• All submissions to this site are governed by Second Life Project Contribution Agreement. By submitting patches and other information using this site, you acknowledge that you have read, understood, and agreed to those terms.
Issue Details (XML | Word | Printable)

Key: SVC-3344
Type: New Feature New Feature
Status: Reopened Reopened
Priority: Normal Normal
Assignee: Unassigned
Reporter: Vex Streeter
Votes: 8
Watchers: 2
Operations

If you were logged in you would be able to see more operations.
2. Second Life Service - SVC

Automatically migrate Open Space sims to be co-resident with other Open Spaces that have similar resource demands

Created: 29/Oct/08 09:59 AM   Updated: 30/Oct/08 05:25 AM
Return to search
Component/s: Performance
Affects Version/s: None
Fix Version/s: None

Issue Links:
Duplicate
 
Relates
 


 Description  « Hide
Background:
  • Open Space/Void/Quad sims compete directly for CPU resources with 3 others, leading to issues where over-use of one causes significant problems on the other co-resident sims.
  • The recently announced Open Space SIM repricing is reacting to the over-use ("abuse") of OS Sims by upping the price to match the aggregate average use.
  • IMHO, this will have exactly the opposite of the intended effect, driving out those who do not require high resource use and retaining the abusers. The latter group will still see Open Spaces as a bargain, and the former were using such sims in the way intended - a relatively inexpensive way to have light-use decorative sims requiring full space but few resources. Even those light users who pay the additional tier will be dissatisfied as their low resource sims will be sharing a cpu with (more than likely) 3 high-use sims.

Rather than attempting to fix the real problem with a repricing approach, I propose to solve the underlying problem using basic load-balancing algorithms in reverse:
1. Meter all OS Sims for a load factor, which is a combination of measured avatar traffic, script load, physical object count (or maybe just colliders?), and prim load (maybe only temprezzed?).
2. Compute a longer time load factor determined statistically. I'd suggest something like a max load for the previous 10 days.
3. keep database of sim and current average load.
4. on sim reboot, choose target based on minimum delta between "my" load factor and the average of other sims on the target.
5. force migration if the disparity of load factors grows too large (this would be per machine or per-cpu to avoid chaotic migration).
6. increase visibility of load factors, in viewer and LSL. Also make it possible for the viewer to see the names and load factors of co-resident OS sims. (optional, but helps people understand why they are migrating)
This sounds pretty straightforward and can be rolled out in phases. It will take some time to determine the correct factor values.

This puts everyone in control of their own destiny: Everyone rises (or sinks) to their level of resource use, and tolerates (or not) living with their peers.

  • Too much lag? reduce your own resource use and you'll get migrated down.
  • Want to open a round-the-clock club on an OS sim, go for it! but be you'll be sharing the CPU with 3 other clubs.

Note that this proposal very clearly changes the definition of "open space" sim somewhat. One could easily imagine sims at the top levels choosing to upgrade to a "half sim" and the ones at the bottom downgrading to an "eighth sim", rather like tier on mainland.



 All   Comments   Change History      Sort Order: Ascending order - Click to sort in descending order
Beware Hax added a comment - 29/Oct/08 05:31 PM
duplicates SVC-3344
(when i proposed the idea, someone else already had it earlier, but well hidden away (meaning, not directly linked from MISC-1776))

Beware Hax added a comment - 29/Oct/08 05:32 PM
resolved the wrong one. sorry, im being awake for too long already today.

Beware Hax added a comment - 29/Oct/08 05:38 PM
my version, made up independently, now posting it here. ultimately it means the same thing.

i think only a percentage of the openspace sims is actually offending (meaning: using enough resources to lag other openspace sims on the same server)

i propose a plan which:

  • saves the homes
  • punishes the clubs
  • doesn't increase costs for LL
  • doesn't increase prices for (innocent) users
  • does not change the original idea behind openspace sims (meaning: light use only)
  • needs very little development effort

here goes: measure the resource usage of openspace sims, possibly averaged over a timeframe (a week?), divide the statistic in a number of classes, for example:

90-100th percentile: worst offenders
70-90th percentile: moderate offenders
0-70th percentile: not offenders

warn offenders that if they don't fix their sim, they will be moved to a server with only other sims in the same class. then they punish each other, and non-offenders can live peacefully without lag.
after having moved, they can move back on request, and paying a fee, if they fixed the problem (closed that club they were running, for example)

combine this with publicly showing in the client that a particular region is openspace sim and that it lags because its using too much resources.

also, offer a "half sim" - 2 on a core, $150 per month, and suggest offending openspace sims to upgrade to one.


Vex Streeter added a comment - 29/Oct/08 07:44 PM
@Beware - you've certainly got a similar idea, with (hopefully) similar results. I intended my proposal, however, to be fully automated based on objective measures of resource use. Questions on your proposal:
1. I think you are making some assumptions about the percentage offenders there are - I suspect you are correct, but I'd worry about the subjective nature of the measures. In fact, why categorize at all? There are plenty of OS sims out there to support a lot of performance levels.
2. I'd not use subjective terms at all - if 90-percentiles only compete with each other, they're not really hurting anyone...
3. your procedure appears to require a human in the loop - this sounds expensive (and gets LL employees yelled at). Why not just automate?

I think the half-sim is a good idea, but maybe belongs in a different entry. I think with a proper automated self-organizing network, you could also step down and offer 8ths and 16ths. More information on the overhead of sim machinery is needed however - you really don't want to run 16 copies on one CPU - better to have one simulator process running a 16 logical areas..


Vex Streeter added a comment - 29/Oct/08 08:05 PM
Minor changes to make clear that this is a serious technical proposal and not just grousing - the misc issue scares me with too many complaints without productive comments.

Beware Hax added a comment - 30/Oct/08 05:21 AM - edited
yes your version is cleaner/better. however, interference between sims on different cpu cores is known to happen (with full sims), so i say openspace sims compete with 15 others, not 3 others.