| Author |
|
grcgrc
Intermediate
Joined: August 16 2006
Posts: 6
|
| Posted: August 16 2006 at 11:27am | IP Logged
|
|
|
Are there any performance benchmarks, or better yet, real deployment stories regarding how many simultaneous calls can be placed/handled by different hardware platforms? My target application is an IVR based survey system, which basically plays many recorded questions and accepts DTMF digits for answers. Open ended questions can also be recorded (voice).
In particular, I am interested in 3.2 GHz HT Xeons, Dual Xeons, Opterons, etc. Do things scale well onto two processors? Does everything run the same under AMD or Intel processors or is the code optimized for either platform?
|
| Back to Top |
|
| |
support
Administrator
Joined: January 26 2005
Location: United States
Posts: 1666
|
| Posted: August 16 2006 at 1:56pm | IP Logged
|
|
|
Hi Guillermo,
Your VOIP IVR application sounds cool. You asked many questions, so we will try to break them down individually:
Performance Benchmarks:
As far as media engine performance benchmarks, we do not publish these figures yet. This is a good question and we understand why you ask it. We are working to create some kind of “normalized” data for the media engine as far as call loading on different platforms.
The main reason we do not have this data for customers is because it’s a lower priority relative to getting customers new functionality and features into the product.
There is another post you might want to check out. It talks about our SIP/RTP proxy products but some of the content is also applicable to the VOIP Media Engine. See this URL: Deploying ISTP of 5000 users - Questions
Real Deployment Examples/Stories:
None at the moment. We have made a note of your suggestion for this information. We are glad you mention this. If enough customers request something, we generally set off to offer it to them.
Do things scale well onto two processors:
Yes. The threads of the media engine will run on whatever processor is least loaded and available. Here is an excerpt from the book “Inside Microsoft Windows 2000 – 3rd Ed.”:
Code:
Overview of Windows 2000 Scheduling
Windows 2000 implements a priority-driven, preemptive scheduling system where the
highest-priority runnable (ready) thread always runs, with the caveat that the
thread chosen to run might be limited by the processors on which the thread is
allowed to run, a phenomenon called processor affinity.
By default, threads can run on any available processor, but you can alter processor
affinity by using one of the Win32 scheduling functions.
|
|
|
So, the threads of the VOIP Media engine (and all other LanScape VOIP applications) will run on any available processor in the system. The media engine has been developed as a multi-threaded product that takes advantage of all available CPUs.
Intel or AMD processors:
This does not matter. Performance on any Intel or AMD core works sell. Internally, there are signal processing optimizations that will be used as a function of what underlying CPUs are available.
How many call can be processed and how fast:
The max number of phone calls that can be processed by the media engine is limited to the number of phone lines licensed. Currently we offer up to 64 line versions. We will be offering 128 and 512 line versions in a future release. Due to the design of the media engine and the max number of process threads, there is an operating system limitation on the max number of lines we can license. I forget what this high end number is at the moment.
For strict IVR based apps, the fast host machines you mentioned above should be able to handle 250 to 500 calls concurrent – maybe more. The speed at which calls will be answered by the media engine basically comes down to how fast the host machine can start a new internal “call thread”.
Good questions. Repost as needed,
Support
|
| Back to Top |
|
| |
grcgrc
Intermediate
Joined: August 16 2006
Posts: 6
|
| Posted: August 16 2006 at 2:43pm | IP Logged
|
|
|
I´d be interested in seeing those performance numbers as soon as possible.
I read the post you mentioned (Deploying ISTP of 5000 users - Questions), and although it talks about the media proxy product and not the engine, I supposed that the proxy is based on the engine. As far as the estimations made in that post, I think there are some calculation errors: I would REALLY love to be able to squeeze 4500 concurrent calls into a 1 mbps pipe!! Even 3100 would be nice... From everything I´ve read and all the VoIP bandwidth "calculators" available on the web, a G.729 call consumes about 30kbps per call, enabling about 35 calls on a 1mbps pipe. Is this correct? If so, can you please recalculate the previous post and re-post it?
As far as hardware is concerned, what server NICs have you found that work best/ lessen the load on the CPUs?
Also, how does the codec being used affect the CPU usage? I hear that although iLBC is great under bad connectivity conditions, it is about twice as expensive CPU wise as G.729.
In an IVR application like I mentioned, there is a lot of sound file playback. The sound files must be transcoded from PCM to G.729 (or iLBC). This is also CPU instensive. Do you think that 200 to 500 simultaneous calls can be handled? Asterisk manuals state that a high end system can handle about 100 concurrent calls, and that is just voice. Is your product that much more efficient? (I hope so!)
Regards,
Guillermo
|
| Back to Top |
|
| |
support
Administrator
Joined: January 26 2005
Location: United States
Posts: 1666
|
| Posted: August 16 2006 at 3:20pm | IP Logged
|
|
|
Hi Guillermo,
Bandwidth Calculations:
Yes, we’ve seen numerous different VOIP bandwidth calculations too on the web. Regardless of how you calculate it, the limiting factor will probably be the interrupt service handling of the host PC – not the bandwidth of the network hose connected to the server.
Server NIC:
We do not favor one over the other. We use 10, 100 and gigabit Ethernet I/O. All operate well.
CPU loading due to CODECs:
LanScape’s implementation of G729/G729A and iLBC are fast. Relatively speaking, G729A requires the least CPU. iLBC and G729 are about 2x the CPU compared to G729A.
200 to 500 simultaneous calls:
The only way to know is to design the app and test it. NASA has an old saying: “Fly like you test and test like you fly”. We can toss out number and guess all day long about performance for a particular host PC, CPU(s), NIC, operating system, network bandwidth, etc. The only true way is to put it together and see how it performs.
Repost as needed,
Support
|
| Back to Top |
|
| |
|
|
Active Topics 
Member List 
Search 
Help 
Register 
Login
LanScape Support Forum -> LanScape VOIP Media Engine™ - Pre-Sales Technical Support
Topic: Performance benchmarks
