We have a complex character (telnet/host/terminal) integrated
application with several WebSpeed interfaces. We are using WS3.0 and
V9Db connected in shared memory on SCO Unixware with Apache. We have
used our existing 4GL code (as much as possible) run as embedded
speedscript. There is a performance issue I would like comments on.
With simple transactions, say user fills in HTML form with Order Entry
details, submits, creates one record in one table and WS TRNSVR returns
an updated web page with allocated Order Number, we get a transaction
time under two seconds. Given the interaction with Web Server, TRNSVR,
etc this seems reasonable
The performance issue occurs when we run the programs to allocate stock,
update debtors, general ledger etc. These additional .p's update several
tables and perform complex transactions. They add only about 0.25 of a
second when we run them in the character interface in host/terminal mode
on same server against same Db, using the same code. However if we
insert this code into our WebSpeed interface they add 3 seconds to
execution time ie over 10x slower than character. The code is very
efficient, running in multiple sites for many years, so doesn't seem to
be a coding/indexing issue.
This is leading us to conclude that if you want sub two second
performance WebSpeed code has to architected to only run the simplest
transactions and all back end processing must be run on an AppServer or
batch mode or similar.
Do other WebSpeed developers experiences correspond with ours?
Paul Royse
Applied Logic Systems
Auckland, New Zealand
paul@applog.co.nz <maimailtoaul@applog.co.nz>
application with several WebSpeed interfaces. We are using WS3.0 and
V9Db connected in shared memory on SCO Unixware with Apache. We have
used our existing 4GL code (as much as possible) run as embedded
speedscript. There is a performance issue I would like comments on.
With simple transactions, say user fills in HTML form with Order Entry
details, submits, creates one record in one table and WS TRNSVR returns
an updated web page with allocated Order Number, we get a transaction
time under two seconds. Given the interaction with Web Server, TRNSVR,
etc this seems reasonable
The performance issue occurs when we run the programs to allocate stock,
update debtors, general ledger etc. These additional .p's update several
tables and perform complex transactions. They add only about 0.25 of a
second when we run them in the character interface in host/terminal mode
on same server against same Db, using the same code. However if we
insert this code into our WebSpeed interface they add 3 seconds to
execution time ie over 10x slower than character. The code is very
efficient, running in multiple sites for many years, so doesn't seem to
be a coding/indexing issue.
This is leading us to conclude that if you want sub two second
performance WebSpeed code has to architected to only run the simplest
transactions and all back end processing must be run on an AppServer or
batch mode or similar.
Do other WebSpeed developers experiences correspond with ours?
Paul Royse
Applied Logic Systems
Auckland, New Zealand
paul@applog.co.nz <maimailtoaul@applog.co.nz>