On Sunday, 10/19/2003 at 10:01 EST, Adam Thornton <[EMAIL PROTECTED]> wrote:
> How much does a large MTU actually help, even when everyone supports it? > > I usually leave mine at either 1492 or 1500, regardless of the allowable > interface maximum, because there's often something in the path that > doesn't understand how to fragment, and it seems like the gain is > negligible. You're right that using a large MTU on a path where fragmentation is required hurts the equation. That's the great thing about Linux path MTU discovery (PMTUD); it will adjust the MTU for a particular destination to ensure the packet is as large as it can be without fragmenting, regardless of what you code for the MTU. If you are streaming data (e.g. ftp) then large MTUs help by reducing the number of times the stack must build a frame to send, and it reduces the number of times that the receiving stack must read another frame off the device. Since you *know* all the data has to get from A to B, you might as well send as much as you can all at once with as little overhead as possible. So, a 1500 MTU is great for the conversation of client to web server. A larger MTU for web server to app server. The largest MTU for app server to db server. Basically, wherever you've got a DMZ, you can consider whether a different MTU is more appropriate (if allowed by the h/w, of course). The more data that has to flow, the more likely it will benefit from a larger MTU. (One has to actually measure the real workload.) Alan Altmark Sr. Software Engineer IBM z/VM Development
