H.264 - Polycom and Codian

[EDierks]

I am doing some testing in our environment with an RMX 2000. In production we are running Polycom HDX 8000 endpoints and a Codian 4505 bridge.

Initially I was having an issue connecting the RMX to endpoints in 6 remote offices, which showed transmit packet loss (from the respective endpoints at 10-20% consistently). All but one of these was connected to our MPLS network via a 1.5 MB link. The other was a large office that was pushing the limit on its 3 MB link. The Codian was connecting without any such issue.

I replicated the issue on the Codian by shutting off "flow control on video errors" and enabling the H.264 protocol. With H.264 disabled, I got no packet loss, but SIF video. With H.264 and flow control, I get 720p and no packet loss.

Unfortunately, I'm not able to replicate the good behavior on the RMX by enabling LPR. I also have PVEC enabled on the endpoints.

Just wondering it is about H.264 that would cause it to have so much packet loss and what might this be telling me about the network connections for these 6 offices?

[phantom_iv]

Just wondering it is about H.264 that would cause it to have so much packet loss and what might this be telling me about the network connections for these 6 offices?

I think the problem is more to do with bandwidth utilisation. While h264 uses less bandwidth to achieve the same video quality that h263, I believe the Polycom HDX is only able to encode CIF resolution at h263, meaning with h264 enabled the HDX will be transmitting ~9x the amount of pixels (for 720p) as when you're using h263. Even allowing for a greater compression efficiency of h264 this will use a significant extra bandwidth to transmit all that extra information - probably of the order of 4x the amount.

This produces significantly more load on all your switches, routers, firewalls etc - it only takes one component of the network path between the MCU and your endpoints to get overloaded and you've got some nasty packet loss.

You say you've got a 1.5M link to each office - is there any other traffic on this link at all, or just this one video system? On one of your other threads you say you're trying to make 1.5M calls - I would suggest you're not going to be able to acheive this on a 1.5M link. Try turning down the bandwidth you make the call at to 1M (or even lower?).

The HDX endpoints may not be able to transmit 720p video at these speeds however, but you might find your MCUs can transmit HD back to them (which should be fine for a CP layout in many cases, obviously not so great for full screen video).

[AcidF]

Initially I was having an issue connecting the RMX to endpoints in 6 remote offices, which showed transmit packet loss (from the respective endpoints at 10-20% consistently). All but one of these was connected to our MPLS network via a 1.5 MB link.

Hello EDierks

I am not sure if you have looked at this but QOS also plays a Big role in the VC network. Even if it is going over the MPLS or the normal LAN they will need to have QOS on the network so the packets dont burst over the nework and give you more errors.

So that nothing else will take over your pipe you have and if there is anything big ( email or internet ) using the pipe with in your VC you will know that the VC gets the bandwidth needed.

Setting your QOS : set it for the amount of bandwidth you would like to make the call at, with a 20 % overhead for the odd burst.
384KB + 20% = 406KB ( just an example )

Check the ports on the routers and switch as well...I have had problems with that as well.


Hope this helps and you have come right.

[EDierks]

Thanks very much!

I experimented with QOS for the RMX. QOS was implemented by using the IP of the Codian. I swapped that to the MPM+ address on the RMX and took the Codian down. After this I really didnt see any difference in packet loss. The other part that is puzzling is that calls at 384K or 1.5 MB show the same amount of packet loss and it is only packet loss on transmit from the endpoints, receive is 0.

We're taking this step-by-step, but ruled out duplex mismatches on the RMX and endpoints, confirmed that packets from the RMX and Codian are following the same network path via traceroute, and I'm still puzzled with the 8 endpoints/5 offices that have no such packet loss.

I will definitely keep you all posted.

Thanks again!

[robertk]

You can also try looking at the MAX MTU SIZE value.

Sometimes you have WAN links that fragment the packets, lovering the MAX MTU SIZE will make the packets go trough your network unfragmented.

//Robert

[kr244]

Hi there,

have you had a chance to sort this issue out?

Since I've experienced exactly the same problem. (high packet loss on RMX calls)
For VSX to VSX calls everything seems to work fine. But if I connect these endpoints with the RMX I experience high packet loss.
Not sure if this is because VSX to VSX calls use CIF and RMX-calls use 4CIF?!

Cheers!

[Bullcrappy]

Bump
- any word or fix for this?


Hi there,

have you had a chance to sort this issue out?

Since I've experienced exactly the same problem. (high packet loss on RMX calls)
For VSX to VSX calls everything seems to work fine. But if I connect these endpoints with the RMX I experience high packet loss.
Not sure if this is because VSX to VSX calls use CIF and RMX-calls use 4CIF?!

Cheers!

[Gargamel]

Basic answer but if you have packet loss on every call to your RMX ;

# first look at the switch port negociation. The switch should tell you that it has negociated Full Duplex, and 100 or Gig.
# then, if you still have loss, connect 2 endpoints (could be PVX) to the same switch and you'll see if it comes from RMX.