Take the 2-minute tour ×
Server Fault is a question and answer site for professional system and network administrators. It's 100% free, no registration required.

Networking is not my main job and i am hoping you guys can comment on my test procedure so far to identify PCs best to host a few voip softphones. Our 100Mb LAN comprises of 40 Win XP/2000 PCs and a Windows SBS server, an enterprise switch and various consumer switch/hubs around the office.

I have run a udp test (using iperf) between all nodes of our LAN against a single workstation (on the same switch/hub as our server) primarily interested in the jitter results. I used the JPerf user interface which sets bandwidth max setting at 1Mb by default. I ran each workstation test individually so as not to overload the single PC they tested against. Each test runs for 10 seconds.

Ive collated the results: a few workstations stand out with having poor results (jitter above 1ms), some workstations have 0ms jitter and others have 0.0x jitter. All workstations were able to sustain the 1Mb connection, though some workstations jitter results varied greatly during the 10 second test. I have identified 3 workstations on one switch/hub with poor results to i will change about that setup.

Having run the test once, how much faith can i put in the results as being representative of the quality of the connection paths tested?

It may be time to just trial the softphones avoiding those PCs with poor results rather than spend too much time trying to decipher test results i am not experienced enough to fully appreciate, but your input would be a greatly appreciated on what i have done so far.

Other info: I will not be utilizing QoS, softphones in use will be limited, our internet connection is than sufficient for the voip, other network load is limited.

share|improve this question
    
Is it a switch or a hub? The two terms are mutually exclusive, and, depending on which one it actually is, could have bearing on the behavior you're seeing. –  EEAA Nov 24 '09 at 1:19
    
Also - how much jitter are you seeing? –  EEAA Nov 24 '09 at 1:19
    
They are switches, belkin mostly 8 port ones off the shelf from the like of PC World. Some connections have 0ms jitter, some in the region of 100ths of ms of jitter, and then a small number of workstations with >1ms jitter. Im more interested in the validity of my test method, would you not expect to see any jitter on a small lan? –  Gary B2312321321 Nov 24 '09 at 1:27
    
If you were using decent switches, I wouldn't expect to see much jitter, but with consumer-grade switches, all bets are off. Do you see jitter consistently from the same computers/circuits? If so, there may be a cabling problem. –  EEAA Nov 24 '09 at 1:30
    
what do you classify as "not much jitter"? –  Gary B2312321321 Nov 24 '09 at 1:37

2 Answers 2

up vote 0 down vote accepted

I'll agree - 100ms of jitter is quite a bit. A properly-configured jitter buffer in your VoIP system can help with that, but at much more than 100ms, the delays introduced by the buffer can make conversation quite awkward.

I'd recommend trying narrow down the source of jitter. You're running consumer-grade switchgear. While that's not definitively the source of your jitter, it's certainly not helping. Cheap switches nearly always switch in software and as such, are very susceptible to delays cause by other network traffic traversing its backplane.

Poor wiring can also contribute to jitter, so you ought to check into that as well.

Overall, I think you're going down the right path, and being very intentional. Though you didn't state this explicitly, it sounds like you're trying to determine if your network is robust/stable enough to support a VoIP network. At this point, I'd say go for it. You may uncover jitter issues on a couple circuits - when this happens, you'll have some troubleshooting to do, but for the most part, you shouldn't have many issues.

Also, as an aside, you were testing to 1Mbit streams. That's much higher than any VoIP traffic is going to be. If you're using G.711 (which you most likely are), you'll only need about 88Kbps per call - much less than the 1Mbit you were testing with.

share|improve this answer

Hmm. You can sometimes get really good throughput with wireless N. I do at my house, from a 20-ft distance. I get 40% connectivity at 20-ft through a wall. I use a $35 Dlink DIR-615 N-Router connected straight to my cable modem. The other D-Link routers have disconnectable antennas if you need more gain/distance.

Then, put a wireless-N adapter on each PC that wants the VOIP. Even if they get 25% connectivity, that might be better than the LAN wire. I measure the percentage connectivity from the router side, not from the client side.

share|improve this answer
1  
Why would you advocate adding in another possible source of jitter into this situation? Unless you have the money to shell out for enterprise APs and WLAN controllers, VoIP and wifi don't mix. –  EEAA Nov 24 '09 at 2:13
1  
Also - bandwidth != latency. Though you may be able to higher bandwidth using 802.11n, you'll certainly see much higher variance in latency (this is what jitter is, essentially) than when using wired ethernet. You don't need oodles of bandwidth to run VoIP. What you do need, however, is a fairly jitter-free connection. –  EEAA Nov 24 '09 at 2:27

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.