[gt-user] WS call performance... doesn't seem to be linear with respect to the input/output argument's size

[Tim Freeman]

What is the element data type you tested with? 

Thanks,
Tim 

On Thu, 21 Sep 2006 14:39:02 -0500
Ioan Raicu <iraicu at cs.uchicago.edu> wrote:

> Hi everyone,
> I have a question regarding the WS call performance.
> My input argument contains an array of elements.  My instincts tel me 
> that the larger my array of elements that I send in 1 WS call, the lower 
> the overhead of the WS call will be.  For example, if sending an array 
> of size 1 takes 50 ms, I would expect sending an array of size 10 to 
> take <= to 500 ms, size 100 to take <= 5000 ms, and so on... at least 
> until you run out of memory on the JVM. 
> 
> Here are some numbers I have observed in one of my web services I have 
> written.
> 
> Array length to be sent in 1 WS call 	Time (ms) to make 1 WS call
> 	Time (ms) per array element
> 1 	51 	51
> 2 	50 	25
> 4 	54 	13.5
> 8 	53 	6.625
> 16 	73 	4.5625
> 32 	72 	2.25
> 64 	98 	1.53125
> 128 	126 	0.984375
> 256 	211 	0.82421875
> 512 	406 	0.79296875
> 1024 	843 	0.823242188
> 2048 	2525 	1.232910156
> 4096 	9838 	2.401855469
> 8192 	41496 	5.065429688
> 
> 
> Notice that the WS call overhead is indeed amortized as the array length 
> increases up to about when the array length is between 256 and 1024 (my 
> guess is that the array length is really a function of each element 
> size, so depending on how complex each element is, the array size will 
> differ).  My question is why is the WS call overhead increasing again 
> once we get to larger (1024+) array sizes?  The problem is that the time 
> per element array keeps increasing to very larger numbers with every 
> increase in array length... at least up to about 128K for the array size 
> that I tested.  In a perfect world, I would have expected that the Time 
> per array element to flatten out until the JVM ran out of memory.  Why 
> is this not so?  What exactly makes the processing of WS call arguments 
> not be linear in terms of the argument size?
> 
> Thanks in advance,
> Ioan
> 
> -- 
> ============================================
> Ioan Raicu
> Ph.D. Student
> ============================================
> Distributed Systems Laboratory
> Computer Science Department
> University of Chicago
> 1100 E. 58th Street, Ryerson Hall
> Chicago, IL 60637
> ============================================
> Email: iraicu at cs.uchicago.edu
> Web:   http://www.cs.uchicago.edu/~iraicu
>        http://dsl.cs.uchicago.edu/
> ============================================
> ============================================
> 
>