I've seen it mention a few times but no actual explanation of what it is. How does it improve performance and is it worth the risk to try it?
Google "Cut-through forwarding", and search this forum for "CTF" - there were a few posts.
As for the Broadcom' implementation - it's a black-box…
If you would like to try it - build the latest version from git. There's a high risk of the router functioning incorrectly, or even being bricked, if you activate CTF in released builds 53 and 54.
Any clue whether it requires certain hardware support? I know it was only added with the stock firmware of the latest generation of devices…
I ran a quick test with it (from current git) on my old WRT54GSv4 last night, comparing CPU use under heavy traffic. Saturating my downlink at ~750 KB/s, with one or two active connections, I saw values like:
with CTF enabled: CPU: 6% usr 5% sys 0% nic 42% idle 0% io 1% irq 44% sirq CPU: 1% usr 1% sys 0% nic 55% idle 0% io 0% irq 40% sirq with CTF disabled: CPU: 6% usr 6% sys 0% nic 40% idle 0% io 3% irq 42% sirq CPU: 4% usr 7% sys 0% nic 39% idle 0% io 1% irq 46% sirq
…so: enabling it does seem to offer a reduction in CPU use for me, but only a marginal one in my current config. I haven't tested it with QoS disabled to see if it has a bigger impact then.
I saw that TB created a checkbox to enable/disable this in the latest git version, but as I can't find it on the Advanced/Miscellaneous page, I asume it's hidden for further testing?
Does anyone have a K26 Build MIPS R1 NoUSB/VPN for WRT54GL featuring this? I'd love to give it a try..
Don't really want to clog my linux server up with build environment :)
Pick one of these for MIPSR2 only if you would like to try CTF and report back if it does anything!
Download site http://www.4shared.com/dir/v1BuINP3/Toastman_Builds.html (inc. torrent downloads)
Make a Paypal Donation http://toastmanfirmware.yolasite.com/donations.php
The CTF Discussion sort of died. Has anyone done anymore testing?
I am far from having done any real scientific testing, but I can at least say that the post-build-54 implementation seems solid enough (i.e. it doesn't break anything observable). It _may_ be placebo effect, but it would *seem* that things are considerably more responsive, especially VPN traffic (go figure).
My Tomato utilities site: http://multics.minidns.net/blog/articles/tomato_utilities