January/February always seems to be a rockier time for the Second Life grid. The grid tends to behave quite unreliably through this period, though there never seem to be metrics that suggest that it is being placed under unusual user-loads during the unreliable stretches.
Nevertheless there's a lot going on under the hood as Frank Ambrose, Linden Lab's Senior vice-president of Global Technology reminds us. Ambrose outlined a number of Q1-deliverable initiatives for the Second Life grid and has provided a status report outlining where most of those are at, and noting a few additional, and unexpected difficulties.
LLNet seeks to move away from what are, ultimately, unreliable VPN appliances, instead securing communications directly between servers over whatever network medium is being employed (in this case, a big fiber ring). Unfortunately, a number of server-hosts kept trying to pull resources from the old network -- to their detriment. The Lab's having another look to see what's been overlooked.
Yes, waiting for things to rez actually is slower. Most everyone you've probably talked to in-world lately has been commenting on it, and the Lab has noticed the issue too. This appears to be a tricky one. The slowdown has occurred in several stages, each stage coinciding with the migration of infrequently used assets from the Isilon asset cluster to slower bulk-storage.
The peculiar thing that's going on here is that few or none of the things that are rezzing more slowly are actually among those assets that have been moved. That's a mystery that needs solving. It is possible, of course, that extra-decision-making has been introduced into the pipeline for looking-up and retrieving assets. Fractions of a second in delays may not seem like much, but when you're doing something 50,000 times each second, tiny delays tend to add up.
Additionally databases continue to be stabilized and optimized. If anything, the grid should be feeling faster. It just ... well, isn't. We love a good mystery, though, and we're looking forward to finding out just what the culprit is, in this case.
Ambrose also says that a new team has been formed to evaluate the existing architecture and infrastructure (hardware and software) and to propose improvements, replacements and restructuring where necessary. Actually, this comes as a bit of a surprise to us, and to quite a number of other users, as we had long been under the impression that such a team already existed.
There's no additional word on the status of the Agent Inventory Service, destined to improve inventory reliability and performance. It may well be contingent on the LLNet project which, as mentioned above, has run into a rocky patch.
|Are you a part of the most widely-known collaborative virtual environment or keeping a close eye on it? Massively's Second Life coverage keeps you in the loop.|