VoIPNorm MythBusters: The Great Server Myth



Myth #4: “You need to buy hundreds of servers, one at every site”

I absolutely love this one, because it is so easy to bust. The competitors to Microsoft love to show the expanded OCS Enterprise Edition configuration because it makes it look true when in fact in 99.999% of cases expanded Enterprise Edition deployment is not recommended or required. They also forget to mention other improvements like hybrid gateways. How many servers you deploy is really up to what services and options you want to roll out. I think customers can also be confused by some of the terminology around OCS. The term server is used to describe what really can be a service or role running on an OCS Consolidated Front End. This in turn is confused as requiring a physical server when it does not.

Competitors also like to keep the conversation telephony or vertically focused. With OCS this just isn’t the case as the one platform is meant to be able to take on all workloads. Although I am not going to offer up a competitor comparison here, if you were to take the example UC services with server count I have presented and do a comparison, I think it would surprise most people that OCS and especially wave 14 typically would do more with less servers.

Example deployment

Let’s take as an example a 5000 user deployment for both R2 and wave 14. This really is a pretty typical number and there are great deal of organizations that are either at or under this number. The size of a deployment for this many users can vary from 1 Stand edition server doing IM and presence only (which could also be virtualized for this workload) right up to a fully blown UC deployment with high availability.

The server count below includes all OCS workloads on what is referred to as a consolidated Enterprise Edition configuration with high availability (99.999’s reliability for all workloads) around the front end servers and PSTN access. This configuration would allow remote VPNless connectivity, Public IM access, federation to external companies, SIP trunking, Unified Messaging, PSTN dialin Conferencing, Single Number reach, IM/Presence, web collaberation etc, etc.

OCS R2 Central Office for full UC with high availability:

2 Front end servers (SIP register, AVMCU etc)
2 backend SQl database cluster for HA
2 mediation servers (transcoding and encryption for PSTN connectivity)
2 PSTN Gateways (Audiocodes, NET, put vendors name here)
1 Consolidated Edge Server (Remote VPNless access for all UC workloads)
1 Monitoring server (CDR Records and Quality Statistics)
1 Communicator Web Access (web client access and dial-in audio conferencing web pages)
1 Exchange UM server (Unified Messaging, for voicemail HA +1)

Branch site

1 Hybrid Gateway

And of course with Wave 14 this number will be lower with integration of the mediation server and CWA on to the Front End Servers. The additional workloads plus the addition of survivability at the branch has reduced server count while increasing features.

Wave 14 with high availability and survivability for branch sites:

2 Front end servers (SIP register, AVMCU, Mediation server, Reach web client, etc)
2 Backend SQl database cluster for HA
2 PSTN Gateways
1 Consolidated Edge Server (Remote VPNless access for all UC workloads)
1 Monitoring server (CDR Records and Quality Statistics)
1 Exchange UM server (Unified Messaging, for voicemail HA +1)

Branch site

PSTN Gateway with onboard Survivable Branch Appliance (AudioCodes, NET, Dialogic, etc)

As for having a server at every site, this really depends again on your requirements. With the mediation function now available on the gateway itself in OCS R2 this is only true if you require local MCU resources and you do not want to have a centralized datacenter deployment. Again with Wave 14 the survivable Branch Appliance (SBA) will be an option but its only required if the customer deems it to be. The SBA is also cohosted on the gateway, so no increase in server count.

In all honesty if a thrifty company wants to remove high availability and settle for disaster recovery, still maintaining the magical 5 nines (99.999% uptime) for voice they could do so with less servers again in wave 14 using side by side Standard Edition Servers.

Wave 14 Stand edition with disaster recovery for voice and peer to peer UC.

2 Standard edition Servers (SIP register, AVMCU, Mediation server, Reach web client, SQL database etc)
2 PSTN Gateways
1 Consolidated Edge Server (Remote VPNless access for all UC workloads)
1 Monitoring server (CDR Records and Quality Statistics)
1 Exchange UM server (Unified Messaging, for voicemail HA +1)

Branch
PSTN Gateway with onboard Survivable Branch Appliance

Of course you could add more server roles to add different functionality. A good example is archiving, but most companies do not want to archive their IM sessions so I didn’t include it in this example. I also included the monitoring server in all these scenarios but it is an optional component as is the edge and CWA. It really just depends on your requirements.

So with 5 servers and 3 PSTN gateways for 5000 (SE supports up to 5000 users for all workloads) users, you could enable the full UC workloads including PSTN dial-in conferencing. Of course the number of PSTN gateways will vary depending on your branch count and voice load but I think you get what I am saying.

This has been a rather long myth to break down but I think we can safely say it is BUSTED.

Final thought…….

My advice to anyone doing an evaluation of Unified Communications architectures from the different vendors is to get a deep dive of the technology from the vendor in question. Map out all your requirements before the meeting as it relates to UC and not just voice, and get your questions answered in relation to your business needs. Having a different bunch of widgets is great from a choice perspective, but if they don't meet your business needs you might be buying into the wrong solution for the wrong reason.

Comments welcomed.

VoIPNorm

2 comments:

  1. Well busted indeed!

    One comment. You said:
    "if a thrifty company wants to remove high availability and settle for disaster recovery, still maintaining the magical 5 nines (99.999% uptime) for voice they could do so with less servers again in wave 14 using side by side Standard Edition Servers"

    That is correct but I think it is important to provide more explanations as in my opinion this is a killer-capability for small businesses.

    Let's first understand what we mean by high-availability. It's a server redundancy mechanism (typically achieved via costly hardware load balancers) so that a server failure would be virtually transparent (minus some brief potential impacts).

    Meanwhile the new failover/resiliency capabilities in CS14 enable clients (end points) to register to a primary registrar and instantly failover to a backup registrar when the primary is no longer available.

    The beauty of it is that as long as your client can see a registrar (and assuming a PSTN connection is available), you will be able to at least perform the most important functions (place and receive calls, etc).

    Say your company has 2 main sites, each with say 500 employees, connected by a decent WAN (preferably with WAN resiliency but not an absolute must). Suppose you put a Standard Edition server in each site, and even better put in those servers a gateway on a card (such as the one offered by Ferrari) - now you have 2 registrars and 2 survivable branches. If one of the servers is in maintenance or fails, the other will be the registrar and provide services for all users.

    The resulting availability from a basic scenario (place and retrieve calls in particular, but not just that) is very high, without having to use hardware load balancers. That's the beauty of it - very cheap, easy, functional.

    What's the catch (there is always one, of course)? Well, during the failover, only some (most important but basic) capabilities will be available to the users homed on the server that is unavailable. Typically what will be missing is contact list and presence and things that derive immediately from presence such as presence based routing, things that have been scheduled (meetings...). But telephony works (*), and many other capabilities, including peer to peer capabilities in the branch.

    If you want a very economical solution, and can accept the potential impact (loss of some advanced features, but voice works) of occasional server failure (and this is 2010, servers' MTBF is pretty good); and manage around the impact of server scheduled maintenance, then this may well be a solution for you.

    Work is still going on with respect to what will be supported virtualized, but (crossing fingers) if the Standard Edition server is supported in that mode, you could also have say a Domain Controller or maybe an Exchange Server on that box... Sweet!

    (*) note that inbound calls may require rerouting on failure by your PSTN provider or a separate gateway - rather than the blade in the server - that would redirect inbound calls while the server is unavailable.

    Also note that if you are experiencing a longer term outage, it is possible to move the users to the remaining server - restauring all (incl. advanced) capabilities.

    ReplyDelete
  2. All great points Nemo. This is going to be a killer supported scenario that will be welcomed in the small busniess space. This also may make a great blog post to talk about it.

    ReplyDelete