Re:Google and Microsoft Cheat on Slow-Start. Should You?
by ergo98 on Friday November 26, 2010 @10:16AM
(#1942520)
Very interesting. Is such a thing configurable in Apache or nginx? It seems to be a rather rude behavior, but I'm curious how accessible it is.
[ Parent
| on HN ]
Re:Google and Microsoft Cheat on Slow-Start. Should You?
by riffraff on Friday November 26, 2010 @10:24AM
(#1942539)
I don't think any web server gets to this level of control on the tcp/ip level, this is something that should be addressed in the OS' network stack.
Re:Google and Microsoft Cheat on Slow-Start. Should You?
by eru on Friday November 26, 2010 @11:02AM
(#1942615)
Unless you have an exokernel operating system.
Re:Google and Microsoft Cheat on Slow-Start. Should You?
by riffraff on Friday November 26, 2010 @04:06PM
(#1943306)
Nah, that's not even necessary, I believe: the congestion window is a TCP feature, but any normal user space program (with a few privileges) can create IP datagrams with custom content, so you would only need to implement a bit of tcp in a normal old style monolithic kernel
And since you control the only higher level protocol (your http implementation) you can probably cut some corners and get done quickly :)
Re:Google and Microsoft Cheat on Slow-Start. Should You?
by panic on Saturday November 27, 2010 @02:51PM
(#1945706)
You may be able to create them, but as a server, how would you receive incoming connections?
Re:Google and Microsoft Cheat on Slow-Start. Should You?
by daeken on Friday November 26, 2010 @04:04PM
(#1943301)
No need for an exokernel here. A microkernel could enable the same thing, since your network stack would be Yet Another Service (TM). This could easily be implemented on the likes of L4.
Re:Google and Microsoft Cheat on Slow-Start. Should You?
by wmf on Saturday November 27, 2010 @12:38AM
(#1944387)
I don't really see the difference between patching the kernel and patching the privileged network stack server in a microkernel environment.
Re:Google and Microsoft Cheat on Slow-Start. Should You?
by eru on Saturday November 27, 2010 @09:58AM
(#1945038)
In an exokernel environment, the network stack server wouldn't be privileged.
Re:Google and Microsoft Cheat on Slow-Start. Should You?
by bstrong on Friday November 26, 2010 @10:25AM
(#1942541)
It's not tunable at the app level. On linux it requires a kernel patch to change. I'm not sure about other OS's.
Re:Google and Microsoft Cheat on Slow-Start. Should You?
by pquerna on Friday November 26, 2010 @10:35AM
(#1942568)
I wish there was a patch where you could enable it on linux via a ioctl/setsockopt. Would be very useful to those of us who aren't google.
Edit:
I wished, google already sent the patch in May for this:
http://www.amailbox.org/mailarchive/linux-netdev/2010/5/26/6...
Nice!
Re:Google and Microsoft Cheat on Slow-Start. Should You?
by 8plot on Friday November 26, 2010 @10:34AM
(#1942566)
something like: ip route change default via 0.0.0.0 dev eth0 initcwnd 10
Re:Google and Microsoft Cheat on Slow-Start. Should You?
by self on Monday November 29, 2010 @07:16AM
(#1950056)
Perhaps "ip route change default via <your gateway address> dev eth0 initcwnd 10" -- your gateway address, not 0.0.0.0.