Hybrid Centralized Call Processing Model (Media Resources)

[MarkMcGrath]

We currently have one CUCM cluster running 8.5 with 40 remote sites hanging off of it. We're also utilizing CUC, UCCX and CUPS. Each site has its own connection to the PSTN via local PRI's. According to RTMT, our cluster is taxed out which is causing issues such as having to restart misc services to restore functionality.

I believe part of the problem is due to the way the MRG/MRGL's are setup. While there is no uniform standard across the board (another problem in itself), digging through the configs I see most remote sites use the first MRG that has an Annunciator, MOH, and the builtin SW conference bridge on CUCM. The next MRG has a centralized HW conf bridge and transcoder.

While I would like to add another server to the cluster, management will not allow approve any purchases, despite my extensive lobbying. Each remote site does have a surplus of DSP's free on its gateway, so what I am trying to figure out is if I can create an MRG with a site specific transcoder, conference bridge, and MTP. So I will retain the centralized call processing model, but decentralize the media resources to the remote sites.

Looking in the SRND, this is discouraged seeing as voicemail often uses g.711 which means I'd have to transverse the WAN with extra bandwidth. Other than that, very little is said though. I'm trying to untangle a rats nest here and my options are limited, so any suggestions are appreciated. I'm still drawing out the network but it does not appear like the centralized HW conference bridge and transcoder can support all the sites at once, it would need to be configured as an overflow.

[Jonathan Schulenberg]

I wouldn't go so far as to say the SRND discourages local media resources. Without knowing the exact paragraph you're referring to it is more likely that it is suggesting transcoding a call isn't recommended for voicemail since Unity Connection also does a second transcoding internally (aka the line codec vs. the recording codec) which can cause tandem compression if the wrong events coincide, degrading the voicemail quality.

Local media resources, especially when the site also has a local PSTN circuit is actually quite useful and a legitimate design, at least for audio calls. Here's how I frequently build the MRGL/MRG, one per-site:

• <Site Name>_MRGL
• <Site Name>_CFB_MTP_MRG
  Add the dspfarms from the local router that perform conferencing and software MTP. Note that a software MTP is done in router CPU and is sufficient. Hardware MTP is effectively a transcoder.
• <Site Name>_XCODE_MRG
  Add the dspfarm(s) from the local router that perform transcoding. This is important to separate from MTPs because they will get inappropriately chosen for calls that would get by with just an MTP when in the same MRG.
• Also, you should consider whether you actually need transcoders in your environment. In most cases you can avoid these. For example, if you're using G.729 across the WAN: 1) that sucks, stop doing that; and, 2) you can still create a dedicated region for CCX that allows everything to use G.711 natively when calling CCX instead of requiring a transcoder.
• (Optionally) Add the CFB_MTP_MRG and XCODE_MRG from the central site to serve as overflow or backup to the local resources.
• <Data Center Site Name>_IPVMSA_MRG
  Add all of the MoH, Annunciator, software MTP and software conference bridge resources here. If there is a DR data center you may want to put that in a separate MRG so you can maintain the primary vs. DR roles.

Also, just a reminder to use the DSP Calculator to decide how many sessions you can enable on your conference and transcoding dspfarms.

[MarkMcGrath]

This is the section of the SRND I was going off of. I do not want to have to reconfigure the regions for VM, if possible.

Digging deeper today, I noticed that the Centralized-DP is assigned to all the VM ports. There is no MRGL assigned to the DP at all. So remote sites may be using G.729 and invoking a stray unassigned transcoder. I would like to know a little more about the built-in unity transcoder as well, it's possible that the server may be doing the transcoding, will know more once I run some traces.

[Mohammed al Baqari]

Hi,

Centralized deployment means centralizing call processing nodes. It doesn't mean centralizing media resources. In fact centralizing media resources is a bad idea as it will cause quality degradation and BW waste .

Ideally, in centralized model you should have and MRGL per site which contains the following:

1. Transcoder provided by the local gateway DSPs

2. MTP (software or hardware) provided by the local gateway

3. CFB provided by the local gateway

4. MoH Provided by Centerlized CUCM

This MRGL should be assigned to all endpoints on the site including phones, gateways, softclients.

Also, you should have an MRG called DONT-USE-MRG which isn't assigned to any device. This should contain your software CFB and MTPs provided by CUCM. You should use software CFB/MTP as they are limited to G711 codec, 3 participant conferencing, etc. Additionally they consume CPU resources