[Spread-users] spread delay problem

[John Lane Schultz]

I don't think Spread tracks how long it takes for the token to circulate.
Besides, the token circulation time can vary drastically depending on how
much data traffic your daemons are sending as tokens are sent AFTER data
traffic.  If you instrument Spread to measure the token circulation time,
then make sure you only use these measurements from runs where you are
heavily loading the system with data traffic (e.g. - users on every daemon
sending very fast).

You can put an estimate on the token circulation time by taking the point to
point latency on your network, multiplying it by the maximum number of
daemons you ever expect to be running together at once and then multiplying
that by a fudge factor of 10-100 to allow for processing, loading and bulk
data transmission time.

Then test your parameter by running the largest configuration you expect and
having users on every daemon sending very quickly.  If you get partitions,
then you need to up the parameter significantly.

If you do your testing iteratively and you find the timeout right around
which the system starts becoming unstable under heavy use, then take that
timeout and multiply it by a factor of 3-10 and that should probably be
good.
 
Cheers!

---
John Lane Schultz
Spread Concepts LLC
Phn: 443 838 2200 

-----Original Message-----
From: Adrian Revill [mailto:adrian.revill at shazamteam.com] 
Sent: Friday, November 07, 2008 9:20 AM
To: John Lane Schultz
Cc: spread-users at lists.spread.org
Subject: Re: [Spread-users] spread delay problem

Hi John,

I have tried the Hurry_timeout at 100ms and the pauses have now gone to 
100ms. To help me tune the value, is there any way i can see the actual 
time it takes for a token to rotate?

An odd thing though, before i tried the lower time-out, i re-ran my 
tests, but forgot to remove the PROTOCOL logging. All the pauses changed 
to approx 300ms. As soon as i turned off the logging, they went back to 2s



John Lane Schultz wrote:
> It definitely sounds like you are losing the token.  You can use spmonitor
> to connect to the daemons and see what kind of losses they have.  You can
> also lower the Hurry_timeout, set at the top of membership.c, to reduce
how
> long the pauses from a loss of token take.  On a LAN you could try
something
> like 10-100ms depending on the size of your configuration (bigger ->
> longer).
>
> Cheers!
>
> ---
> John Lane Schultz
> Spread Concepts LLC
> Phn: 443 838 2200 
> -----Original Message-----
> From: spread-users-bounces at lists.spread.org
> [mailto:spread-users-bounces at lists.spread.org] On Behalf Of Adrian Revill
> Sent: Thursday, November 06, 2008 11:39 AM
> To: spread-users at lists.spread.org
> Subject: [Spread-users] spread delay problem
>
> I found a similar problem to mine on these lists 
> http://lists.spread.org/pipermail/spread-users/2004-December/002367.html
>
> Did anyone ever get a solution?
>
> regards
>
> Adrian
>
> ______________________________________________________________________
> This email has been scanned by the MessageLabs Email Security System.
> For more information please visit http://www.messagelabs.com/email 
> ______________________________________________________________________
>
> _______________________________________________
> Spread-users mailing list
> Spread-users at lists.spread.org
> http://lists.spread.org/mailman/listinfo/spread-users
>
>
>
> ______________________________________________________________________
> This email has been scanned by the MessageLabs Email Security System.
> For more information please visit http://www.messagelabs.com/email 
> ______________________________________________________________________
>   

______________________________________________________________________
This email has been scanned by the MessageLabs Email Security System.
For more information please visit http://www.messagelabs.com/email 
______________________________________________________________________