Stack Configuration

---

In many embedded TCP/IP stacks, the ability to reconfigure them for small footprint or to simply reduce the functionality that’s available can result in greater performance. For example, if only one interface is available for the protocol stack, removing the logic required for multiple interfaces can reduce footprint and increase performance. Configuration of upper-layer protocol options such as SACK can also lead to better performance. For example, SACK is less useful for applications that communicate over a LAN, so if your application does not require a WAN, this option may be compiled away for better performance.

Other configurations include support for jumbo frames. If an application can support larger MTU frames, better performance can result. Consider the amount of time required for processing packets at a variety of rates. For line-rate performance at 10 Mb/s, standard MTU packets (1500 octets) require 1.2 ms per packet (833 packets per second). For 9000 MTU packets (standard jumbo frame), only 138 packets per second are processed, or 7.246 ms per packet. This benefit becomes even more clear when operating at very high performance rates. Consider now 10 Gb/s links. Standard MTU packets require 1.2 us of processing per packet (833,333 packets per second). For jumbo frames at 10 Gb/s, 7.2 us are required per packet (at 138,888 packets per second). Clearly, increasing the packet size has a beneficial effect on the time required to process each packet, making it more likely that line rate can be supported.

---