[Tutor] Python on network problems

[Carroll, Barry]

Hello, Diana,

I have had a little experience with networking problems.  Here's my take
on your situation.  

Regards,
 

> Date: Tue, 15 Aug 2006 11:19:35 +1000
> From: Diana Hawksworth <dianahawks at optusnet.com.au>
> Subject: Re: [Tutor] Python on network problems
> To: Alan Gauld <alan.gauld at freenet.co.uk>
> Cc: tutor at python.org
> Message-ID: <200608150119.k7F1JZ9O022639 at mail31.syd.optusnet.com.au>
> Content-Type: text/plain
> 
<<snip>>
> 
> We are running Windows XP.  The students log in to a network server
that
> allows them access to their user accounts as well as various group
> folders.  We have 4 rooms full of computers - but Python is installed
on
> each workstation in this room only ie, it is "active" in this room
only.
> It is not on the network and no other computer rooms have access to
it.
> The students in this room save their Python files to a networked
"user"
> account - as they do with all their other files.
> 

>From your description, I gather that a single computer provides
authentication service and file service to the entire network.  Is this
correct?  If so, that machine is a single point of failure in your
network.  If it hangs for any reason, the entire network "goes away".  

> Two failures have occurred.
> 
> The first has been intermittent and to individual students only at
rare
> times.  A student will attempt to start IDLE - and nothing will
happen.
> Sometimes, changing that student's log in user name will solve the
> problem, and he then has access to Python again.
> 

I don't haven't used IDLE that much, so Alan's ideas are more reliable
than mine here:

>> That could be a local machine/user setup issue.
>> Given the various issues with IDLE and firewalls I might
>> look at the firewall settings on the Python PCs and make
>> sure IDLE is happy with them.

> The 2nd is more pervasive, and that is, whenever I have the class
working
> with Python - the entire school network becomes inoperable, and the
system
> administrator needs to reboot it again. Because we have been working
on
> Python each time this has happened, Python is being blamed for the
system
> failure. Inoperable - no one is able to open any files in any program,
> save any work they have been working on, open files - or even log in
if
> they haven't already. The whole system is " frozen", from the office
to
> every other computer in the school.

> Now I am inclined to think it is not a Python problem at all.  Python
is
> locally and individually installed on computers in this room only.  It
has
> no network access at all  - apart from files being saved in networked
user
> accounts.
> 

>> The only slight possibility is if any of the students are
>> writing code that accesses a networked folder, to open
>> a file say, and is causing some high volume network traffic. 

I think Alan is right here, too, and here's why. (Python internals
gurus, please correct any bad assumptions in the following.)  Lab #4 is
full of users running Python programs.  The interpreter is running on
each local machine, but the program (and data) files are all stored on
the lone network file server.  Since Python programs are executed by the
interpreter, they access file storage more heavily that compiled
programs do.  So, every time any of the (15, 20, 25?) computers in the
Python lab needs a file, or needs to save a file, it sends a request to
that one computer.  The potential bottleneck is obvious.  

I think the network server's capacity is being exceeded in some way.
When this happens, instead of degrading service gracefully (e.g.
rejecting a request and instructing the client to try again later) the
server is getting lost somehow and hanging, causing the failure modes
you describe.  

Here are some areas for your network administrator to investigate:

   * maximum number of concurrent disk access requests, 
   * maximum number of open files,
   * file server disk usage,
   * file server disk fragmentation,
   * server processor utilization, 
   * network bandwidth utilization,
   * other limits imposed by the networking software or the 
     server's hardware.

> What I would like to know is if anyone else has had a similar problem
-
> and it has been proven that Python is the cause?  If so - how was the
> problem solved?
> 

The best solution (IMHO) is also the most expensive: add a redundant
server to your network.  This would eliminate the single point of
failure weakness and reduce the load per server.  Also, is one server
should hang or fail for any other reason, the duplicate server can keep
the network running while the failed machine is rebooted, repaired, etc.


If possible, I would recommend using a Unix clone as your server's OS
instead of Windows.  Even Linux, which is available free of charge,
provides a more stable platform for a network server than does Windows.
This is relatively easy, especially if your network is based on TCP/IP.
Many Windows intranets use Unix/Linux servers.  The biggest obstacle, in
many cases, is the learning curve required of the network administrator.


> Be grateful for any advice - and hope I have made the situation
clearer
> now.
> 
> Diana
> 

HTH

Barry
barry.carroll at psc.com
541-302-1107
________________________
We who cut mere stones must always be envisioning cathedrals.

-Quarry worker's creed