[MUD-Dev] The Lag monster...

[Harlan Beverly]

--- Kiztent Hatepriest <kiztent at gmail.com> wrote:

> I don't know about ideas, but I know that in my experience,
> network performance isn't the problem.  Usually when I play it's
> rendering and loading information from disk into memory (lots and
> lots of memory).

> I get good performance at 14K/sec (too far for ADSL, no cable),
> and that barely touches the network inside my firewall.

> I think Amanda's question is really valid.

I'm familiar with bottleneck theory.  The theory goes that making
changes other than to the bottleneck will not cause ANY marginal
improvements.

Thus the definition of bottleneck precludes the possibility that ANY
inprovement in anything else will cause ANY improvment.  I do not
think that is the case in the lag problem.

For example, would it not be beneficial for a server to aggregate
state updates and pre-compute some interaction, then at a regular
rate, update the server program?  This would give less time for a
server to deal with the interrupts of packet collection and parsing.
Further, packet filtering for stale data could occur reducing the
need for unneccesary system interrupt.

If any improvement is possible (even marginal), then the bottleneck
is either a parallel path, simultaneous bottlenecks, or the TWO
components that actually work together to cause 1 bottleneck.

In systems as complex as computer, I believe it is the last
scenario.  The components of internet traffic (particularly at the
server) interacts with overall system capabilities creating a
dependancy situation that broadens the variables apon which the LAG
problem can be ameliorated.

The question becomes, to what effect would adjusting the internet
variables (such as throughput, latency, interrupt frequency, PCI bus
utilization, memory copy operations, etc.) cause an improvement in
percieved lag of the users?

Further, would any adjustments at the client machines improve the
system lag as percieved by the users?

Any other thoughts are fully appreciated.

Harlan
_______________________________________________
MUD-Dev mailing list
MUD-Dev at kanga.nu
https://www.kanga.nu/lists/listinfo/mud-dev