So I am a part of building the NAG LAN at the biggest Gaming Expo in South Africa.
rAge

My main purpose at the event will be to set-up the the LAN cache, as done here by Multiplay.

I have 2 x the below to work with:
2 x Intel Xeon CPU E5-2690 v2 @ 3.00 GHz, 10 Cores, 20 Logical Processors
1 x SSD 160GB
5 x SSD 120GB
64 GB RAM
8 x 1GB Eth

These are some pretty BEASTY servers.

Lets just crunch some numbers quick:
There are 10 + 10 physical cores – Which gives us 20 physical cores.
There are 20 + 20 virtual cores – Which gives us 40 virtual cores. Madness?

Then if we were to put all 10 x 120GB SSDs in RAID 0 in one server…
Lets assume that the SSDs have a read/write of around 500MB/s these days.
That is 5GB/s of space. Pure space.

And we have 8GBps of Network traffic, so even in a setup like this, the storage is still a bottleneck.
We have a 5.2GBps internet connection coming into the event, so this would be the “backhaul” for us to connect to Steam servers and get stuff for the cache.
But why would we even want a SteamCache with a connection like that?!
Well, we would like to make sure as much of that pipe is left open for games, other downloads, torrents, whatever. If we can take some duplication off of that connection and keep things outside the LAN as quick as we can, that would be a win.

So if we were to put all the 120GB SSDs into one server, we would only have 1.2TB of space. This really concerned me at first, but then I was made aware that nginx has something to deal with this:
“The special “cache manager” process monitors the maximum cache size set by the max_size parameter. When this size is exceeded, it removes the least recently used data.”
==
if all-cache.size > max-size:
del *the most inactive cache* until size < max-size
==