[HCoop-Discuss] hardware advice

Adam Chlipala adamc at hcoop.net
Tue Jun 9 10:18:42 EDT 2009


Adam Megacz wrote:
> You should buy reliable hardware so that this doesn't happen very
> often.  When it does, the result will be several hours of downtime but
> (unless somebody drops the ball) less than a day.  And the downtime
> would only be experienced by those users whose data resided on the
> failed machine.
>   

You don't think it's worth replicating enough data that the back-up 
machine can take over for the main machine immediately, albeit probably 
starting from an old snapshot?

>> Use Xen virtualization on both machines, so that we can create many 
>> virtual servers and move them among machines as needed.
>>     
>
> Regarding v12n, I won't say I'm against it, but I think it creates a
> good bit of hassle that I'd rather not deal with.  The way I think of
> it, domtool is a v12n technology that works at the application layer
> (for each application domtool knows about).  I'm not sure that
> Xen/OpenVZ/whatever really adds much once you've got domtool.
>   

We're not considering virtualization in any sense where each user gets 
his own sandbox, so I'm not sure there's much of an analogue with what 
Domtool provides.  I think it's more about placing hard resource limits 
on particular shared daemons, which Domtool doesn't do at all.  It's 
also nice to contain the effects of bugs in daemons and to make it easy 
to move slices of our infrastructure across physical machines.

(And this discussion ignores Domtool bugs, which show up every now and 
then. ;])



More information about the HCoop-Discuss mailing list