[Asterisk-Users] Hardware to build an Enterprise AsteriskUniversal Gateway

[Steven Critchfield]

On Sun, 2004-01-04 at 13:28, WipeOut wrote:
> Steven Critchfield wrote:
> 
> >On Sun, 2004-01-04 at 10:14, Doug Shubert wrote:
> >  
> >
> >>I would set the "Enterprise Class" bar at five 9's reliability
> >>(about 5.25 minutes per year of down time) the same
> >>as a Class 4/5 phone switch. This would require redundant
> >>design considerations in both hardware and software.
> >>
> >>In our network, Linux is approaching
> >>"Enterprise Class" and I don't see why *
> >>could not achieve this in the near future.
> >>    
> >>
> >
> >I may be wrong, but I think the 5 9's relates to full system not to
> >individual pieces especially when talking about a class4/5 switch. On a
> >small scale deployment, that will be a problem as you won't implement
> >full redundancy. Redundancy adds quite a bit to the cost of your
> >deployment. 
> >
> >As far as linux goes, it is at that level if you put forth the effort to
> >make it's environment decent. I have multiple machines approaching 2
> >years of uptime, and many over a year of uptime. I have not had a
> >machine in my colo space go down since we removed the one machine with a
> >buggy NIC.
> >
> >So next step, is asterisk. Outside of a couple of deadlocks from kernel
> >problems when I was compiling new modules, I haven't had asterisk knock
> >over while doing normal calls.
> >
> >The downtime could have been dealt with by having some redundancy in the
> >physical lines. I would have lost the calls on the line, but the calls
> >could be reconnected immediately. 
> >
> >I can say up front that I have asterisk installs running multiple months
> >without problems. 
> >  
> >
> Steven,
> 
> You often mention your servers uptime, I am assuming you don't count 
> reboots since you must have had to patch your kernel at least a few 
> times in the last year and the reboot would have reset your uptime..

Why do you assume I would have to patch a kernel? Not all machines must
run the most current kernels, and some kernels can be such that they are
sufficiently minimal enough to present low risk. Plus all the recent
problems require a local user to exploit. I subscribe to the theory to
only give access to critical machines to people I can quickly level a
shotgun to their head. With that knowledge, and my users acknowledgment
or witness to my accuracy, they don't wish to screw with the systems. 

BTW, my accuracy goes up with the number of concurrent targets by about
4 percent. 

> If that is the case then I have a server that is also around the 2 year 
> uptime mark.. The longest single runtime between reboots for updated 
> kernels is only 127 days.. :)

I have 2 machines at this moment that are halfway to looping the uptime
counter again at 497 days.

Webserver is at 497 + 197 days
Old almost decommissioned file server is at 497 + 194 days
A VPN machine is at 414 days
DB server is at 245 days
A almost decommissioned distro server is at 497 + 165 days


due to some upgrades, I now have fewer machines holding high uptimes. My
mail server was updated just over 2 months ago and it was swapped to the
distro server. So the distro server that is about to be decommissioned
is really just waiting for me to go take it out of the rack. 

Those are real uptimes with no reboots. What makes those 4 machines with
more than a year uptime interesting is that 1 is a dell, one is a
supermicro, the other 2 are homebuilt systems. So I can attest to x86
being able to be stable. Maybe not always, and I would like some more
swappable parts.
-- 
Steven Critchfield <critch at basesys.com>