Determining how many front-end servers to deploy in Skype for Business Server 2015

I figured it was time to update this article to reflect Skype for Business Server 2015, from the previous Lync Server 2013 version.  For those that would like to reference the previous article I wrote, you can find it here Determining How Many Front-End Servers to Deploy in Lync Server 2013.  Through many deployments of Lync Server 2013 and newly released Skype for Business Server 2015 the calculations haven’t changed all that much, but I still find plenty of confusion in the community regarding planning and sizing the appropriate amount of Front End (FE) servers for a pool deployment.  So I thought it best to go ahead and give another stab at the article, but this time making it relevant to the latest version of Lync Server 2013 being Skype for Business Server 2015.

The Requirements

One of the areas that has changed from Lync Server 2013 is the maximum number of supported FE Servers in a pool has increased from 10 to 12 now with Skype for Business Server 2015.  The total number of users supported per pool still stands at 80,000 users; technically it’s 79,920, but we will come back to that number a little later.  So let’s begin with a few facts:

Recommendations

The recommendation typically is to have no less than 3 FE servers in a pool.  Now mainly we know about this from the fabric and pool quorum which is true; but another area regarding capacity numbers we will see that having at least 3 FE servers is quite more beneficial than going anything less. We know that the more the better, but anything less than 3 FE servers, really doesn’t do us much good.  So for the sake of conversation, let’s say we go with 3 FE servers; let’s start doing the math.

Calculations: 3 FE servers = 19,980 (20,000); so that means each FE server will handle 6,660 users. 

If we are maxing out this Skype for Business pool with 20,000 users, we are going to plan for the N-1 scenario.  That is where we basing our capacity on the fact that if a single FE server goes down out of the 3 total, that the remaining 2 FE servers can handle the load of our total user population.

With that assumption this means we have 6,660 users per FE server across 3 servers; which also means 6,660 is divided into the 2 servers remaining FE servers in the event that a single server goes down.  This leaves 3,330 users per FE server to the already existing 6,660 users already on the FE server which means have a total of 9,990 (10,000) users per FE server, which is over the supported Microsoft limit of 6,660 user per FE server.  As a result, the decision to have only 3 FE servers in a pool if we have 19,980 (20,000) users for the pool is not a good one, considering the N-1 factor of losing a single FE server and still be in a supported state. 

Mitigating the N-1 Factor
 

So what happens if we add a single server to the existing 3 FE server pool to make 4 FE servers?  By adding an additional server to the pool we now can have a max of 26,640 users to the pool.

Calculations: 4 FE x 6,660 users = 26,640

Now by adding just a single extra FE server, this makes each server in the pool of 19,980 (20K) users able to handle 4,995 users per FE server. 

Calculations: 19,980 users / 4 FE servers = 4,995

With each FE server in the 4 FE server pool is handling 4,995 users; if a single server goes down this means 4,995 users will be spread across the remaining 3 FE servers in the pool.  That makes 1,665 users per FE server now added to the existing 4,995 users for a total per server 6,660 user per FE server.

Calculations: 1,665 users + 4,995 users = 6,660 (funny how we come back to the number 6,660 again)

So the calculations tell us if we have 19,980 users in a pool we should deploy 4 FE servers considering the N-1 factor which means that our pool can still handle the load of all the users in a supported manner with respective a single server going down.

Calculations: (N-1) 3 FE servers x 6,660 users = 19,980

Keep in mind that the FE server is at its max with users for 3 FE servers and cannot handle another FE server going down at that time; even to be rebooted for maintenance. If that was the situation the remaining servers shouldn't even be rebooted with a single FE down and only 3 FE servers remaining.

If 1 of the remaining 3 FE servers where to be rebooted during this time means the 6,660 users on the rebooted FE server would temp have to be moved to another FE server which would make us non-compliant with the supported number of users on a FE server.  For what if the rebooted FE server didn’t come up?

1.N-1 Scenario (Single FE server goes down)

2.FE server A = 6,660 users

3.FE server B = 6,660 users

4.FE server C = 6,660 users

5.Then if FE server A goes down or rebooted for maintenance 6,660 users are evenly spread across the remaining 2 FE servers

6.3,330 users go to FE server A and FE server B

7.FE server A now has 9,990 users (Supported Max = 6,660)

8.FE server B now has 9,990 users (Supported Max = 6,660)

That would leave 19,980 users on 2 FE servers and I don’t have do the math to show you that would mean there is more than 6,660 users per FE server then.

Take a What-If Approach and Do the Math

Determining the right number of Skype for Business FE servers can be tricky when we don’t take time to do the math.  Take into consideration I tried to keep a simplistic scenario of the N-1 approach; you are more than welcome to go deeper such as N-2 or play the what if game to your heart desires.  Just keep in mind that whatever scenario you use that you end up having enough FE servers to handle the user load adequately and not impact the performance of the IM, conferencing, and Enterprise Voice features.