[asterisk-users] Building the Perfect Box
Jay R. Ashworth
jra at baylink.com
Thu Sep 28 18:19:21 MST 2006
On Thu, Sep 28, 2006 at 08:44:55AM -0600, Colin Anderson wrote:
> >I concur with your approach, but "Tier 1" means as little here as it
> >does when evaluating Internet backbone carriers. could you expand on
> >what evaluation criteria you use? I'm going to be pre-speccing some
> >stuff myself this month...
>
> Sorry I should have been more clear. A good Asterisk install needs a
> holistic approach to use a hippy dippy phrase.
Yeah; I'm a Carlin fan, too.
> A Tier 1 server, which is a
> midrange to high end name brand server from the Big 3 (Dell, HP/Compaq, IBM,
> am I missing someone?) is usually highly optimized for bus bandwidth
> although that design was intended for a different use - usually massive disk
> I/O. As well, a Tier 1 server will have two seperate, independent PCI buses
> and this to me is a critical feature - it allows you to completely separate
> your TDM traffic from network, disk I/O etc. On my big production Netfinity,
> I took great care to ensure the Digium cards were all on their lonesome on a
> single bus, and everything else on the other bus. This is how I can run two
> TE110's in a single box with no problems. zttest does not give me 100% all
> the time, but on the other hand it *never* drops below 99.9987%, even under
> load.
Can I suggest that if the metric on zttest is 100=acceptable, 99.9987=
borderline, then someone needs to rewrite zttest? :-)
> I selected this Netfinity because of the obvious care put into it's
> design, but the specs are unimpressive: quad Xeon 700's. CPU is over rated
> for Asterisk, IMO unless you are doing tons of transcoding and if you are
> doing that, then your design is flawed.
Well, I think so. But I'm planning to have most of my audio on GigE
between media gateways, anyway.
> Anyway, the holistic approach (to go on a small rant for the newbie lurkers)
> be summed up as follows:
>
> 1. Good box, see above
To be a bit contrarian: big-name boxen aren't *always* the best choice
on engineering grounds, *because you can't always find out what they're
built with*... and sometimes it matters.
> 2. Good LAN - this is so critical and so often overlooked in the day and age
> of guys crimping their own cables and running $150 switches. You can't do
> that, and if you do, you do so at your own peril. Managed swiches,
> professional cable installation. This is not a problem for me since I *am* a
> professional cable installer but I have actually witnessed people making
> patch cables with a flat blade screwdriver and a hammer!
Hee.
> 3. Tuning of the LAN - VLAN's are good. QoS packets are good. Switches that
> honor the QoS packets are good.
Switches aren't *that* expensive; you would really VLAN the media with
the ops?
> 4. Handset selection - this is another biggie. I've selected Snom 360's, and
> yes they have warts, but they are feature rich for the price and Snom is
> really good about revising firmware. When you select handsets, GET YOUR
Oh, yeah.
Though, 2500's are hard to get wrong. ;-)
> USERS INVOLVED.
> 5. Tuning of Asterisk box itself - this cannot be under emphasized. This is
> a very important step and tuning methodologies vary according to distro,
> skill of the admin, and particular circumstances. I've learned *way* more
> than I ever wanted to about processor affinity sinc I started using
> Asterisk.
Write a book. :-)
> 6. Termination of PSTN. Basically I would never do an Asterisk install where
> I was forced to do something stupid like aggregate a dozen Centrex lines or
> some mickey mouse deal with FXO ATA's or whatever except for a hobby or
> prototype install. PRI, BRI, IAX or SIP, don't mess around with anything
> else.
Yep.
> 7. Relationship with provider. What is their SLA? Is it the incumbent or the
> clec? An incumbent will be more expensive and more difficult to deal with
> but they will tend to be more reliable. A clec will be cheaper and they will
> be way more accomodating but you will most likely not get five 9's from
> them. A VoIP provider should never be trusted, period. You will not get five
> nines from them, ever. Plan failover situations accordingly.
"SLA? What's a 'SLA'?" :-)
Amusingly, a client's * box went down this morning. I didn't get the
washout, but the mitigation wasn't well planned either -- everyone with
an Asterisk box should know what they're going to do if it falls over,
in detail. In a notebook. Just like when the nuclear missles start
going.
> 8. Plan plan plan plan. A good install of ANYTHING is 80% planning 20% doing
> it. What is your plan when your primary PSTN provider fails? What is your
> plan if your Asterisk box goes pear shaped? My dialplan can survive either
> PSTN, WAN or LAN failure (albeit with reduced functionality). I also keep a
> cold spare, an identically configured box that I can literally throw into
> the rack, turn it on, plug in the PRI's and no problem.
TXT! TXT!!
> 9. Internet bandwidth and latency. I am fortunate enough to have a great IP
> provider. Ask for demos - most guys will install a 90 day trial or something
> like that. Do not believe the brochure, get the product installed and put it
> under load.
Yep.
> 10. Traffic prioritization at the IP demarc - total no brainer.
Never hurts.
> 11. Constant, constant user feedback and remediation. If you are not talking
> to your users, your install will ultimately fail even if you have the best
> of everything. Underpromise and overdeliver. Never loose sight of the basics
> - they have to pick up the phone, and it has to work. Always.
I was musing on giving station users a list of pseudo-CLASS dialcodes
they could punch to mark that there was a problem with a previous call,
so it would go into the logs and could be checked latter.
Cheers,
-- jra
--
Jay R. Ashworth jra at baylink.com
Designer Baylink RFC 2100
Ashworth & Associates The Things I Think '87 e24
St Petersburg FL USA http://baylink.pitas.com +1 727 647 1274
"That's women for you; you divorce them, and 10 years later,
they stop having sex with you." -- Jennifer Crusie; _Fast_Women_
More information about the asterisk-users
mailing list