Linux TCP stack and network cards performance: packets per minute before drop?

Hi all, how much TCP packets should come at 100/1000mbit network card (not an embedded Realtek's home solution) before the Linux kernel will become unable to process the NIC's buffer resulting in packet drop?

Hardware: let's say about dual-core Xeon + Intel's gigabit PCIE card. The final product is planned to capture and process the packets coming from a large net (doing the same job that Netflow does, but with few specific features). But currently, have no test environment available to pre-calculate possible load per server.

Software for such a basic load test: linux's kernel ~2.6.30, netfilter, 2-3 iptables rules. Some kind of network similation is in progress, too.

From serverfault mhambra

• Hi all, how much TCP packets should come at 100/1000mbit network card (not an embedded Realtek's home solution) before the Linux kernel will become unable to process the NIC's buffer resulting in packet drop?

  If all processing is in two iptables rules, we can expect performance 100 MBps and 0.3~0.4 MPps.

  Intel PCI-E NIC good choice for this purpose, they are well performance tuning. Dual-core Xeon not the best choice. Core 2 Duo/Qaud >= 3 Ghz or Core i3/i5/i6 faster.

  From bindbn