[Opensim-dev] Concerning recent grid instability and inconsistency

[Charles Krinke]

I guess I need to weigh in here a little bit and express a bit more of the testing opinion from the viewpoint of OSGrid.

We have a UGAI , some community regions (plazas), test regions I have set up (Yang & the moons), test regions others have setup & regions others have set up where they are using OpenSim.

In generaly, we want to keep the UGAI and the plazas up to svn as much as practical and reasonable for the developers. Also, we want to use only svn for the UGAI and the plazas. Others are encouraged to use other branches on test regions and compare their performance with svn, but we are trying to set the standard with the UGAI and the plazas. One of the primary short-term goals is to have a controlled environment to provide feedback to keep OpenSim moving in the direction of stability and reliability.

There have been and will still be times of instability. At those times we have 3 choices. 1) Go back to the last perceived stable revision 2) Stay at the current revision or 3) Move forward to the latest revision.

Each day when this decision is made, it is a balance to try to help us move forward in the most straight-forward manner. I cannot say that I always make the correct decision. Any input that the developers wish to make to effect choice 1, 2 or 3 on a regular basis will be given great weight. After all, we are trying to make it more reliable and at different times we need to adjust our update strategy for the benefit of OpenSim development and balance that for the greatest perceived stability in this major public grid.

This morning, the number of regions topped 200. The logins are running over 300 per day on average. Dozens of new accounts are added each day. There are 40000 assets in the mysql database. There is significant traffic on Wright Plaza on the order of dozens if different users.

So, we are beginning to deeply exercise the software and we find that in a heterogeneous network environment with a mixture of Windows and Linux servers spread over several continents running on pipes that go from cable modems to GigE interfaces is exposing data we would not normally see with quick stand alone tests after a commit.

I would urge us to use this resource more to wring out the details of how our software is performing as others begin deploying it in other, even less controlled setups. Please invent tests that allow all of us to compare Windows/Linux, differing revisions, inventory usage scenarios and others to allow us to make our software more robust.

Charles

----- Original Message ----
From: Brian Wolfe <brianw at terrabox.com>
To: opensim-dev at lists.berlios.de
Sent: Friday, February 29, 2008 6:48:48 AM
Subject: Re: [Opensim-dev] Concerning recent grid instability and inconsistency


Well, 
so 
far 
as 
one 
of 
those 
many 
that's 
handled 
the 
packet 
code, 
I've
been 
chasing 
down 
some 
inefficiencies 
in 
the 
login 
sequence. 
I 
could 
use
some 
more 
people 
trying 
out 
the 
UserThreadPoolBranch 
of 
the 
repo.

If 
you 
understand 
git 
then 
you 
can 
fetch 
a 
more 
recent 
version 
that 
was
just 
merged 
against 
trunk 
yesterday 
at
http://brianw.terrabox.com/OpenSim/OpenSim.git  
as 
branch 
UTPBranch 
(or
was 
it 
UserTHreadPool, 
brain 
wonky 
before 
first 
coffee)

It 
replaces 
the 
thread 
per 
user 
creation 
with 
a 
dedicated 
threadpool.
Some 
people 
report 
one 
impression, 
others 
another, 
andit 
works 
"great"
for 
me 
personally. 
So 
more 
eyes 
on 
what 
it's 
doign 
to 
syetems 
would 
be
appreciated. 
:)




-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://opensimulator.org/pipermail/opensim-dev/attachments/20080229/53af15c7/attachment-0001.html>