We are hosting some big websites in our data centre with two GigaBit Uplinks to Internet. A Cisco 6503 (only switching) speaks Netflow (layer 2) to a Dell 1860 (Dual XEON with 4GB RAM) with Debian Sarge on it.
We are confronted with 'event-times' and 'non-event-times'. On non-event-times we have a traffic noise about 20-40MBit/s. (about 2000 flows/s)
On event-times we experience traffic over 1GBit/s (estimated 40.000 flows/s and more, damn much - I know ).
Till today we used a Dell 850 (2.8Ghz) with 1GB RAM, and with traffic about 100MBit/s the machine was completely overloaded. (Netflow said that switch/router time is not synchonized, load was about 10)
If I am right sampled netflow dont't help me, because it only decreases performance used by the router. And netflow aggregation is not supported by Netflow Analyzer because of missing version 8 support of netflow.
Now the new Dell 1860 (with traffic about 40MBit/s and about 2000 flows) although it's a bigger machine, runs at a load of 2. So I guess with the next event/s (which are actually the soccer-worldcup-games) my new server is also totally overloaded.
I allready did the 'database-optimizing" to the accordant amount of RAM (--key_buffer_size=720000000 --innodb_buffer_pool_size=848000000), but with no improvement.
What else can I do to account all traffic? Is there another possibility for optimizing the database? Installed RAM is completly used by java, CPU is nearly idle.