Cisco Support Community
cancel
Showing results for 
Search instead for 
Did you mean: 
Announcements

Welcome to Cisco Support Community. We would love to have your feedback.

For an introduction to the new site, click here. And see here for current known issues.

New Member

CUCM cluster design

Hello,

I'm trying to put together the best design for the CUCM cluster.  We are looking to deploy 15,000 total.

I currently have one publisher and 6 subscribers that will be involved with call processing, (3 Subs for Active Call Processing) and (3 Subs for Backup for Active subs).

We also have 2 subscribers dedicated just for TFTP and MOH.

What is the best design for setting up device pools to these 3 Active Subs and 3 Backup for the Active subs?

We want redundancy combined with distributed call processing.

I was thinking we need 6 Communication Manager groups alon with 6 device pools.

What do you guys think?

Thanks

23 REPLIES
Hall of Fame Super Silver

CUCM cluster design

First all of for 15K devices I would maximize CUCM call servers and deploy 8 CUCM call processors in 1:1 redundancy model, so 4 primary and 4 backup then building CUCM Groups comes down to the CUCM server placement, are all servers at one location, or are they split between 2 locations?  Hopefully no more than 2 as that makes things more complicated and you quickly run out of servers to build good redundancy model.

Chris

VIP Super Bronze

CUCM cluster design

Two follow-on comments to Chris's excellent response:

  • If you deploy anything except a 1:1 model with CMGs you will experience call processing outages during upgrades of the cluster. The short version of this is that phones registered to nodes that have been upgraded cannot call phones that are on nodes still running the old version. With a 1:1 model you can upgrade all of your backup subscribers, failover to them, and then upgrade all of your primary subscribers. In every other model you end up having phones spread between versions which segments your cluster until the upgrade is complete. This is messy and at 15k phones probably not acceptable.
  • The UC SRND also states that you should split TFTP and IPVMSA into their own dedicated nodes. These should not co-exist on the same node. Configuration changes in TFTP could spike the CPU which would negatively impact IPVSMA services. Also note that there is a service parameter that you need to change when deploying dedicated TFTP nodes.
The Maximum Serving Count service parameter, which can be configured, specifies the maximum number of requests that can be concurrently handled by the TFTP server. (Default value = 500 requests.) Use the default value if the TFTP service is run along with other Cisco CallManager services on the same server. For a dedicated TFTP server, use the following suggested values for the Maximum Serving Count: 1500 for a single-processor system or 3000 for a dual-processor system.

For a cluster at this size you should spend some time reading the UC SRND.

New Member

CUCM cluster design

Well I do have a 1:1 for cluster model don't I?

3 Active Subs, and 3 Back up Subs

1 Publisher

2 Dedicated TFTP Servers.

New Member

CUCM cluster design

for such a large environment, is load balancing not recommended among the primary and back up server?

Hall of Fame Super Silver

CUCM cluster design

Yes, you  do with 6 servers. You are load balancing between the 3 primary servers, the reason to have 3 backups is in case any of your primary one goes down so that you  can continue running without impacting any servers by  placing additional load.  Cisco only supports 1:1 and 2:1 deployment model, meaning you have 1 backup server for each primary server or 1 backup server for a pair of primary subscribers.

I have done quite a few large deployment and typically I do not run any tftp or IPVMSA services on the primary subs, but  I do run them on the backup subs.

So, for 15K devices I might propose something like:

9 servers

1 pub

8 call processing subs with 1:1 model

4 backup subs are running IPVMSA

2 of the backup subs are also running TFTP, if desired dedicated tftps cen be deployed, with vitalization and free node licenses this should not be a big deal

HTH,

Chris

New Member

CUCM cluster design

Chris,

Going off of your design, for the call manager groups, would you create a primary, backup, tertiary sub?

Also would you load balance all these phones among all 8 call processing subs or just create device pools to point to the 4 active subs?

Thanks

Hall of Fame Super Silver

CUCM cluster design

Assuming 8 Subs, all at one location, I would do:

CUCM 1:

Sub1

Sub2

Sub4

CUCM2

Sub3

Sub4

Sub6

CUCM3

Sub5

Sub6

Sub8

CUCM4

Sub7

Sub8

Sub2

notice all even numbered subs are backup servers and only used as backup, tertiary subs.

Note, this would change if servers are split acorss 2 data centers.

Chris

New Member

CUCM cluster design

Thank you, and they are split across two data centers.

So how would you balance that out? 

Hall of Fame Super Silver

CUCM cluster design

Assuming both data centers also have local phones I would do:

DC A:

sub1

sub2

sub3

sub4

DC B:

sub5

sub6

sub7

sub8

CUCM Goups to server local phones at DC A:

sub1

sub2

sub4

and

sub3

sub4

sub2 

for load balancing if DC A is large

same for DC2 with local subs

then create additional groups for remote sites

sub1

sub2

sub6

sub3

sub4

sub8

sub5

sub6

sub2

etc, etc depending on how much load balacing is desired, the idea is that tertiary server would be from remote DC to support remote site in case DR.

HTH,

Chris

New Member

CUCM cluster design

Sorry, maybe I wasn't clear.

Both of my data centers are on site.  4 subs are in one building and the 4 other subs are in another building.  I'm trying to load balance the phones properly between the two data centers.  The 4 subs in data center A would be the primarys and the 4 subs in data center B

With the above scenario of putting all local phones in a cucm group pointing to just to data center A subs I would lose all my phones.

Hall of Fame Super Silver

Re: CUCM cluster design

So one physical data center?

Or campus, if so what is the link between buildings?

Sent from Cisco Technical Support iPhone App

New Member

CUCM cluster design

They are two physical separate data centers on the campus.

The link between these two buildings are a 1 gigabit link.

Hall of Fame Super Silver

CUCM cluster design

I would balance between them, so:

sub1

sub2

sub6

sub3

sub4

sub8

sub5

sub6

sub2

etc, etc

HTH, please rate all useful posts!

Chris

New Member

CUCM cluster design

So in reality, my back up call processing subs would have phones on them correct?

Also how many call manager groups would you create with a 4 Active Sub and 4 Back Up Sub design?

Thanks for you help

CUCM cluster design

well Chris provided a very good info and details here

and based on what being described the with 1:1 back server shouldn't have devices registered to it in the case the primary server is while the tertiary might have

if you have 4 active servers then you can have 4 groups

with 4 backups to get the 1:1

HTH

New Member

CUCM cluster design

So here is the design I've come up with.

I would have 4 different device pools, each pointing to the 4 different call manager groups below.  Each device pool would allow for a max of 3,750 phones. I want my primary data center to contain the active subs and the backup data center to contain my back subs.

Can you guys validate if this is a good design?

Thanks!!

------------------------------------------------------------

D1 - indicates the primary data center

D2 - indicates the backup data center

CUCM1

1.) D1-cucm001 - sub1

2.) D2-cucm001 - sub2

3.) D2-cucm002 - sub4

CUCM2

1.) D1-cucm002 - sub3

2.) D2-cucm002 - sub4

3.) D2-cucm003 - sub6

CUCM3

1.) D1-cucm003 - sub5

2.) D2-cucm003 - sub6

3.) D2-cucm004 - sub8

CUCM4

1.) D1-cucm004 - sub7

2.) D2-cucm004 - sub8

3.) D2-cucm001 - sub2

------------------------------------------------------

Hall of Fame Super Silver

CUCM cluster design

This would work fine.

HTH, please rate all useful posts!

Chris

New Member

CUCM cluster design

Thank You for your help guys, especially to Chris.

Seeing now that I have 4 CUCM groups, how would I properly setup MGCP redundancy?

I have 4 Voice Gateways, 2 are 3945's and 2 are 3845's.

I was looking at this example from a post on the forums:

---------------------------------------------------------------------------------------------------------------------------------------

This will be like this in your ccm-group

server a

server b

server c

where server a is ocnfigured in your mgcp call agent config as follows:

mgcp call-agent 10.176.105.1 2427 service-type mgcp version 0.1

and server b abd c are configured as shown in the ccm-manager redundant host section

ccm-manager redundant-host 10.146.90.200 10.146.200.200

ccm-manager mgcp

ccm-manager music-on-hold

ccm-manager config server 10.126.5.200 10.127.7.80

ccm-manager config

----------------------------------------------------------------------------------------------------------------------------------------

So for each CUCM group that I have (4), would I need to do the above example on each Voice gateway that I have?

Also woud I have to configure a dial peer with a target host for each CUCM server pointing outside calls into my CUCM?

This is where it gets confusing.

Hope this makes sense.

Thank You again

CUCM cluster design

using MGCP you can use the concept you put above with redundant CUCMs

but talking about dial peers this is something related to H323 and with H323 you will need a dial peer with higher preference for each CUCM in the group to have it fully redandant

hope this help

please rate the helpful posts

New Member

CUCM cluster design

I guess my only concern with my concept of the redundant CUCM groups with MGCP is that if I lose a gateway, my whole CUCM group will lose connectivity to the MGCP gateway.

Super Bronze

CUCM cluster design

Hi

MGCP gateways work just like phones in terms of registration and fault tolerance. You put them in a device pool they download a list of servers from the CM group assigned to that device pool, and register to them in the order specified by that pool. This download happens as a result of the two 'ccm-manager config' commands you posted, which should specify the IP of a CUCM TFTP server.

If you lost an MGCP gateway, you lose the connections that are physcially on that gateway. In a cluster this size, you should not have just one MGCP gateway.

The resources that are assigned to a CM group are NOT limited only to that one group. For example, you have your four CM groups - if you only have two MGCP gateways, and they are both registered to CM group CUCM4, then all devices in the cluster can use that MGCP gateway regardless of what group/callmanager they are registered to.

I would probably have at least two gateways per data centre for resilience; you would also want to consider the number and type of PSTN circuits you need to terminate on them. You would also want to consider how you will get your calls delivered resiliently inbound to the two data centres from your service provider.

Two other things:

1) It's customary to rate responses you receive; I see you've got a lot of good help and advice here and have not done so.

2) All the help and advice you will receive from this forum isn't going to remove the need to read the SRND for the product. If you are designing a whole system yourself, you need to understand it and you don't seem to be familiar with the basic building blocks of the system, let alone the details. http://www.cisco.com/en/US/docs/voice_ip_comm/cucm/srnd/8x/uc8x.html

Regards

Aaron Harrison

Principal Engineer at Logicalis UK

Please rate helpful posts...

Aaron Please remember to rate helpful posts to identify useful responses, and mark 'Answered' if appropriate!
New Member

CUCM cluster design

Thank you for replying and that helps a lot.

In my current enviornment we use a combinatino of MGCP and H.323.

I was reading with H.323:

For devices using Unified CM Groups, it is important to know which nodes  will run H.225 daemons because calls sent to an incorrect subscriber  might be rejected. For example, this situation would occur if a Cisco  IOS H.323 gateway is configured with dial peers that send calls to  subscriber C in a Unified CM cluster but the Unified CM Group for that  gateway has only subscribers A and B in its list. In such a case, the  call will fail or be handled by an H.323 trunk daemon if one happens to  be configured on the subscriber.

So I guess that means I would have to create Specific device pools just for the H.323 Gateways so I balance out the CUCM's correctly so that if I lose a gateway that I don't lose H.323.

am I assuming correctly?

Hall of Fame Super Silver

CUCM cluster design

What you need to ensure of is that the CM Group applied to the Device Pool used by the gateway needs to reference the same CUCM Servers as the dial-peers on the GW, otherwise if a dial peer sends the call to CM "A" which is not referenced in the CM Group used by the GW the call will not be routed.

Chris

6735
Views
28
Helpful
23
Replies
CreatePlease login to create content