Re: o Re: sequence of octets bug in Orbit when running server as root?t

[Claude Beazley <cbeazley hgmp mrc ac uk>]

>On Tuesday 24 Sep 2002 11:39 am, Michael Meeks wrote:
> Hi Claude,
>
> On Mon, 2002-09-23 at 11:35, Claude Beazley wrote:
> > When printing out the length of the octet-sequence that the client tries
> > to pass to the server, the maximum sequence length that the server as
> > root accepts (on my PC. anyway) appears to be  262044
>
> 	Well; there are likely to be a number of issues with really large
> sequences. The sequence gets to be marshalled in-memory into GIOP wire
> format; which means a copy - unless it's a simply layed out
> sequence<octet> [ eg. ]. So, memory may be 1 problem, also - there are
> limits on the maximum size of incoming buffer in ORBit2 - that perhaps
> need re-visiting.
>
> 	What type does this sequence contain ? and why are you passing so much
> in one chunk ? typically in Gnome we break down big lumps into ~4K
> sequences so we don't introduce abnormally long processing delays
> anywhere.

Basically the (protoype) server requires the client to send it protein and DNA 
primary structures  so the server can perform  multiple alignments and other, 
more heavy duty,  manipulations. The client can then download the results. 
The data files are in ASCII and so the client loads up the sequence<octet> 
with chars.  As you can imagine these sequences can be rather long and any 
processing bottlenecks tend to be the rather intensive biocomputing 
applications which the server calls. 

I have setup the server so it can dictate the maximum size of the 
sequence<octet>  chunk that the client can transfer. The client can send and 
receive files in any chunk size it wants (up to the maximum size) . 
This allows me to more easily tailor the server according to the  
hardware/bandwidth used, it also gives the client as much flexibility as 
possible.

I also have to consider grid compliance, fibre optic bandwith rates and use of 
large beowulf clusters and I don't want to introduce any uneccessary 
bottlenecks, when the server gets implemented on these systems, by 
overlimiting the size of the sequences.