[Spread-users] Token Message
Yuval Lifshitz
yuvalif at yahoo.com
Sun Jul 17 04:40:39 EDT 2005
In the lab I have the following setup:
Spread_Segment 10.10.0.255:4803 {
P1 10.10.0.180
}
Spread_Segment 172.16.0.255:4803 {
P2 172.16.0.129
P3 172.16.0.126
}
Setup is identical on all 3 machines. Each segment is
on different LAN, connected through a router.
I have ~0.05msec latency within the LAN and ~0.6msec
latency across the router (according to ping rtt/2).
Field deployment will have similar characteristics
only with more segments (all connected to a private
backbone) and more daemons in each segment.
--- Yair Amir <yairamir at cnds.jhu.edu> wrote:
> Hi Yuval,
>
> Spread has some heuristic to understand your network
> and that does not work
> always well for people who use local addresses (e.g.
> addresses such as 192.168.0.x,
> 192.168.1.x ..). In such a case, there will be a
> change you will need to
> do manually that may improve things.
>
> E-mail your configuration file and we'll be able to
> comment. Explain also the
> latencies between the segments (and how they are
> connected).
> Cheers,
>
> :) Yair.
>
>
> Yuval Lifshitz wrote:
> > I have a spread network with two segments, one
> with
> > two daemons and one with one.
> > All is working fine except that for each message I
> > send through spread there are about 25 extra
> messages
> > sent between the daemons. These are small token
> > messages (most are token hurry) messages, but this
> 25
> > factor (with my high rate of messages) flood my
> > network.
> >
> > Spread is compiled with:
> > Token_timeout.sec = 20
> > Hurry_timeout.sec = 6
> > in the wide configuration the daemons use.
> > Is that a known problem or configuration problem?
> >
> > I tried to investigate the code and saw that the
> token
> > messages is sent in many places (sometimes twice
> in
> > the same place), so I tried to impose the
> > "Hurry_timeout" in an ugly way in the code, in
> > network.c I added:
> >
> > #include <sys/time.h>
> > ext sp_time Hurry_timeout;
> >
> > int Net_send_token( sys_scatter *scat )
> > {
> > struct timeval now = {0 ,0};
> > static struct timeval last = {0 ,0};
> > token_header *token_ptr;
> > int ret;
> >
> > gettimeofday(&now, NULL);
> >
> > if (now.tv_sec - last.tv_sec < Hurry_timeout.sec)
> > {
> > printf("too much tokens, time
> diff
> > from last is: %ld\n", now.tv_sec - last.tv_sec);
> > return 0;
> > }
> >
> > last.tv_sec = now.tv_sec;
> > ...
> >
> > Is that bad as I think it is?
> >
> > Yuval.
> >
> >
> > __________________________________________________
> > Do You Yahoo!?
> > Tired of spam? Yahoo! Mail has the best spam
> protection around
> > http://mail.yahoo.com
> >
> > _______________________________________________
> > Spread-users mailing list
> > Spread-users at lists.spread.org
> >
>
http://lists.spread.org/mailman/listinfo/spread-users
> >
> >
>
>
> _______________________________________________
> Spread-users mailing list
> Spread-users at lists.spread.org
>
http://lists.spread.org/mailman/listinfo/spread-users
>
____________________________________________________
Start your day with Yahoo! - make it your home page
http://www.yahoo.com/r/hs
More information about the Spread-users
mailing list