Re: (forte-users) Communication Pbs between a NT client& a HP 11 load-b

Answer 1

Hi,
I think you misunderstand the anatomy of an Array object. The size of an
Array containing 63 elements can never be 1638 bytes when each element
contains a 100 byte string.
If you use the sizeof function, you'll only get the combined size of all
attributes of the object. But if an object contains attributes that are
objects themselves (e.g. TextData), then only the size of the pointer to
this object is counted. Not the object itself.
Array and LargeArray offer the same functionality but use different
algorithms. LargeArray is better suited for large lists and Array is better
suited for small lists. Array is faster, but requirs more memory. LargeArray
is slower, but requires less memory.
Array reserves a piece of memory that holds 63 pointers to objects. Within
these 63 positions, you can read and write randomly. When you add a 64th
object, Array locates a new piece of memory that holds 127 pointers. The
first 63 pointers are copied to this new location and the original location
is free for garbage collection. When you add a 128th object, Array locates a
new piece of memory that holds 255 pointers. etc.
Check if you're network communication time is closer to what you expect when
you have 127 elements in your list. And then check if performance drops
again when you have 128 elements.
Also, take into account that Forte is not capable of measuring time in very
small units. 80 ms and 150 ms might really just be 3 ms apart. Also, garbage
collection might kick in at any moment, slowing down performance. If you
have a multi-user OS, it might suddenly start doing maintenance tasks, that
reduce the amount of processor time available for your partition, either the
partition that sends the array, or the partition that receives it.
Sending an object requires the object to be serialised, send and
deserialised. The partition that invoked the Send() method blocks during
this whole process. If deserialising at the other end takes longer, then the
process simply waits longer.
-----Original Message-----
From: Olivier Brousseau [SMTP:Olivier.Brousseausema.fr]
Sent: Thursday, April 06, 2000 7:05 PM
To: kamranaminyahoo.com
Cc: Eric ABECASSIS; Giulio CALIENDO
Subject: (forte-users) Communication Pbs between a NT client & a HP
11 load-balanced server : 100ms to transfer 1600 bytes on a 10Mbits link.
Hi,
we have done a very small application to instrument & verify the
communication capabilities of our soft & network :
1. a win NT client that generates an array of object ( nb of object is a
parameter ), and that send it to a server.
2. a HP11 server, that contains a load-balanced SO ( with only one
replicate ). The prototype of the SO entry point, which does nothing, is
as
follows :
void send( input GenericArray of Element );
And Element is a class that contain a DataValue attribute with a 100 bytes
string.
Between the client & server, we have a 10 Mbits link ( it has been
verified
using FTP ). From 1 object sent, to 63 objects, ( size of the array with
63
objects = 1638 bytes ), everything is OK. The elapsed time is 10
milliseconds.
With 64 objects ( size of the array = 1666 bytes ), the elapsed time is
100
milliseconds ( in fact for one iteration, it's 80 msec, for the next one,
150 msec, and then it's 80 msec, etc... ).
All the results with more objects than 63 are awfull.
Has anyone enccountered such a problem ?
Can anyone advise ?
Thanks,
olivier
Nota. Both server & client are built from Forté 3.0.M
O. Brousseau
Architecture & Performance Team
Sema Group France ( DTS )
For the archives, go to: http://lists.xpedior.com/forte-users and use
the login: forte and the password: archive. To unsubscribe, send in a new
email the word: 'Unsubscribe' to: forte-users-requestlists.xpedior.com