Zauber Paracelsus

LSL Memory Usage: Phlox vs LSO vs Mono

One of the big advantages for the Phlox script engine at InWorldz is that LSL scripts run under it use significantly less memory than they do under the LSO and Mono engines in Second Life.

For about two years now, I’ve been showing some memory usage stats that measure the memory use differences between Phlox, LSO, and Mono.  Those measurements, however, are from two years ago, when Phlox was still in beta testing.  So today, I’ve decided to take some fresh stats.

The original script was a proprietary one I wrote for one of my products, so I wasn’t inclined towards releasing it for others to confirm my results.  So, for this test I am using a public domain script I received back in 2008, a particle script based on code written by Ama Omega and Jopsy Pendragon.  This script is heavier than the original one I used, so the results will be much different than the original’s.

Memory usage is reported by executing llSay(0, (string)llGetUsedMemory()); at the end of the script’s state_entry event.  But, because LSO always reports memory usage as 16384 bytes (see notes below), I substituted llGetUsedMemory() with llGetFreeMemory() for LSO tests, taking the reported value and subtracting it from 16384 to get the correct value.

If you would like to test the scripts to see for yourself, I’ve packaged them up and made them available for download.  Here are the results:

Initial Test:

• Mono: 13308
• LSO: 3946
• Phlox: 3124

As you can see, running the script under Mono uses substantially greater memory than it would under LSO or Phlox.  But, this script can be stripped down to its bare essentials, by removing all the handling code for the particle system and stripping it all down to just a single call to llParticleSystem with a raw list.  To get the raw list, I took the processed list that is passed to llParticleSystem and fed it into iwMakeNotecard to get the raw list values, and plugged those in.

Optimized Test:

• Mono: 4388
• LSO: 1372
• Phlox: 473

In this test, the script is substantially lighter: 3x smaller under Mono and LSO, and  6.6 times smaller under Phlox.  But what about overhead, where each script has a certain baseline amount of memory that it uses?  That’s easy to measure.  The next test strips out the llParticleSystem call, and does nothing but report memory usage:

Minimum Test:

• Mono: 3364
• LSO: 221
• Phlox: 74

LSO and Phlox are both substantially smaller with this bare-minimum test, but Mono is only about a kilobyte smaller compared to the prior test, and is several times larger than LSO and Phlox.

So, as you can see, Phlox has much better memory usage than the Mono engine, though LSO’s memory usage is not actually much better (read the notes below).

Phlox is designed for high performance and efficient memory usage, allowing you to do more with your scripts without having to worry as much about lag.  Contrast to Second Life, where Linden Lab is attempting to impose more limits on scripts rather than trying to improve their system.

I honestly do hope that Linden Lab does turn around and try to make some common sense changes to their system, but continual disappointments have given me a rather cynical opinion of the lab.

Additional Notes:

• LSO Memory Usage: Above I noted that I had to use llGetFreeMemory and subtraction in order to obtain the memory usage for the LSO engine.  Linden Lab’s reasoning for this behavior with llGetUsedMemory is that all LSO scripts, no matter how much memory they are actually using, will always take up 16 kilobytes of memory.
• Dynamic Allocation: Phlox and Mono use a technique called Dynamic Allocation, where they only use the amount of memory they actually need.  So, instead of using up 64kb of memory for every script, Phlox and Mono scripts just use what they’re actually using, with 64kb only serving as a maximum on how much memory they can use.
• Bytecode Sharing: This is another memory-saving technique that Phlox and Mono use.  The “bytecode” is the compiled version of a script that gets executed and run.  If you have 20 copies of the same script rezzed out, the bytecode is shared amongst them, substantially reducing the memory used. LSO doesn’t use bytecode sharing, so 20 copies of the same LSO script will use the memory of 20 copies instead of 1 copy.

The Original Tests from 2010:

The original tests had different results than the one above: llGetUsedMemory was not available in InWorldz at the time, (and possibly not Second Life either), so llGetFreeMemory was used for all tests, and the subtraction was performed within the script code.  The testing script was a proprietary one for one of my products, so it will not be released.  I will note, however, that it was lighter than the one used in the initial test, and the optimized version of it was heavier than the one above test.  The minimal test was roughly the same, with slightly more memory usage due to the use of in-code subtraction.  Here are the results:

Initial Test:

• Mono: 9308
• LSO: 2220
• Phlox: 2113

Optimized Test:

• Mono: 5916
• LSO: 932
• Phlox: 632

Minimum Test:

• Mono: 3360
• LSO: 228
• Phlox: 80