|
Dan Cox wrote:
> Hello..
>
> I was thinking about using the VS for linux.com's current layout (we are
> expecting
> massive amounts of hits). unfortunately certain issues and "The
> Boss" deems I
> must run 2.2.x, so it looks like I'll have to wait a bit to implement
> it. Anyways, I'm
I think the VS-Tunneling or VS-DRouting is good for a heavy load
site like linux.com.
For the linux.com site, I will make the VS patch for kernel 2.2 running
well as soon as possible, I now have a little more spare time to play
with LVS because now I cannot make the trip to the LinuxExpo
which I had planned for a long time. I am regret that I couldn't see
you there. Anyway, Dr. Joseph Mack from National Environmental
Supercomputer Center will give the presentation of LVS for me on
(1:30 Thurs, Conf A.). Thank Joseph. If any of you go to the
LinuxExpo, you can have lunch together. :-)
Peter, I think I will take your current patch as a basis. I will try to
make license and authoring information that everyone is happy with.
>
> using the portfw ipmasq implementation in 2.2.x of load balancing via
> NAT, and
> this is the current problem I'm running into on my load tests..
> I believe this would
> also be a problem when using VS/NAT, so that's why I'm posting my
> question
> here.. I think I saw somebody else was having the same problems..
>
> Apparantly the max number of masq'ed entries is 4096, as (I believe from
>
> somewhere I read) is defined in include/net/ip_masq.h
>
> #define PORT_MASQ_BEGIN 61000
> #define PORT_MASQ_END (PORT_MASQ_BEGIN+4096)
>
> I tried fidgeting with these settings. ie. 56904 for BEGIN and
> BEGIN+8192 for
> END, but I am still running into the 4096 limit. Does anyone know what
> the
> problem/fix might be? or am I totally looking in the wrong place? :)
>
The number of free masquerading ports doesn't matter, because they
are used for internal masquerading users to access the external
services.
Please send me your current network layout and some testing result
that you have done, which will help me tune the performance, because
you have a really good virtual server testbed, and I don't.
>
> Thanks-
>
> Dan Cox
> Linux.com - Systems Administrator
>
>
Thanks too,
Wensong
|
