Kitely Forums

Try logging into the Kitely Welcome Center and teleporting to the merchants sandbox from there. We're currently investigating the issue that is causing some teleports to fail.

I'm out of time for running tests today. The tests that I ran took over three hours.

Not sure there is a significant advantage to running the same tests in the merchant sandbox when they've already been confirmed by a second party on a different region. While I admit that running these tests on a bare sim would possibly yield better results, no region is bare. "An empty sim is a happy sim" may be true... but none of us have empty sims. ;D

I ran some quick tests on Merchant's sandbox tonight. Didn't take very long.

In all intances of testing, textures rezzed as follows:

#1: 1024x512 2.3 seconds
#2: 256x128 1.5 seconds
#3: 512x512 1.6 seconds
#4: 1024x512 2 seconds
#5: 1024x1024 1.5 seconds
#6: 512x256 1.5 seconds
#7: 1024x1024 2.75 seconds

This was either directly from inventory, when applied to a prim, or pulled from a box. Figures were very similar.

So it's evident that an empty sim rezzes textures much faster than a populated sim... at figures that are quite acceptable. This indicates there may very well be a priority issue or bottleneck somewhere that is causing excessively slow texture rezzing.

Of interesting note, to help locate where the bottleneck may be:

* The test region for OSgrid was ElvenSong, which is one of the most heavily populated, heavily textured, and heavily scripted VARs on OSgrid. Yet texture rezzing times are competitive with SL.

* The test region for SL was Raglan Shire, home of the tinies, which again is one of the most heavily populated, heavily textured, and heavily scripted regions on SL. Yet textures rezzed in times equivalent to the totally empty Merchant's sandbox.

So my conclusion is there's probably an issue somewhere on Kitely populated sims-- an issue that doesn't exist on the other test grids. Something appears to be taking high priority over texture loading. This creates a user-experience issue because if you can't see something, you can't interact with it. This is especially troublesome to builders trying to figure out which texture to use-- and shoppers waiting for vendor textures to rez. Unfortunately I don't know what that issue/bottleneck may be... but perhaps this information will assist in tracking it down. If I may be of further assistance, please don't hesitate to beep.

Thank you for the additional information Snoots, we'll investigate this issue to see what's causing it.

*** appaws ***

To assist in verification, I ran some additional tests today. Same test... very different conditions.

On prior tests I went into three different grids (including Kitely) at a draw distance of 32m, into a sandbox area. I also included the Kitely Merchant Sandbox, which is for all purposes an "empty world". I totally cleared my viewer cache prior to entry of each area. So all prior data was formed on heavily minimizing texture and object impact on the viewer.

In this new series of tests I did just the opposite. I stood in the middle of a heavily-populated and heavily-textured area. I didn't clear cache, and I set my draw distance to 256m. So my viewer was full, impact maximized. In all tests I tried loading textures both DURING the region texture load cycle, and following loading of local textures (ctrl-shift-3 menu).

Rather than listing all the data test by test, I'm going to list averages for the sake of brevity. These figures are for texture loading directly from inventory.

OSgrid, Wright Plaza (a particularly laggy area filled with freebies): average 4 seconds

Second Life, RacerX's Giant Snail Racetracks (insane number of textures and builds): average 3 seconds

Kitely Welcome Area (no other avatars present): average 2 seconds (similar times to Kitely Merchant Sandbox)

Kitely Wellspring: average 7 seconds

Kitely Sylvan Retreat (a fairly light-duty world): average 4 seconds

(Note: I consider the difference between Wellspring and Sylvan to be a significant clue in our mystery hunt.)

Aberrations: On all grids, without exception, there was the occasional texture loading experience of 10 to 17 seconds. This would be considered a performance aberration that could be caused by anything from temporary excessive asset server access to Internet problems. These aberrant-behavior loads were discarded as extremes of the bell curve... as was the occasional texture that loaded instantly. I found the later to be of special interest though: if every once in a while a full-size texture (512x512) loaded instantly with almost zero delay... why is that? How did that happen? What was working right for once? This was indication to me that if all things were working right all the time (a "utopian" situation), all inventory texture loads would be instantaneous. Most likely instant texture loads happened at very rare instances when nothing else was impacting any server anywhere. Again, impossible to tell, but there were occasional (rare) aberrations on both ends of the spectrum... textures loading instantly and textures taking far too long. Standard bell-curve averages however, are indicated above.

Of importance in this however, is that on the prior test occasion when tests were run by another user at the same time, average Kitely texture load was 15-17 seconds whereas elsewhere it was appx 2 seconds. That tends to rule out the Internet itself as an issue and brings us down to grid and viewer software and performance. That there are often times when textures take an excessively long time to load is evidence that "aberrant behavior" can without notice become the norm. This is the area of concern and interest, namely: why does that happen? If this was rare it would be of no concern. That it happens often enough to draw attention is the source of this forum.

These figures bring several questions: Why did Kitely perform significantly better today? Were there fewer avatars on the grid this early evening and thus less impact on the asset servers? Was something fixed already? (Not to my knowledge.) All we have is the data... and the data is accurate. Of greatest interest is the inconsistency of information. The prior readings differ significantly from today's readings. That is a bit of a head-scratcher.

One thing that is evident: heavily-populated areas take longer to load textures from inventory than non-populated areas. That was proved in the prior test comparing the Kitely Merchant Sandbox to a populated world. But a texture load from inventory is a specific on-demand request to the asset server, so should have priority as a user-focus item (a user is directly and currently involved in the request). The same would hold true for a merchant vendor. When a script calls for a new texture load it is usually because a user is specifically focused on that item, and logically would demand high-priority response.

Just some thoughts on the texture loading process... and additional data to confuse the issue.

I took some tests on Kitely again today at about noon-thirty. Texture-from-inventory loading time was back up to 15 seconds average again... which verifies the erratic and inconsistent nature of this beast. As stated prior there were extreme ends of the bell curve. Very occasionally a texture would load in 2-4 seconds, but such was rare... and no indication as to why that particular texture loaded quickly (sizes of the textures I tested averaged from 512x512 to 512x1024). One texture that loaded almost instantly was a 512x1024, so there's just no immediately-observable explanation for that.

The vast majority of textures took 15 seconds to load today... the same types of textures that were loading in 7 seconds last night (under identical environment). I'm starting to wonder if something needs banged on with a hammer. ;D

Additional information regarding texture loading performance:

I was on Wellspring today trying to decide between eight textures on an object surface. As expected, it took about 15 seconds per texture to see each one on the object.

What surprised me was when I went back through the textures again. Instead of each texture appearing instantly as one would expect, it seemed as if each texture was being loaded from the asset server again. Each texture took significant time to load and rez, as if they hadn't just been loaded earlier.

I repeated this test several times. Each time the textures seemed to be reloading rather than being pulled instantly from my viewer cache. I'm not sure whether that's a viewer or world server issue (or both), but it's evident there is some cache failure there somewhere. My avatar remained in the same position, not moving a centimeter, so those eight textures should have been instantly available after the initial load.

If this is a universal trait, one can imagine the excessive bandwidth use and drain on the asset server in repeatedly and unnecessarily loading the same textures over and over throughout the grid.

Just an observation for whatever value. /me pulls out magnifying glass... the game is afoot! ;D

Hi Snoots,

Was the world already cached in your viewer? Were there other avatars in that world when you ran your test?

Ilan Tochner wrote: Tue Oct 27, 2020 10:19 pm Hi Snoots,

Was the world already cached in your viewer? Were there other avatars in that world when you ran your test?

Yup and nope. I'd been in the world for some time, was studying the object (side of a house) for possibilities, and I was the only avatar on the world.

As I've mentioned before, I don't know how the Viewer or region server works. But logically according to computer standards, when a user is fetching a texture from inventory, the system should 1) Check to see if that texture is already in the user's computer cache 2) If it is, instantly load it from that cache 3) If it isn't in cache, fetch it from the asset server or region server cache.

They should be in a cache somewhere, since the same eight textures were just fetched a few moments prior.

If the texture is in user cache, retrieval should take a fraction of a second. If it's in region server cache, retrieval should be fairly fast-- direct communication between the server and user. If it has to be fetched from the asset server it may take a bit longer as it would involve an extra step.

So in this experiment I wonder: 1) Why is it taking up to 15 seconds to fetch a single texture, at any time? (Texture fetching logically should never take that long) and 2) Why isn't it fetching recently-accessed textures directly from the user cache?

Perhaps in those two questions lies the path to the answer. (Granted, it might be a long, twisting, difficult path...) ; )