[Embedlets-dev] Re: JStamp vs TINI bytecode execution speed?

[Andrzej J. T. <an...@ch...>]

Chris pointed me to James' uVM benchmarks on the JStamp list (now why 
didn't I remember to look there?).

From a quick review of James's testing it looks like a TINI is from 15-20 times 
slower than a JStamp.

This means that a TINI will be running around 150-200K bytecodes per 
second, as compared to the 3M that a JStamp can do, and 15M for a JStik.  
What's interesting is that uVM is is about 1/2 the speed of a JStamp (for the 
20mhz PIC) or about the same (for the 40mhz PIC chip).  

Granted, these tests were simple ones that would not reflect a "real world" mix 
of bytecodes being executed, but their probably close enough.

What is intriguing is that according to the above, the TINI will be the least 
capable target platform (based on processor speed alone and ignoring JVM 
features and memory) that we are considering.  Bit of a revelation for me, 
since I had just assumed that the PIC's running uVM would take that position. 
Goes to show you....always check your assumptions.

Kind of leads me to believe if we can implement Embedlets on a TINI and have 
enough horsepower to do some "useful" applications on that platform, then the 
rest will be no problem.



Andrzej Jan Taramina
Chaeron Corporation: Enterprise System Solutions
http://www.chaeron.com