Re: [cobalt-users] How are people handling fault tolerence?

[Kris Dahl <krislists@xxxxxxxxxxxxx>]

on 6/1/00 7:02 PM, Chip at chip@xxxxxxxxxx wrote:

> I will soon be responsible for administering a Raq server that will host a
> bunch of customers' sites and email.  Currently, we are reselling but we're
> switching to dedicated.
> 
> For those of you that are hosting 100 sites or less, how do you handle
> redundancy and/or failure?  I see several possibilities:
> 
> 1. Keep good backups and hope the Raq doesn't go down, then hustle to get
> another one if it does. (Not a very good choice, IMHO, but a lot of folks
> seem to be doing this)

This is by far the most popular option.  I would highly recommend having
another hard drive ready to rock.

> 2. Install an IDE RAID-1 mirror in the Raq, realizing that a hard drive
> failure is probably the most likely cause of death.

I suppose you could do this, but I don't really think software RAID is the
best way to go, for performance and resilliance.

> 3.  Buy 2 Raq servers, mirror information between them and locate in same NOC
> 
> 4. Same as above, but locate them in different ISPs with different NOCs.

Then you are talking about purchasing some load balancing hardware that will
probably eclipse the cost of going with a heavier duty server that has the
reliability and availability options you require.

> Option 4 seems the most bulletproof, but also the most costly.  I'm very
> curious what others are doing, particularly given the horror stories I've
> been reading about the non-functional Raq backup solutions.

I don't think option 4 really is bullet proof at all.  How are you going to
laod balance between the two?  Round Robin DNS?  Round Robin DNS doesn't
always work out as well as you'd like, I have found.  Sure it can I guess
load balance fairly well, but what happens when a server goes down?
Obviosly you have to host DNS offsite, and you could take it out of the list
for that host, but then what do you do about the places that already have
the record cached?  And a lot of people frown upon low TTLs.

Additionally, How are you going to syncronize the two servers?  There will
be some pretty serious overhead here.

Option 5, which I feel is better than any of those:

Purchase a server with hardware RAID, and host it in a world class
datacenter with redundant routes & carriers.  It should be able to survive a
single disk failure, if not have a hot spare online.  Purchase a decent tape
drive and make a good data backup plan.

-k