DISCLAIMER: These suggestions are by no means acceptable if you are looking for quality/stability, but if you are looking for a cheap/cheerful way to get dialer traffic out the door, then the above should give you some ideas. I've written this response on the basis that quality/stability is not a goal.

Throwing everything onto a single instance of FS is ultimately going to cause bottlenecks anyway (for example, the throughput for IVRs/voicemal would not be the same as the throughput for a simple user bridge) and no amount of hardware would fix that.

You could potentially run multiple FS instances on a beefy box then use resource separation with LXC (you're less likely to run into performance/quality/jitter issues in comparison with virutalization with ESXi)

You'd also probably want to take a look at the Sangoma DSP cards such as;

http://www.sangoma.com/products/d100-30-400-sessions/

For hardware, you'd want something like 2x or 4x SSDs in RAID 1 using a decent hardware RAID card (LSI Logic is good), and make sure it's a proper RAID card (e.g. PERC H800) and not these crappy soft cards (e.g. PERC S100). You could throw a dual quad core at it and put a single CPU in to begin with to save on some cost until the point where you need to expand. Maybe 32-64GB RAM, which is pretty cheap these days.

You could use consumer grade equipment and save on some cost. The RAM would be cheaper as it's non ECC/FB, and the majority of components would be cheaper also, but you'd most likely have to put it in a 4U ATX which could become costly if you plan on co-location. But be careful, this can result in corrupt data being written to disk if the memory goes bad.

In terms of throughput, this has been heavily discussed already and there are many threads and examples of this on the wiki - use some google-fu and you'll find them.

The only time you'd need to run multiple profiles in order to boost performance would be if you are processing thousands of calls per second (because sofia, the SIP parsing library, is single threaded), but as core devs have mentioned before you would have to be pushing some crazy numbers before this started to become a problem.

You'd most likely run into problems when you start running applications such as IVR/TTS/Voicemail, and your throughput will depends on exactly what you are doing. For example, the throughput for 100% of calls going into a voicemail app would not be the same throughput as 100% of calls going through a simple user bridge.

Hope this helps.

Cal

On Tue, Oct 15, 2013 at 11:18 PM, Andre <andretodd@verizon.net> wrote:

Assuming Money is no object and you wanted to get the most CPS/ Ports out of one server and you didn’t care that it’s stupid to put all those customers on one box. What type of hardware is recommended for running a carrier class traffic for media bypass for short duration dialer traffic?

I’m assuming we want to max out the hardware.

Processer, Hard drives, memory etc.? This would only be for the freeswitch server not the database.

Would you also have many instances of Freeswitch to use up the server or just more profiles or just one profile? Also, how many ports/cps do you think one massive beast of a server can handle? 10,000 CPS? More? Less?

I know this is a vague question but I’d like to hear what others think?

_________________________________________________________________________
Professional FreeSWITCH Consulting Services:
consulting@freeswitch.org
http://www.freeswitchsolutions.com

FreeSWITCH-powered IP PBX: The CudaTel Communication Server
http://www.cudatel.com

Official FreeSWITCH Sites
http://www.freeswitch.org
http://wiki.freeswitch.org
http://www.cluecon.com

FreeSWITCH-users mailing list
FreeSWITCH-users@lists.freeswitch.org
http://lists.freeswitch.org/mailman/listinfo/freeswitch-users
UNSUBSCRIBE:http://lists.freeswitch.org/mailman/options/freeswitch-users
http://www.freeswitch.org