[gclist] Garbage collector efficiency question...
Tue, 18 Nov 2003 23:13:28 -0700
I am a student currently working on my research report on incremental GC
techniques in real time systems.
As a goal of the study I list the items/issues to consider which helps a
user choose an algorithm he/she could use as a potential memory manager
for his application.
As I was working through the list of the issues, I had an idea, and I
want to share with you and check the feasibility of it.
Work has been done in Java in which the user can choose the collector he
wants to run with his real-time application.
Now, since the choice of the algo is to choose an algorithm which gives
as much less intrusion as possible, and the user can choose one of many
as suggested by the VM that has the collectors implementation in them,
is it possible to express the efficienty of an algorithm as a function
of the scheduler latency, heap memory size, processor speed and the WCET
of the application the user is running?
Say the Processor speed is P
Heap Memory size is M
Scheduler latency is S
And WCET determined for the real-time application on a no load system
(with no other applications except the system processes running)
And E the efficiency of the algorithm
Cant we say E = Af(P) + Bf(M) + Cf(S) + Df(WCET) + Constant
Where A, B, C and D are the weightages issued to each of the parameters
on the equation which can be determined empirically?
This is just a theory I thought of in my mind, and I cannot find any
literature which mentions that the efficiency of a collector can be
calculated in this manner.
Maybe this will give the application programmer to choose his collector
based on the response time deadlines he needs to meet for his real time
Please let me know if this is just a dumb thought, or if it can actually
be put as a factor that is a part of the conclusion to my study?
Thanks in advance, I appreciate any help.
The human race has one really effective weapon, and that is laughter. --