aboutsummaryrefslogtreecommitdiff
path: root/pipermail/nel/2001-February/000212.html
diff options
context:
space:
mode:
Diffstat (limited to 'pipermail/nel/2001-February/000212.html')
-rw-r--r--pipermail/nel/2001-February/000212.html82
1 files changed, 82 insertions, 0 deletions
diff --git a/pipermail/nel/2001-February/000212.html b/pipermail/nel/2001-February/000212.html
new file mode 100644
index 00000000..2b9cec97
--- /dev/null
+++ b/pipermail/nel/2001-February/000212.html
@@ -0,0 +1,82 @@
+<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 3.2//EN">
+<HTML>
+ <HEAD>
+ <TITLE> [Nel] Dynamic load balancing?</TITLE>
+ <LINK REL="Index" HREF="index.html" >
+ <LINK REL="made" HREF="mailto:archer%40nevrax.com">
+ <LINK REL="Previous" HREF="000209.html">
+ <LINK REL="Next" HREF="000213.html">
+ </HEAD>
+ <BODY BGCOLOR="#ffffff">
+ <H1>[Nel] Dynamic load balancing?</H1>
+ <B>Vincent Archer</B>
+ <A HREF="mailto:archer%40nevrax.com"
+ TITLE="[Nel] Dynamic load balancing?">archer@nevrax.com</A><BR>
+ <I>Mon, 19 Feb 2001 17:04:15 +0100</I>
+ <P><UL>
+ <LI> Previous message: <A HREF="000209.html">[Nel] Dynamic load balancing?</A></li>
+ <LI> Next message: <A HREF="000213.html">[Nel] Dynamic load balancing?</A></li>
+ <LI> <B>Messages sorted by:</B>
+ <a href="date.html#212">[ date ]</a>
+ <a href="thread.html#212">[ thread ]</a>
+ <a href="subject.html#212">[ subject ]</a>
+ <a href="author.html#212">[ author ]</a>
+ </LI>
+ </UL>
+ <HR>
+<!--beginarticle-->
+<PRE>According to EagleEye:
+&gt;<i> I would love dynamic load balancing... just add another server to your
+</I>&gt;<i> cluster, and assign it a piece of the game world to take care of... perhaps
+</I>&gt;<i> that city area that just got really popular in the last few weeks...
+</I>
+That's not dynamic load balancing if you have to assign servers to a
+specific area. That's merely 'reconfigurable static' :)
+
+Ultima Online did this for their servers. They measured during beta
+a lot of statistics, and created a map of areas for allocation to all
+processors (server being ambiguous here, because their servers were
+multiprocessor boxes). That map could be redone at will, depending on
+the number of processors allocated to a shard. But once the shard was
+started, that was it. The only way to add a processor would be to
+split one area in two, and the whole migration process was hard
+enough that it wasn't worth (their words) the development cost vs the
+number of times it would be used.
+
+True dynamic load would mean that each processor would adjust its
+border (i.e. which objets it manages) according to its load, compared to
+its neighbours. If it has too high a load, then its border would shrink
+along its most lightly loaded neighbour, swapping objects between one
+and the other.
+
+Of course, that means a variable, almost fractal geometry of processors,
+and a processor that, at boot time, was serving city A and its adjacent
+areas could well, four days later, be serving chiefly the mountain range
+5km from there, because of simple &quot;pressure&quot;. It looks nice from a
+strict research standpoint, but it's HELL for system management.
+
+--
+Vincent Archer Email: <A HREF="mailto:archer@nevrax.com">archer@nevrax.com</A>
+
+Nevrax France. Off on the yellow brick road we go!
+
+</pre>
+
+
+
+
+
+<!--endarticle-->
+ <HR>
+ <P><UL>
+ <!--threads-->
+ <LI> Previous message: <A HREF="000209.html">[Nel] Dynamic load balancing?</A></li>
+ <LI> Next message: <A HREF="000213.html">[Nel] Dynamic load balancing?</A></li>
+ <LI> <B>Messages sorted by:</B>
+ <a href="date.html#212">[ date ]</a>
+ <a href="thread.html#212">[ thread ]</a>
+ <a href="subject.html#212">[ subject ]</a>
+ <a href="author.html#212">[ author ]</a>
+ </LI>
+ </UL>
+</body></html>