I'm looking into doing an upgrade of our NAV installation, we're on 3.15
and it's time to move on to 4.
I'm looking for advice on hardware setup, filesystem layouts and so on.
Currently we have about 1500 devices, 99% routers and switches, in the
database (mostly Catalysts ranging from 2940 to 3750, a few 6500 and
Nexus 7k/5k, Juniper SRX/EX and a Juniper QFX, around 50 virtual Cisco
ASA contexts as well).
Hardware is a single server, 2x dual core opteron 2.6GHz, 32GB ram and
several raid1's (in hardware). I've put postgres on its own 15k RPM
raid1, the same with /var/lib/nav and confined /var/log/nav to the
OS-drives with its own partition.
I have been and will be using the excellent debian-packages from UNINETT
for my installation.
I've done some tricks for optimizing disk performance, by using the
"noop" disk scheduler I see a big difference in system sluggishness,
even when I have loads around 30. This is mostly related to my hardware
raid, the default debian scheduler does not give a good performance when
operating on a hardware raid (same thing with SSDs).
I'm planning a new installation, this time on two servers, but I have
yet to determine disk and raid layouts and what to put on each server.
Postgres away from the collector is rather given, I'm unfamiliar with
graphite, could/should it be put on another server? (I might find a
third server if its recommended.)
What will be the default datastores be? I'd like to set up separate
partitions before installing the NAV-packages to make sure filesystems
like /var don't run out of space when NAV logs grow, postgres goes wild
and/or graphite stores what I'm actually asking it to store... What
will typically be hit hard with random IO, sequential read/write and so
Any input would be apreciated.