|
On Wednesday, January 10th, Joseph Mac NA3T wrote:
> On Tue, 9 Jan 2007, Dan Brown wrote:
> >>
> >>> I am looking at setting up a geographically separated set of
> >>> directors to perform failover with a pair of routers.
>
> the usual reason for geographically dispersed servers is for
> lower latency for the local users (eg google has separate
> servers for local areas/countries).
>
> You handle failures that must be handled quickly (<1min eg
> disks) locally and failures that can be recovered in 24hrs
> (eg backup/replication) off-site. Nowadays with RAID etc,
> gear doesn't fail a whole lot (except fans which fail about
> as often as disks here). Planned maintenance seems to be the
> main problem where I am - a bureaucrat insists that security
> patches be applied the instant they're released and the
> machine rebooted or else we'll be compromised that very day.
> Otherwise machines (and nowadays networks) just seem to stay
> up forever.
Well, the bureaucrats here seem to have this nagging doubt about dumpster
fires. There was a hosting company in town who had four separate ISP lines
all coming in off the same poll behind their building. One day the dumpster
behind the building, right beside the poll, caught fire (cigarette probably)
and took out all four connections at once and subsequently they were down
for about two weeks before they regained connectivity. During that time
they lost 80%-90% of their clients and went out of business shortly
thereafter.
Failover, as far as I am concerned, could have a lag as long as five minutes
although this would probably not be acceptable in the middle of the day.
Our local LVS takes over a minute from failover reaction to the takeover of
every last IP. We create/run websites, they aren't critical and they don't
save lives. I'd prefer to have colocation within the same city but as it
happens the nearest colocation facility (provided by the ISP itself) happens
to be 650km (+400mi.) away.
-------------
Dan Brown
danb@xxxxxx
|
