3CX Scalability / Hardware Requirements?

[RKMFL]

I thought this would be a common thread, so I searched thoroughly before posting, but no luck. I've found a few comments here-and-there (eg- how many concurrent calls can an Atom CPU handle) but nothing definitive.

Is there a link/article that discusses the scalability of 3CX on different hardware platforms? For example, using g711u (least overhead CPU-wise), all trunks internet terminated (no voice-card hardware), the total concurrent calls that can be handled under a recommended config (and what that is, for example, 3ghz dual-core, w/2gb)?

Or has anyone done any side-by-side's with 3CD v Asterisk v FreeSwitch, etc?

If I'm missed a previous thread/article, I apologize, but after searching quite thoroughly, I came up empty handed!

Thank you in advance!

 

[leejor]

Unfortunately a lot depends how things are set up. If you have all extensions set for "PBX delivers audio", then, of course there is going to be a much greater load put on the processor than just setting up the call then letting the devices "talk" to one another. I imagine that there must be a "breaking point" where the CPU is so overloaded that problems arise. But do you base it on the number of people on hold in a queue or simultaneous voice-mail users, or...? What criteria do you add to the equation until failure? In users each setup it will be different to an extent.

Perhaps someone on the forum has run some tests on various CPU's/disks/memory and can let us know, what straw "broke the camels back" in their case. If you were installing a system for business purposes, you would obviously put 3CX on a system that you would assume would be able to reasonably handle it, making note of any problems that cropped up. If they occurred, and it was because of a slow or under powered system, then you'd better be prepared to substitute a faster PC and fast.

Given the price and size of Atom based PC's/systems, I can certainly understand the popularity of them, I have a couple myself.
In fact one company is offering an Atom 1.1 Mhz PC the side of a couple decks of cards, for use as a VoIP PBX. But , of course, in some configurations, you will hit a limit but it will be different in each case..

 

[RKMFL]

Yes, exactly. I'm not looking for every scenario under the sun, just some standardized benchmarks.

Another way to run this, is to use what one predicts to be roughly 50% of capacity, and benchmark system resources at that rate.

For example, throw 100 concurrent g711u conversations (not IVR states), 10 IVR states, and 10 recordings (voicemail emulation). At this benchmark, approximately what are the system statistics among a P4 Quad-Core, with 2 GB, with everything else relatively generic/default.

For example, if this turns out to be 50% on all four cores, we can estimate that a dual-core is pushing the limits to run this range of volume (and apply that same logic to other resources/statistics).

It doesn't have to be anywhere close to perfect or exactly matching any specific specs -- any empirical data at all -- no matter what methodology, would simply be good information to have, for selling the 3CX solution to others, to defending against the Asterisk-religious, to building hardware specs, etc.

Thank you in advance!

 

[SY]

To be honest, I cannot understand why "transcoding mode" and "recording" are introduced as the killer of 3CX PBX performance...

 

[leejor]

Running what applications on 3CX causes the largest "drain" on CPU resources? The least?

 

[SY]

Running what applications on 3CX causes the largest "drain" on CPU resources? The least?

"Reliability and performance monitor" is the objective tool.

Each component does its own job and doesn't dublicate others. So it is very difficult to compare them because it is very dependent on "typical load", environment and configuration.

Sure, media server will do less work in case if transcoding service is not required but it doesn't mean that you can dramatically rise overall performance of the system just by making everything work in by-pass mode...

 

[RKMFL]

I want to be wrong, but it sounds as if 3CX doesn't want to give performance benchmarks of any sort. Is it that the product is that significantly more resource-intensive than Asterisk (this would be hard to believe as Asterisk is not known for scalability)?

Regarding the performance hit in transcoding ... What components of the system require the transcoding process? After all, if calls are all limited (incoming/outgoing) to the same codec, all prompts/recordings are done in native format (surely they wouldn't be recorded in one format and transcoded?), where will the transcoding come in?

 

[SY]

I want to be wrong, but it sounds as if 3CX doesn't want to give performance benchmarks of any sort. Is it that the product is that significantly more resource-intensive than Asterisk (this would be hard to believe as Asterisk is not known for scalability)?

Regarding the performance hit in transcoding ... What components of the system require the transcoding process? After all, if calls are all limited (incoming/outgoing) to the same codec, all prompts/recordings are done in native format (surely they wouldn't be recorded in one format and transcoded?), where will the transcoding come in?

I cannot find any benchmarks for following systems:
1. Asterisk+FreeSwitch,Fonality, Trixbox
2. Other PBXes

Do you have any links where it is avalable?

Thanks
P.S. 3CX PBX uses transcoding service only in case if it is really required. The cost of this service (CPU usage) depends on codecs negotiated with call participants. G711<<GSM<g729

 

[RKMFL]

First, I should make it clear that I don't know anything about other PBX's (beyond Asterisk and/or SwitchVox, and FreeSwitch); however, for Asterisk and FreeSwitch, there are definitely benchmarks / performance recommendations available. I'm aware of this because we once quoted a job and had to supply empirical data which we could source back to the findings.

I'm sorry that I don't have them off the top of my head, but they are definitely out there. For example, I remember one that discussed a very simple rule-of-thumb for the number of g729 channels/CPU requirements (versus) g711/requirements. I remember the difference being quite radical. I believe it was something to the effect of (remember, this was a legacy Asterisk version) 2 Ghz (single core at the time) would reliably safely run 80 concurrent channels at conventional breakdowns, whereas it would only support roughly 30 at g729.

Again, that's just shooting from the hip -- I don't have the specifics off hand.

I've also read performance reports for FreeSwitch, discussing 4-fold increases over Asterisk, for example, running 200 ports of concurrent voice on a single-core system.

Again .. No one is trying to be difficult; I'm simply suggesting it would be good information to have, and it could be in any format you prefer; you don't have to conform to the system requirements of everyone, simply pick a recommended system and give people an idea of what that can safely handle -- it doesn't sound like a bad idea for business, unless I've missed something.

Also, your confirmation of the transcoding issue makes perfect sense; it really is more of an "exception" than the "rule" and if you were to issue some sort of performance recommendations, it would be very easy to leave that as a foot-note, or use a simple 0% v. 25% baseline (channels using transcoding versus conventional compatible codecs), etc -- I'm sure that would be plenty of info and very helpful to those considering 3CX.

Myself personally, I'm not a Linux admin and would prefer to switch from a Linux-based PBX to 3CX; however, I generally like to have some empirical data on resource allocation before I make the decision to invest time in investigating a new solution. I could be different than most, but I can't imagine that catering towards your prospects who seek this information -- would be a bad business decision on your part.

 

[nlandas]

First, I should make it clear that I don't know anything about other PBX's (beyond Asterisk and/or SwitchVox, and FreeSwitch); however, for Asterisk and FreeSwitch, there are definitely benchmarks / performance recommendations available. I'm aware of this because we once quoted a job and had to supply empirical data which we could source back to the findings.

Again .. No one is trying to be difficult; I'm simply suggesting it would be good information to have, and it could be in any format you prefer; you don't have to conform to the system requirements of everyone, simply pick a recommended system and give people an idea of what that can safely handle -- it doesn't sound like a bad idea for business, unless I've missed something.

Myself personally, I'm not a Linux admin and would prefer to switch from a Linux-based PBX to 3CX; however, I generally like to have some empirical data on resource allocation before I make the decision to invest time in investigating a new solution. I could be different than most, but I can't imagine that catering towards your prospects who seek this information -- would be a bad business decision on your part.

This seems like a completely reasonable request. I too was looking for some form of benchmarking to see if 3CX would scale to my environment before I take the next step in the investigative process. We are a mid-size business with two location about 132 phones at our one location and 75 phones at the other. If we purchase new SIP phones from say Polycom and run them Polycom HD Voice technology, including support of G.722 wideband codec what kind of hardware would I be looking at for each location.

Obviously, what other things we do would require more resources but even a base system requirements for X number of phones running at G.XXX or G.XXX would be a great starting point.

We don't plan on recording converstations but would be using voicemail. So as a new potential customer and someone new to VoIp but not networking, how do I know what hardware to specify for a project?

 

[RKMFL]

@nlandas,
That's exactly my feeling. I can't invest the time in testing ever new product out there, until I've gone through the specs and made sure it makes sense to do so.

In this case, I still haven't tested 3CX for that reason. Further, I have to be able to defend myself when the Linux guys come in and bash me for selecting a Windows product in a market that is dominated by Linux.

This is simply the reality in many of our roles as CTO's, etc. Benchmarks are, to many in our field, more important than all of the sale collateral you can assemble. Now, if you want to focus on SOHO, it's not going to kill you to overlook this. But anyone selling software to mid-market businesses, simply won't be as inclined within very good empirical specs, IMHO.

I'm trying to help 3CX for than simply satisfy my due diligence ... yet, I get the feeling they aren't comprehending this.

 

[SY]

CTO should know that the main problem of banchmarks is the specification of environment and methodology.
Could you please provide any hints how to do it to get _real_ results?
Thanks