On Mon, Dec 04, 2006 at 07:15:58AM +0000, Gerrit Renker wrote: > the top performance even with CCID2 is 15Mbits/sec. CCID2 used to be 1gbit/s in august with my "performance" patches, now commited. I ran on an intel xeon 3ghz i think, with PCIe and e1000 gbit cards. It's stopped being gbit/s around september I think, towards the time that the TX buffering stuff + some other was introduced. This doesn't mean that those patches are ``problematic'' but rather that something in the kernel changed around that time... maybe even other random code. I don't know what the deal is now. I highly suggest the use of oprofile in case you aren't using it yet. To relieve CPU usage: * boot parameter: clock=tsc * the most expensive thing should be checksum calculating [~25%]. If it's not, then something fucked up in DCCP. * After checksum calculation, the profile should be flat. That is, 100000 functions, each taking 0.1%. Regarding checksums, have a look at: http://darkircop.org/check.png Different hardware has different checksum calculation capabailities [same checksum code in user-land [i386 code]] =D. We need to offload to the NIC at some point and see if we can gain that extra speed. Also, partial checksums in DCCP will help a lot [cf. TCP]. - To unsubscribe from this list: send the line "unsubscribe dccp" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
