To: lvs-users@xxxxxxxxxxxxxxxxxxxxxx
Cc: Julian Anastasov <ja@xxxxxx>
From: Wensong Zhang <wensong@xxxxxxxxxxxx>
Date: Tue, 26 Mar 2002 21:14:06 +0800 (CST)
Hi Alex,

On Tue, 26 Mar 2002, Alexandre CASSEN wrote:

> Hi,
> I have a question regarding KTCPVS. I have started learning/entering the
> code and I have a question regarding implementation.
> A nice piece of technology is TCP-SPLICING which define a nice design for
> enhancing TCP forwarding soft. For a layer7 switching piece of code there
> is mainly 2 designs :
> 1. Userspace => Here we have to deal with heavy overhead since soft spend
> most of its time copying buffer for kernel-space to user-space (and
> reciprocity)
> 2. Kernelspace => With nice design we can have a enhanced forwarding soft.
> TCP-SPLICING was designed to work for SOCKS design but can be extended to
> loadbalancing. This design is mainly used in userspace. So I was wondering
> :
> Do the use of TCP-SPLICING in a userspace piece of code will be equivalent
> in performance as putting the code directly into the kernel ?
> The best will be both => Kernel prog using TCP-SPLICING when forwarding
> between sockets pair. But the most consuming task is the stream handling
> not the accepting stream ...
> I am looking for TCP-SPLICING code to watch into ? can some one can provide
> me that kind of soft ?

I know TCP Splicing, but there is no code available on Linux (as far as I
can search on the Internet). I started to write one a month ago, but I 
didn't get time to work on it this month. The draft code of TCP Splicing 
for Linux is available at

Maybe I will polish the draft code and release the first version in the 
following weeks. :)



<Prev in Thread] Current Thread [Next in Thread>