Subject | FB 1.5 and Win2K Server issue |
---|---|
Author | Rafael Szuminski |
Post date | 2004-05-28T04:36:12Z |
Hi All,
I have run into a "small" issue while stress testing a FB 1.5 prototype database
on Win2K Server yesterday. It seems like some/all records get "lost" in the
translation.
Ok, here is the background:
OS: Win2K Server SP2
Database Engine: FB 1.5 (installed with the installer as a service)
Clients: Win2K, WinXP, Win98 (all together 10 machines)
Drivers: ADO 2.6 / IBProvider 2.3
Isolation level: Read Committed
Database: Small database, 20 tables, 120 SPs, main table has 4000 records, each
record might have up to 20 detail records scattered across the rest of the tables.
Test Scenario:
Please remember that this was a stress test. Each client computer was running 5
instances of the test application. Each test app would query the db to retrieve
approximately 1000 records and then would choose a random record and pull all
the detail records for the chosen record. This was happening in a loop for 10
min. at a time. So, all together we had 50 connections the database executing
selectable stored procedures in a constant loop.
After starting the test, 5 of the machines (25 instances of the app) have
returned 0 records where they were expected to return the 1000 records. The
other 5 machines worked properly. After restarting the test, again 4-5 machines
failed to retrieve any records, but...this time some of the failed machines were
the ones that worked in the previous test. We have repeated this test
approximately 15 time and each time it was unpredictable which machine would
fail. We have had even an instance where only 2 of the 5 instances failed on a
given computer.
This test have been performed multiple times against FB 1.5 running on WinXP and
Win2K pro and each time it worked flawlessly. The weirdest thing was there were
no errors and timeouts, just empty recordsets. Oh one more thing, we have
rebooted the Win2K Server after a failed sessions and surprisingly the next test
worked. However, all consecutive tests had some issues :-)
Anyhow, sorry for the long post, hope it makes sense and that somebody will be
able to give me a hint on whats going on.
Thanks
Raf
I have run into a "small" issue while stress testing a FB 1.5 prototype database
on Win2K Server yesterday. It seems like some/all records get "lost" in the
translation.
Ok, here is the background:
OS: Win2K Server SP2
Database Engine: FB 1.5 (installed with the installer as a service)
Clients: Win2K, WinXP, Win98 (all together 10 machines)
Drivers: ADO 2.6 / IBProvider 2.3
Isolation level: Read Committed
Database: Small database, 20 tables, 120 SPs, main table has 4000 records, each
record might have up to 20 detail records scattered across the rest of the tables.
Test Scenario:
Please remember that this was a stress test. Each client computer was running 5
instances of the test application. Each test app would query the db to retrieve
approximately 1000 records and then would choose a random record and pull all
the detail records for the chosen record. This was happening in a loop for 10
min. at a time. So, all together we had 50 connections the database executing
selectable stored procedures in a constant loop.
After starting the test, 5 of the machines (25 instances of the app) have
returned 0 records where they were expected to return the 1000 records. The
other 5 machines worked properly. After restarting the test, again 4-5 machines
failed to retrieve any records, but...this time some of the failed machines were
the ones that worked in the previous test. We have repeated this test
approximately 15 time and each time it was unpredictable which machine would
fail. We have had even an instance where only 2 of the 5 instances failed on a
given computer.
This test have been performed multiple times against FB 1.5 running on WinXP and
Win2K pro and each time it worked flawlessly. The weirdest thing was there were
no errors and timeouts, just empty recordsets. Oh one more thing, we have
rebooted the Win2K Server after a failed sessions and surprisingly the next test
worked. However, all consecutive tests had some issues :-)
Anyhow, sorry for the long post, hope it makes sense and that somebody will be
able to give me a hint on whats going on.
Thanks
Raf