From: Evgeniy Polyakov <[EMAIL PROTECTED]> Date: Mon, 31 Jul 2006 14:50:37 +0400
> In syscall time kevents copy 40bytes for each event + 12 bytes of header > (number of events, timeout and command number). That's likely two cache > lines if only one event is reported. Do you know how many cachelines are dirtied by system call entry and exit on typical system? On sparc64 it is a minimum of 3 64-byte cachelines just to save and restore the system call time cpu register state. If application is deep in a call chain, register windows might spill and each such register window will dirty 2 more cachelines as they are dumped to the stack. I am not even talking about the other basic necessities of doing a system call such as touching various task_struct and thread_info state to check for pending signals etc. System call overhead is non-trivial especially when you are using it to move only a few small objects into and out of the kernel. So I would say for up to 4 or 5 events, system call overhead alone touches as many cache lines as the events themselves. - To unsubscribe from this list: send the line "unsubscribe netdev" in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html
