You have two  
> different ways of ensuring redundancy and seem to want both.
> 
> Option 1) Reliable servers: mirror'd raid, dual psu etc etc.
> Option 2) Shonky servers but load balanced.

I don't think we want (2) really.  IMHO load-balancing is only necessary 
if we have a performance issue which we don't yet and it just adds 
another thing to go wrong.  My plan is to have a single server as 
reliable as can be bought which fails onto a very cheap and unreliable 
server designed to cope for the two or three days it takes to get a 
repair guy out.  We know that the cheap and unreliable server can handle 
everything because it has been doing for two years (apart from hardware 
failures recently).  Having been bitten three times with unexpected 
hardware failures (one of which would not have been helped by RAID or 
UPS) I do want the cheap backup machine there.

> There's another option; put the db on a single reliable server class  
> machine and run the web stuff on some shonky hardware but load  balance 
> it (this assumes your app will work in this setup of course).  

I'm not sure this adds anything we need does it?

> For dr 
> purposes you should replicate the db onto something else, but  that can 
> be cheap as well since it'd only be used in extremis.

We already have a daily backup onto something cheap somewhere else so 
that part is taken care of.

> UPS for the db server + one of the web servers and you'd be good to  go. 
> It wouldn't be totally 100%, but it'd cover most eventualities  (the 
> main failure mode would be a power cut if you were running on  the 
> non-ups'd web server only). Adding a bit more ups would solve  that, and 
> it's something that could be done either now or later.

I think we'd just have to accept downtime in the event of a powercut -- 
it's likely the routers in the wiring cupboard would also be down anyway.

-- 
Richard G. Clegg,
Networks & NonLinear Dynamics Group,
Dept. of Maths, Uni. of York.
http://www.richardclegg.org/