Re: Something to read overlunch: Eclipse IDE in thebrowser

[Joe Pluta <joepluta@xxxxxxxxxxxxxxxxx>]

Aaron Bartell wrote:
> Use data queues and send "end of data set" messages so the queue knows when
> it has received all entries.  A similar approach would be to make data queue
> records as long as possible and "chunk" multiple logical entries onto one
> data queue record.  Each data queue record would have to have a header area
> where you would denote whether there were additional records needing to be
> read.
>
> Thoughts?
Data queues have certain properties that make them appealing, 
specifically the ability to have a single server support multiple 
clients.  I've done a lot of that over the years; it has pluses and 
minuses.  It's best suited for situations where your server load needs 
to be controlled; you can more easily control the resources assigned to 
a given activity by controlling its server.

However, for pure OLTP (and the even faster response time requirements 
of Web 2.0), I've moved almost entirely to a program call approach.  As 
server CPW grows in leaps and bounds, its more about call-to-call 
performance rather than load control, and for that nothing beats direct 
program calls.  And its quite easy to do "chunking" with program calls - 
you simply call the program multiple times.  One of the simplest 
approaches is to call a program with two data structures: a control DS 
and a data DS.  The control DS is used to indicate the operation to 
perform and to return the result code.  The data DS is simply for 
passing data.  To emulate a "result set", you perform an "init" opcode, 
and then start doing "get next" operations.  The control DS returns a 
flag indicating EOF.

One of the many benefits of this approach is that paging is entirely 
under control of the client.  It can start or stop reading records at 
any time and reset the control pointers.  You can even use embedded SQL 
in your called program with a cursor and in effect page your SQL calls.

It's a great architecture.

Joe