Skip to main content

Setting up MTE for the first time

Posted by cbbs70a on Fri, 01/23/2009

All;
I've been asked to come up with a multi-tenant solution for a customer who wants to do hosting for his customers. I need to come up with a solution that starts with 50 customers but will grow to 500.
Of course, uptime is a priority. Are people doing a single host with manual failover to a backup, or are they doing some sort of clustering solution from the start? Or are people using some sort of layer 4 switch that will redirect to a secondary host if the primary goes down? What about hardware specs for the computers? Single CPU, multiple CPU, RAID, etc? I most certainly appreciate any insight at all.
Thanks
Frank


Submitted by eeman on Fri, 01/23/2009 Permalink

many ways to skin that cat..

the simplest would be 1 MTE box active doing nightly backups of the important stuff + 1 warm spare that its provisioned enough that a restore of the backup tar's and sql and ip change would return to service. As tenants grew and load increased you could bring about another MTE box into production for the next batch of customers putting you at 2 MTE box + 1 warm spare and rinse/repeat to scale as needed. Your customer is still going to want to install a 'gateway' or 'routing' asterisk server to handle your PSTN interfaces, ensure numbers you own remain on your voip network and dont traverse the PSTN, and maintain cleaner, 1line per call, CDR entries for billing software.

you could also create 2 boxes in active/passive arrangement using DRBD to replicate partitions and DRBDlinks to dynamically create symlinks to the syncronized partition and use Heartbeat to determine if failover needs to occur.

layer4 or layer7 switches tend to have issues with SIP due to phones seeing data coming from a different IP than they sent data to. Sometimes this can be overcome in sip.conf to overwrite the URI using externip= . Given that you couldn't use this in some active/active scenario, you'd still have to use DRBD+heartbeat on the MTE boxes. thus the layer4 switch is unnecessary as Heartbeat can be told to issue a arp ping to register the new MAC to its gateway.

Submitted by cbbs70a on Sat, 01/24/2009 Permalink

Erik;

This is good stuff you gave me, it really is. Let me ask you 2 questions. I know that the numbers are not exact, but ballpark, what are we talking about as far as users per Asterisk server? It's probably better to talk about concurrent calls per server rather than users per server. Secondly, at what point do I need to add an OpenSER box? Thanks for all your input with this.

Regards;

FSD

Submitted by eeman on Sat, 01/24/2009 Permalink

given sufficient resources, you should be able to reach, ballpark-wise, 1000 calls per server maybe more. There are a lot of things that affect load, such as codec, transcoding, conferencing, type of trunk connection etc.

Submitted by cbbs70a on Sat, 01/24/2009 Permalink

Are you using OpenSER? I know a couple hosted service providers who swear by it. To them, the sun rises and sets on it.

Thanks

FSD

Submitted by eeman on Sat, 01/24/2009 Permalink

I push the sip proxies all the way to the customer (having them use siproxd) so that i can seperate their network with vlan's for improved call quality.

Submitted by cbbs70a on Wed, 01/28/2009 Permalink

Has anyone every tried using a layer4+ switch in an active/active mode in a real-world scenario? Isn't one of the purposes of the switch to present a single IP address to the world without the need to ever show how many servers/IP addresses there really are. Plus, it seems that most of the better switches claim SIP compatability. I don't know if it works, just that most everyone claims it. Just a thought.

Submitted by eeman on Wed, 01/28/2009 Permalink

you cant active/active... imagine 50% of your call being delivered to server A and 50% delivered to serverB or imagine a tenant were phones register with different servers; so when you try to call Bob at extension 110, it went straight to voicemail because you are on ServerA and Bob was registered on ServerB.

your comparing something stateless like HTTP with something where not only do all _your_ packets need to consistently reach the same asterisk box, but everyone else in your office need to register on the same box so that other features like Call transfers, Parking, Presense, Busy Lamps, Voicemail and everything else can work the way its intended. Even where state matters in HTTP my browsing needs never require other browsers to also connect to the same backend server. How frustrated would your customers be if they could only access some of their voicemail messages because the remaining were stored somewhere else?

there is only 1 company performing an MTE type solution for that, in that event asterisk is not much more than an IVR and voicemail server leaving SER as the actual SIP engine. I've only seen deployments for residential services so I don't even know how more advanced features like busy lamps etc work. You don't buy his software, you lease it. It starts at 45,000 Euro per year. In his solution he had 2 SER servers in active/passive and a bunch of Asterisk boxes and NAS. You also had to buy your NAS and servers, the 45,000 Euro was just the licensing cost.