With the database enabled, the key factor becomes the number of datasets that you're exporting. By default that's response time and short-term packet loss for all devices, plus sent/received bytes/sec for all interfaces. So if each of your devices had 16 interfaces, that would be 2000 * 2 + 2000 * 16 * 2, for a total of 68000 datasets.

For architectural reasons, performance degrades heavily with more than 20-40k datasets. We're working to increase this limit, but for now the best performance investment is reducing that number. A couple of hours spent narrowing the datasets to just those that you *really* want to report on has a greater impact than a million dollars of hardware.

Generally, though, the software is I/O-bound, particularly when using the database. And with a single machine, the bottleneck becomes contention rather than bandwidth. If physical space isn't a concern, it's therefore much more effective to get three lower-powered systems, running IM, IMDC and Flows separately, than a single high-powered one. SAS doesn't help nearly as much as avoiding having three processes constantly fighting over the position of the disk head.

Ideal setup on a budget would be three machines with the cheapest consumer-grade Core i7s and 4/8/8 GB of RAM for the IM, IMDC and Flows machines, respectively. If still under-budget, get an SSD for the IM machine and SATA RAID0 for the IMDC & Flows machines, with some money set aside to cover the fact that SSDs and consumer SATA drives tend to fail a lot faster than server-grade SAS.

System Requirements represent the high end of the hardware required to run InterMapper. More CPU, RAM are not likely to improve performance beyond that point. Disk storage requirements may continue to increase due to the data storage requirements; this is dependent on the data retention policies in use.

The upper limit on the number of devices which can be monitored on a single server is a function of a number of factors other than the total number of devices, including:

- the type of probe used to monitor the devices (ping/echo, SNMP, TCP, WMI, etc)

- the number of interfaces monitored using SNMP

- the polling frequency

- the data retention policies in use

- what other applications are running on the same server (including InterMapper Flows and InterMapper DataCenter)

At the upper limit, the bottleneck will normally be disk I/O.