[vworld-tech] Grid/Cluster technologies and MMO server architectures

[ceo]

Glyph Lefkowitz wrote:
 >
 > On Apr 13, 2004, at 2:53 PM, Jeff Thompson wrote:
 >
 >> Has anyone looked into using Grid or Clustering technologies for use
 >> in MMO servers. I am aware of butterfly grid but didn't find much >
 >
 > I've looked at some documents describing the butterfly architecture, and
 > at the time (a few years ago) it sounded like complete garbage.  The
 > technical documentation was riddled with simple errors.  I was looking

 From what I've seen, they were really non-gamers who have entered the 
market thinking they have something special to make a lot of money with. 
This may well be true; certainly, their ability to write a good business 
plan, and to pursue it, is a generation ahead of practically all their 
competitors (although there are problems for their customers that make 
it in the opinions of myself and others rather like Garage Games' Torque 
used to be: lethally poisonous).

 > As I understand it, "grid" is simply a buzz-word for "thingy that
 > commoditizes hardware scalability".  I don't know of an architecture
 > that fulfills IBM's grand vision for the word.

AFAIAA it's mainly an alternative clustering scheme. For the rest of it, 
I'm not sure about their applicability to the games industry; I'm going 
to wait and see what happens. So far, despite 7-digit funding, they 
don't have any major customers. That's not impressive, for a product 
released more than a year ago, in a market that is supposedly desperate.

If anyone has more detailed knowledge, I'll happily bow to that; I only 
know of them as a competitor, and what I know from observing them in 
this capacity. I also met one of their founders incognito back when they 
started, but he wasn't exciting.

 > I'd like to add to the questions: does anyone have any luck applying
 > principles learned from academic literature on distributed operating
 > systems, or research code platforms, to this problem?

Yes, definitely. Although a large chunk of "modification" has to be 
done, much as with the conversion of academic AI routines in order to 
use them in games.

The most basic is that the traffic patterns that most dist-sys research 
is done for are n/a for most games. Digging deeper, the balancing act 
between different requirements tends to be weighted in different 
directions too (repeatability vs. throughput vs. 
burstability/sustainability vs. reliability etc).

But we gained more from outside academia - i.e. other industries - than 
from the academic papers. Even the "best of the best" of the papers I've 
seen never really go *far* enough with any concepts. It's all "OK, I've 
shown this kind of works, it clearly has potential. I'm stopping here", 
or else "This basic technique I've refined to exquisite perfection", but 
there's little that does *both* :).

Over time, we've got used to mainly using the latter: where we find a 
self-contained little non-trivial problem that could take months to 
solve satisfactorily, there's very often a researcher who's already done 
it. But I believe this is just the same approach that is generally taken 
in all industries :).

Disclaimer: we started with particular search paramters for the relevant 
research we were looking for. E.g. don't care about dist-obj arch's. For 
someone taking "a different path" you may find much more directly 
applicable works.

IMHO, academic research is definitely worth using. Especially since 
dist-sys research tends to be expensive (more in time than in hardware 
or etc) and difficult, so it's not stuff you want to be doing yourself 
if you can find prior art.

I would also suggest that patent applications are a damn good place to 
look for stuff. Some companies file excellent patents (e.g. BT), others 
file 3,000 words with not a single sentence worth reading (e.g. the guys 
behind World War 2 Online - they *really* suck at patent filings).

Of course, then you have to license it :). But it's probably better than 
spending months making something work, then thinking you've invented 
something yourself, but discovering 6 months later that it's already 
been patented and now they can name their fee and you've got to pay 
back-dated royalties and punitive damages :(. God, don't you just *love* 
the patent system?

Adam