We have 40,000 Elements and are attached to a SAN. However, the amount of netflow we pile on top of the SNMP data is causing us issues with response time. We see long disk Queue's at times which in turn cause slow response times when querying data. I would suggest local disk(s) if possible.
We're using DAS with a Dell MD1000 and still have disk queue issues. The controller has a 12Gbps bus and the SAS15k RAID5 has 3Gbps x 5 drives. We're getting ready to light up the second controller and format the disks as a RAID10 with more spindles. I have asked tech support for "SQL Bulk Insert" as a method of fixing I/O issues that EVERYONE should be feeling. If you add KIWI Syslog Server to the front end and have it pipe to NPM, KIWI can handle the bulk insert better. We're writing a .NAT app that will insert 500 lines of data as fast as NPM currently adds 1. I'm still puzzled why this product can be considered "Enterprise "class when it is so ineffecient at handling large amounts of syslog.
We've got about 20k total elements and run our DB server from a SAN. (Our Pollers are VM's running on EXS boxes that are attached to the SAN as well.) We have a few performance issues but it is much better than we had previously seen.
At one point or SQL Server was a VM that was attached to the SAN. The web interface was pretty slow... After a few meetings, we were able to get the DB moved to a physical server and things were much much faster. We've never ran the DB from local disk so I can't give a good compairison.
I've been running VM SQL 2008 EE and SAN (for both VM disk and data/log disk.
I've also used local (way back).
Recommendations/gotcha's - try to get different LUNs for data and logs and separate the two "files" across the two LUNs.
- the more disks in your LUNs, the better the perfomance
- make sure you have decent/appropriate amount of RAM on DB system.
we also have almost 20k elements, and our entire Orion environment runs on VM/SAN.