Replacing PBX with Asterisk, need feedback on my new architecture.
I’m about to receive approval to design and deploy an Asterisk-based
phone system for my company. I will immediately have to start writing
specifications. I’m working on the hardware design and the architecture
right now. I’d like a second, third, fourth, 1,000th opinion.
800 SIP phones. All will be G.722. I expect 200 concurrent calls, with
20% leaving to the outside world. There will be another 200 analog lines
that will for the time being remain on the TDM PBX switch they reside
on, and will be whittled down and converted to SIP as time and attrition
allows. These are primarily fax machines and conference “spider” phones.
Those are included in my 200 concurrent calls number. I’m looking to get
as close to 5-9′s reliability as I can, with 4-9′s mandatory. Proper
power filtering and backup is already available.
Here’s what I’m thinking for the architecture:
Server 1: PRI Gateway 1 – Support 2 outside PRI trunks for local and
long distance, plus a third PRI connecting to the existing TDM PBX.
Server 2: PRI Gateway 2 – Support 1 PRI trunk for local and long
distance with room for another, plus a second PRI connecting to the
existing TDM PBX.
Reason for two PRI Gateways is for redundancy and fail-over, but
processor capabilities is a concern. I expect in about two years I’ll be
ready to decommission the TDM PBX, but will be left with about 80 Analog
lines across the multiple buildings on my campus. I expect I’ll end up
purchasing channel banks to support the remaining analog lines, and
distribute across the campus using existing copper plant.
Server 3: Asterisk Master Server
Server 4: Asterisk Slave Server
I’m considering a clustered environment, but I believe a fail-over
solution would be easier to implement in the short term. This means each
system needs to handle all traffic by itself. These servers will be used
for Asterisk and Voice-mail. Conferencing will be enabled, but I’m not
considering it in the build. If I see conferencing becoming a factor, I
will build another server and offload that service.
Server 5: Boot Server – DHCP, RADIUS, SNTP, DNS, LDAP, FTP, HTTPS, SNMP,
This service will provide the phone network all the basic services. This
is a stand-alone phone network primarily because it would be too costly
to upgrade the entire data network to support both voice and data. The
phone network will not initially have Internet Access. This server will
be the server all the phones talk to for pulling their configs.
I’m considering a second Boot Server for redundancy, but since the
phones should store their configs, I’m not seeing this as horribly
critical. Am I smoking something?
Finally, I’ll have a Windows-based workstation that will be used to
remote into all the services, for administration, etc…
I need to plan to use FreePBX on all Asterisk Servers, but I don’t
intend to install it until I’m in regular MAC maintenance mode.
I have no plans at this time to build out any databases. I just plan to
use whatever Asterisk has. If it ever comes to that, I would make those
separate servers as well.
My goal is to build Asterisk Servers and PRI Gateways capable of
supporting 150% of what I anticipate, which would come out to 300
concurrent calls. Again, all phones will use G.722. The PRI Gateway
servers will do the heavy lifting of converting G.711 traffic from the
PRIs to G722, and connect to the Asterisk Servers via IAX2 trunk.
It’s my intention to build each server myself with high-quality off the
shelf components. I’d like all servers to be as close to identical as
possible, as I intend to keep spares on hand to facilitate quick repair
and minimize downtime. I’m considering RAID 1 + 0 (mirrored and stripped
drives) for all servers. I am considering dual redundant power supplies.
For a processor, I’m currently looking at the i7-3770K @ 3.5GHz or very
similar. Its Passmark compares to the Xeon E5-2630 @ 2.3GHz, but is half
I have no idea what amount of memory to consider, so I am thinking 8GB
PCI-E is what I plan for all the cards.
Debian is the Linux flavor
A new network will be deployed using PoE layer-2 managed switches.
Battery backup capable of providing 8 hours will be installed as
required. There will be multiple VLANs in the network as I have multiple
dissimilar offices I need to keep separated from each other. We will
also have 802.11 SIP phones, and will be deploying a campus-wide WiFi
network used only by the phone system. Yes, I crunched the numbers. This
will be significantly cheaper than upgrading the entire existing data
network to support the new phone system. …and to be quite honest, I
don’t trust our network folks, and know adding that layer of bureaucracy
will only negatively impact the customer experience. I was a network
engineer for a top-three telecom company for many years, so I do have a
point of reference to make those statements.
…yes, I am one guy looking to do all this, with an estimated
completion date of the end of 2013. I’ll be building all this out in
addition to my normal “phone guy” job. I’ve built servers (hardware and
software) for 20+ years, but my Linux Kung Fu is weak. I’ll be learning
by doing and know there’ll be a lot of extra hours. The boss is good
about training, so I hope I can get into a good Linux Admin class in
addition to dCAP.
So tear it up! What do you think? Does the CPU have the oomph? What am I
missing? What am I overkilling? What would Brian Boitano do?
I appreciate any feedback, and thanks in advance.