Client Management Suite

We have a very large site (3,000 clients) and we are planning on putting two site servers active in that one site. I noticed the setting where you can control the maximum clients, but I was curious what the algorithm was for dictating what site server a client gets?

Appreciate any direction.

You only need one site server to serve that site.  Is there a reason why you've chosen to use two and then limit each in an attempt to load-balance?

We are having issues downloading packages from that server, and our assumption was that it was getting too many requests. We've rebuilt the server with no luck. We had a consultant come to our site when we first moved to NS 7.1 and he mentioned they shouldnt have over 2,000 clients per site server, so based on that, and the issues we thought it would be best.

Whats a good best practice for a maximum number of clients?

Best practice can really depend on your hardware, solutions installed, and what you have the site server actually doing (Task, Package, OOB, Deployment). Also, there are environmental factors. Is the server being used by another application?

In a nutshell, a Site Server can handle 5k nodes if specced properly(more if it is not under heavy load, but thats another story).

How many package requests are you getting a day? When you say you have 3k in one site, is that a physical location or is that a logical 'site'?

Hi Donald. Its actually 3 physical sites. The servers are spec'd pretty low and are virtual. It has 2GBs of RAM and 1 vCPU (Xeon, 1 core) at 2.8 GHz. My assumption is that this is pretty low, but im not sure. The site server is not used for anything else, but it is a deployment server, task and package server. Pretty good package load, plugin push right now, software packages, patches, etc.

Think its spec'd too low? Appreciate your time.

I'd guess you've already checked this, but is it possible that its just a WAN issue?  I run a (virtual) site server at each of my three sites.

Typically, hardware specs are based on a range of nodes. For insance, I cant give you specific specs on 3198 nodes, but there are specs available for 1k - 5k nodes.

1k - 5k should be 2-4 cores @ 2.4GHz or better with 2-4 GB RAM. What usually plays into this also is the IOPS on the disk. For virtual, you are looking for somewhere in the neighborhood of 160 on the package drive. OS drive should be seeing about 180 IOPS.

Going even farther, Zac brought up a good point. How is your WAN handling this? Are you utilizing any type of throttling, maintenance windows, or deployment schedules to not saturate your links?

I know the initial feeling is to get everything out as quick as possible. But, sometimes, a phased and methodical approach will lead to a lot less headaches.

For example, instead of patching 3000 at once, can you stagger the patching schedule? Same goes for plugin pushes.