[Spread-users] Many packet retransmited in WAN
tryteks at pit.edu.pl
Wed Jun 14 02:25:21 EDT 2006
Many thanks for valuable hint. It was definite problem with router
It is enough to place that two traffics in one fifo queue.
But it has one disadventage - token is sent with the same priority as data
So I have one more question as you have experience with WAN deployment.
For example, throughput degradates to a minimum (and spread load is high)
at one channel for a longer time and
having all messages (token , data, ...) in one queue daemons
are unable to switch to operating state, and of course the system is
unable to send any messages.
And that state is going to last until that channel is working fine.
Does exist any idea how to deal with such an unstable channels?
On Tue, 13 Jun 2006 22:49:31 +0200, John Schultz
<jschultz at spreadconcepts.com> wrote:
> Stanisław Trytek wrote:
>> Interesting thing is
>> that the daemon B receives and delivers the first message and ask the
>> daemon A to retransmit next 9 packets. Next, having again token, daemon
>> B delivers the second message and asks about 8, and so on ...
> It seems like one or more of the routers is delaying the data packets
> from daemon A. In your log files you can see that daemon B is getting
> one of the data packets per round for a while but then it gets a ton of
> late retransmits of the same packets. For some reason, it seems that
> your data packets are getting delayed while your token packets are
> actually making it through as normal. When a daemon has the token it
> sends its data messages and then it sends the token. Data and token
> traffic occur on two seperate ports. It is generally assumed that if
> the packets are not dropped along the way, then the data packets will
> generally arrive before the token packet. For some reason this general
> assumption seems to be *grossly* wrong for your environment. It might
> be that your routers are being "smart" and deciding that too much
> traffic is occurring on the data channel port and therefore giving
> priority to the token channel port traffic. As you can see, this can
> cause major problems for the ring protocol in such a small configuration.
>> Why daemon B delivers only one message per token round? Is it desired
>> Is it possible to force spread to deliver more than one message per
>> token round?
> No, this is *not* desired behavior and is almost certainly an artifact
> of something strange either in your network or your machines, as we've
> had plenty of WAN deployments that worked fine in the past.
> John Schultz
> Spread Concepts LLC
> Phn: 301 498 3233
> Cell: 443 838 2200
> Spread-users mailing list
> Spread-users at lists.spread.org
More information about the Spread-users