recently i’ve mentioned some Second Life paradigms or shackles that are slowly falling by the virtual wayside for me
some are easy to grasp and push aside, like prim limits and parcels. with OpenSim many of us don’t think in terms of parcels as far as what you own – now we think in terms of regions (although parcels are great for defining areas, especially for audio or student sandboxes)
in trying to complete a few partially made OARs, i’m hitting what i call the Cool Ice Campus which has a floating amphitheatre. i decided to make semi-schmancy seating for it
once the seat was made, it was time to place them and this is were i came across yet another paradigm -
how many should this build be able to hold?
in Second Life, i made a few similar builds but it was easy to decide how many seats to use. a “good” sim could house 55 avatars (depending on attachments - i remember some big real world band holding a concert and some Lindens were on the scene and asked everyone to remove their hair so that a few more people could attend!) =D
OpenSim changes that because concurrency depends on hardware, connection speed, and efficiencies in deployment
Kitely is an example with their advertised 100 avatar limit which opens the door to thinking about what advances will possibly happen in the next year or so. it’s conceivable that this campus could be in use for a year or two, so how many seats should be placed in it?
i like a lot of seating so that people can spread out but 100 seats isn’t an excess, not in Kitely anyway
the image below is with 144 seats but i’m thinking of adding more, there’s plenty of space for them
if you make public spaces, how many people do you build for?
I tend to take concurrency numbers with a pinch of salt but in principle you could design for 400 if the build overlapped the corners of four adjacent sims.
graham mills
11 Jan 13 at 12:52 am
I think it is worth noting that Kitely uses standard OpenSim running in a ROBUST configuration but replaces some key components with its own proprietary cloud-based systems. Kitely’s distributed assets and inventory systems provide vastly superior performance, reliability and scalability compared to regular OpenSim.
Kitely has also developed a unique Advanced Megaregions system, which enables creating worlds up to 16 regions in size that behave as one big region and don’t have the limitations of regular OpenSim megaregions.
The combination of all of these unique improvements means that your user experience while using Kitely will be much better than what you can get from OpenSim service providers using unaugmented OpenSim.
Ilan Tochner
11 Jan 13 at 4:28 am
i think that may be some SL thinking mixed in graham because Kitely can’t have four adjacent “worlds” – the 100 avatar limit is per world and not per region
now inSL that was a strategy although it was hard to get more than 200 to work like that and of course would cost an initial $4000 plus 295 a month each if you kept them
LL does this regularly for things like their snowman competitions (do they still hold those?)
“user experience while using Kitely will be much better than what you can get from OpenSim service providers”
i can vouch for this statement with a few caveats. firstly, i reserve my use words like “much” and “very” as i tend to overuse them as do many people (the news media is the worst at hyperbole on everything)
Kitely is better than SL in several areas – cost, performance, prim limits, concurrency, ability to back up and export/import, access control, use of additional TOS for your own work, ability to use multiple account methods, website design and functionality, and other things not coming to mind
Kitely earns my opinion of being much better particularly in performance, prim limits, and concurrency
Kitely earns a “holy cow!!!” from me on cost =)
Kitely earns a *confused, can’t compute* on the megaregion front because there isn’t anything to compare it to – the ability to have a 16 region single megaOAR trashes the “region” frame of reference (the confused rating is a very, very good rating, btw)
i do have to say i was happy with SimHost and in some ways it has advantages over Kitely, but for us, Kitely’s numerous pluses simply outweighed any others
Kitely is phenomenal and it is easy to take for granted how good it is and its value is ridiculous, especially when compared to SL
even at double it’s current costs, Kitely would still be an outrageous value and still be our choice
okay, am i a fangrrl of Kitely? yes, but they earned it and anyone who’s read me over the years knows that i will also call out bad service and poor performance
Kitely continues to exceed my expectations =)
Ener Hax
11 Jan 13 at 7:42 am
I suspect you are right, ener. My bad.
I think I assumed that a 4×4 in non-megaregion configuration would take 100 avatars per region but I vaguely recall asking that question previously and getting a negative response.
The SL-based VWBPE conference has routinely used a 4×4 like that, a very atmospheric steampunk build: http://www.flickr.com/photos/ryanrasmussen/5540448304/in/photostream
Graham Mills
11 Jan 13 at 9:24 am
BIG thank you Ener :-)
The reason I emphasized the aforementioned technological advantages Kitely has compared to regular OpenSim is that people who know OpenSim’s limitation might doubt our performance and scalability claims if there aren’t aware that we have done a lot to overcome those limitations. The improvement is very tangible, especially when user concurrency goes up.
Ilan Tochner
11 Jan 13 at 11:23 am
Not to put a damper on things and in no way as a slight against anyone for anything but a point or two to ponder.
Ages ago, we used to try and do Load Testing on OSgrid. Proper Load Testing with operational & loaded (dressed, prims attached like hair and shoes, scripted obj or two) avatars, (bot’s don’t count). Unfortunately wrangling together that many avatars at any point in time was even difficult there.
It would be nice to see a proper load test & comparison done side x side with 25, 50, 75, 100 avatars. To see how the Server Numbers crunch down, how the client lag is affected and then general things like inventory response’s, messaging etc… These could be an invaluable baseline set of tests which could be compared as things progress and change. Other things that would need to be recorded too is where (physical RL City location) the avatar is logging in from, as the number of hops (distance x relays) will affect some of the performance metrics and need to be accounted for. As we know, someone at a greater distance such as Japan logging into a server in the USA will have a different response than a person logging in from New Zealand, Germany or the USA itself. In addition, it would further a bit on the Marketing Side if such tests were published & available to John Q Avatar showing tangible results derived from a structured & proper test suite.
The tests would need to be done for a few different configs such as:
* Grid with Robust only (dedicated commercially hosted)
* Standalone operating on a commercially hosted dedicated server.
* Grid on Cloud with Robust, another like Kitely with their Robust +
NB The first two could be done on same machine with two separate configs, with only ONE version running during the tests. Using a dedicated hosted server versus a home run system is to provide a proper & correct connection to the web.
Done properly with a fixed set of guidelines for the test, meaning that everyone would need to set Max Viewer speed, draw distance etc to be reasonably uniform for a solid baseline evaluation and then to repeat the testing in cycles. Server settings and configuration would also have to remain reasonably consistent as well to determine that same baseline for server deployment.
I’m not entirely sure it would be even feasible to try and assemble 100 willing human’s to do these tests, let alone follow guidelines in regards to their viewer settings / configuration. Even if it is for everyone’s advantage to participate in such exercises to help & improve & benchmark those improvements.
In regards to Bots: The do not qualify for legitimate testing because they are not carrying inventory and attachments which are driven by a viewer for testing. Viewers send data back & forth constantly resulting in network traffic load and subsequently contribute to lag & other performance issues. Because the Bot is “server side” it cannot generate the same types of traffic & metrics needed to make a proper evaluation of the efficiency.
Bot Testing “could” be an additional test to see what the weighting on a server & it’s resources are and how / what the impacts of using bots on a server are in actuality.
WS
WhiteStar
11 Jan 13 at 11:30 am
@Whitestar I tend to agree — hence my reference to “pinch of salt”. Happy to help with any testing.
Graham Mills
11 Jan 13 at 11:49 am
i like what WhiteStar is saying! =)
but . . . that might be more possible with bigger organizations (Intel) or under an initiative by Overte (once it is up)
Ener Hax
11 Jan 13 at 3:30 pm