Balashx
Joined 20 January 2008
Latest comment: 13 years ago by Peterl in topic Deletion: The Performance section of ‘Green Threads’
Deletion: The Performance section of ‘Green Threads’
editClaims in the section titled Performance, seemed counterintuitive so it prompted me to check the references. Reviewing cited references, I encountered quite a few factual inaccuracies, and since I was unable to locate a reliable source of performance measurements, I suggest this section be removed until such a time that a verifiable set of information becomes available. Here is the list of problems:
- The only reference that provides measured performance information is reference number 3, "Comparative performance evaluation of Java threads for embedded applications: Linux Thread vs. Green Thread" which is an unpublished or a self-published source .
- Reference number 3 states measurement were performed on Linux kernel version 2.2 where this Wikipedia article states the tests were performed on Linux version 2.4.
- Here are the software and hardware configurations stated in reference number 3
- CPU Motorola MPC 860 50Mhz
- Main memory 64 MB SDRAM
- Auxiliary memory 8 MB flash memory
- Cache memory 4 KB for instruction and
- 4 KB for data caches
- Communication 1 serial port and 1 Ethernet I/F
- Display 640x480 color LCD
- OS Linux version 2.2.13
- Java VM Personal Java version 1.1.3
- These hardware and software configuration are extremely out of date (over 10 years old) and hence irrelevant. OS Linux version 2.2.was released in 2000). Current version of JVM is 1.6
- Foot notes on page 2 of reference 3 states “so we patched the scheduler appropriately to support real-time scheduling.” It is highly unlikely that scheduler in version 2.2 of Linux behaves correctly after such a modification. Nevertheless, since operating system behavior was modified for these measurements the results cannot be generalized as performance results for readily available unpatched versions.
- This article does not specify what standard benchmarks were used for these measurements. Stating “Old benchmarks” is not sufficient.
- The first line under performance section states “On a multi-core processor, native thread implementations can automatically assign work to multiple processors, whereas green thread implementations normally cannot”. This statement is factually incorrect based on the official JVM documentation published by Sun Microsystem: http://java.sun.com/docs/hotspot/threads/threads.html. At least in Solaris there is an OS thread per each JVM thread and these threads are managed by OS.
- Reference 1 is a short article that incidentally contradicts the finding of reference 3 and reference number 2 is a discussion group.
--Balashx (talk) 22:51, 1 May 2011 (UTC)
- Thanks for this Balashx. This is a good summary, but it's in the wrong place. It should be on the talk page for Green Threads, so I'm copying it over there. Also, when you make a large edit like this (or in fact any edit), you MUST provide an edit summary. peterl (talk) 11:07, 3 May 2011 (UTC)