[vworld-tech] Maximum Latency?

Bruce Mitchener bruce at cubik.org
Thu Feb 19 11:46:22 PST 2004

Jim Purbrick wrote:
> What are the average and maximum latencies graphical virtual worlds should
> expect to deal with?

I don't have an answer for this directly.

But, which latency are you talking about?  Latency is an end-to-end 
problem and can happen at any point along the way:

    * Client is slow because of hardware, anti-virus, Outlook checking
    * User's network connection is congested/overloaded.
    * Network somewhere between the user and you is congeted.
    * Your network connection is congested.
    * Your intra-cluster communications are congested.
    * Your server is running slow for any of a thousand reasons,
      not all of which are due to your code.
    * Lots more reasons!

Some of these you can control, some you can't. Some you can detect, some 
you can't.  Some will impact everyone, others won't. :)

I do think that this is a good reason for having as much per-task 
accountability as possible.

In some current server work, I'm working on making every task 
transactional (both its in-memory side effects as well as disk and 
network I/O).  This will give us the ability to have our scheduler and 
transaction monitor systems report detailed statistics on any given task:

    * Time to execute
    * Amount of disk I/O
    * Amount of network I/O
    * Number of objects/attributes modified
    * Possibly number of objects/attributes read
    * Whatever else we can get

This should give us a good idea of the work being done by any given 
task.  With the right annotation, it might even be possible to track 
resource usage information to particular game systems or types of tasks.

We'll have to see how well this works for us, what use the transactional 
stuff is in practice, and what the stats end up meaning.

  - Bruce

More information about the vworld-tech mailing list