On Wed, 22 Sep 1999, [EMAIL PROTECTED] wrote:
> Hi Tom , tnx for the reply, now I'm getting really confused, as I have already
> set the retry timers to linear,
> but they still behave as exponential timers.
> I was told that the kernel timers in 2.0.36 are exponential,
> Can this be clarrified ?
> the retry timers in tnos can be set to linear and when set are linear, the Tampa
> convers server, and the modified and cleaned up variant HTPP also behave
> linearly.
> But if you either telnet or ftp kernel to kernel over a not to good path, once a
> couple of packets have been missed the backoff time just get worse.
> I cant do much about the path , it got a big lump of iron ore in the way, it a
> great microwave reflector !!!
> but if I can improve the rate of retry attemps it would help..
Well the way in which i would do it, a way which has been flamed many
times is the same way G4KLX once sent out a patch for, i have los the
patch, but the idea is just change the default max backoff time from
120 seconds to what ever you want, i still use 60 for a link which
plays up every now and again. I use 2.0.36, but looking at 2.2.x code
its the same so i see no reason why it should nopt work.
One draw back is if the retry is set to low and your partner missses
all your packets then of course you lose the link, but thats an
extream case.
./linux/net/ipv4/tcp_timer.c (+/-) line 507 is;
tp->rto = min(tp->rto << 1, 120*HZ); just change the 120 to 60 for
example which will half the max wait time.
The same in tcp_output.c at line 1142 (+/-)
Saying that i think it was input.c anyway, the ipv4 directory now
contains all sorts of timers for other things masq, forwarding etc,
so it might not be such a good idea after all.
> 73 Richard
--
Regards Richard
[EMAIL PROTECTED]
