[Asterisk-Users] Re: load balancing 20 asterisk servers

[Miguel Ruiz Velasco Sobrino]

As being said, the cost of HW based solutions is in many cases too expensive to be
practical, leave alone to have a spare one to give you true high-availibility.
If you complain about DNS caching and timeouts not being respected, you can do a fairly
easy thing.

As said in the follow-up, you can drop in a data center many tiny boxes.
When one box stops working or crashes you can bring down it's interface and give a spare
box the same IP of the [now] defunct machine. If the box becomes irresponsive (a real OS
crash) and you have IMPI 2.0 capable MoBo's (all intel server boards have that, also some
other brands), you can remotely shutdown or reset the machine to avoid IP clashes. Intel
has a command line utility (also the graphical console) to manage that, i've using for a
while this and is absolutely wonderful.

Obiously, you will need a separate LAN with privates IP's to make much of the
administration and the DB access, and use the public LAN only for internet-related
things, so each box has it's own fixed private IP and only the public IP changes.

Indeed, with IMPI 2.0 is possible to remotely power-up a machine (if you are
enviromentally concerned... or if the datacenter metters the electricity you use), so you
don't need to be running all your spare servers waiting for a failure, maybe only one and
have the others shut-down until needed.


--- asterisk-users-request at lists.digium.com wrote:
Message: 5
Date: Wed, 2 Feb 2005 21:57:42 -0600 (CST)
From: Joe Greco <jgreco at ns.sol.net>
Subject: Re: [Asterisk-Users] Re: load balancing 20 asterisk servers
To: asterisk-users at lists.digium.com
Message-ID: <200502030357.j133vgmY050836 at aurora.sol.net>
Content-Type: text/plain; charset=us-ascii

> I'm trying to stay away from a software based load balancer cause what
> happens if that server fails?
> Its far less likely for a piece of dedicated hardware to fail than an actual
> computer.

You really ought to open up one of those pieces of dedicated hardware
sometime and see what's inside.

Yep, it's software based.

Heck, many of the so-called pieces of dedicated hardware are in fact nothing
more than a fancy rack mount PC.  Open up something like a CacheFlow server
and you find an Intel server motherboard, some propietary software, and that
is about it.  Heck, go on eBay and pick yourself up some of those nice F5
BigIP ... rack mount PC's.

Some of the newer stuff is software based with some ASIC assistance for
SSL/compression.  I know that F5 has made an effort to not look like a PC
anymore, for example, and has integrated some switchlike capabilities in
their product.

Still, when it comes right down to it, the traffic direction logic in these
things is software based.

Incidentally: one of the /down/sides to these devices, aside from being
hellishly expensive, is that when it blows at 5:01PM on a Thursday evening
when Friday is Christmas, even if you have the best service contract, it
can be a trying experience to get service.  PC's have the distinct
advantage that you can actually plan to have spare parts available, and
on top of it, you can actually build high quality redundant equipment
fairly inexpensively.

AIC RMC2N-XP Chassis	$150
EMACS R2G-6350P Power	$300
SuperMicro P4SC8	$300
Intel P4-3.0 Prescott	$175
Memory		as desired
CF Adapter		$ 20
1GB CompactFlash Boot	$ 60
			----
			$1005

Toss in a monster passive heatsink and you have a system that isn't
particularly susceptible to the loss of any single moving part.

Of course, you have to be able to sysadmin your way out of a cardboard
box, so it's not like it's cost-free, but here's the thing:

If my hypothetical load balancer fails at 5:01PM on Xmas eve, I can:

1) Grab the cold spare I built because it's cheaper to do two of these
   than a single expensive HW based solution

2) Configure the hot spare I built into production (again because it's
   cheaper).

3) Grab a desktop PC and stick a few Intel GigE NIC's in it and go to
   town.

4) At least have a reasonable chance of figuring out some other way to
   fix things temporarily.

So.

What's really interesting is that even some "networking hardware" is
actually just computing gear on steroids.  I recently saw a SMC 8624T
24-port gigE switch, and it appears to be a bunch of Broadcom GigE chips
with a CPU that runs some (can't recall which) embedded OS.  VxWorks?

... JG


=====
Miguel Ruiz Velasco

Version: OpenKeyServer v1.2
Comment: Extracted from belgium.keyserver.net
Signature: 0x59831109


		
__________________________________ 
Do you Yahoo!? 
Meet the all-new My Yahoo! - Try it today! 
http://my.yahoo.com