Hello.
We are trying to determine if we should migrate our customers from a Sliced Trixbox onto a MTE colocated at our SipHost. I would like to know, from people who have MTE running in a production environment, the following questions:
1) Do you utilize clusters or deploy MTE on a single server with contingency plans should there be an outage?
2) How many handsets per server do you typically have on each server? Concurrent calls on each server?
3) What about cloud based servers for scalability and redundancy?
4) Do you give your customers access to the tenant portion of the website or do you find that they screw things up more than self manage?
Thanks.
We a small but growing hosted PBX company and would appreciate any input you could give us.
Regards.
MTE in production
Hi
1)
We run several clusters, per cluster we have 2 identical servers, that run on a Syn3 High Availability distribution (http://www.syn-3.eu) If one machine fails, the second takes over in seconds. In addition to that we make hourly backups and we have a spare machine on which we can restore the backup on. This machine we also use for testing new config / releases.
2) We didn't ran into any limits so far, on an average we do a max of a 100 concurrent calls with 1000's of handsets on a cluster.
3) no experience on that
4) depends on the cluster and the partner / customer. We have dedicated clusters for dedicated resellers, they have more rights then individual endusers with a single ippbx. Besides the obvious functions you would limit (system management etc) we also limit the trunks (so all traffic will flow through our gateway's).
Kind Regards
1. I have a cold spare that i can restore backups from in very short order.
2. I am a consultant to many other MTE customers besides my own the largest was approx 4500 handsets and 160 concurrent calls. This really depends on you, if you engineer the customer side correctly or take shitty shortcuts that put a bigger burden on your asterisk mte server. The more burden you take on, the less scalability per server you will see. Theoretically you could have 10k handsets and 1000 concurrent calls. Do the bandwidth math on 1000 concurrent calls (2000 call legs) and you'll be dealing with bandwidth contracts long before you get to that threshold.
3) virtualization is not a good solution for PBXs. I know digium claims they work in the cloud. They never bothered to tell you that they only tested it with 15 - 20 concurrent calls and 50 handsets. The time inside a virtual machine is not parallel to the time outside. 1 second in a VM might happen 0.9997 seconds after the previous second, the next might happen 1.0012 seconds later but inside the VM they are evenly spaced 'seconds'. I tried running MTE inside Xen for a while. For the first 6 months I thought I had the issue solved. Then, as I took on more customers, and scaled up concurrent calls, I realized that I had to get off the VM. The bridged calls sounded fine, but playback of voicemail, the IVR greetings, the asterisk sound files, conference rooms, they all started to sound like shit. Anything that required a timing source was suffering bad audio.
4. I give them limited access. I dont let them alter outbound routes and things like that. I let them alter inbound routes, extensions etc. For the most part they actually rarely use the web portal.