[dba-Tech] Any ideas

Martin Reid mwp.reid at qub.ac.uk
Tue Oct 21 09:32:08 CDT 2003


Anyone have any ideas re the following?

I have a system whereby each PC in the SCCs sends in one short line per
minute to a central server. Each line is of the form IP address, time,
date[, user id]. The central server is only a P450 with 256Mb memory but
I have used a P733 with the same results.

When a user logs in to a PC, it writes a line to the same file on the
central server as all the other used PCs. Each PC writes at the same
second each minute, but the PCs determine their second to write by
chance, basically. Thus the incoming data for the file is reasonably
well spread across 60 seconds.

On the minute, the software on the central server renames the input
file, thereby causing a new one to be created with the next record sent
to it. The central file is held on a share to which each PC has to
authenticate.

When enough PCs are active, and I have not been able to deduce if there
is a threshold figure for that number, some or most of a record may be
lost. That can be seen from the input files.

During stress tests, when my PC was the only system communicating with
the server, My PC could send in about 630 lines per minute and none
would be lost. And this over a period of say an hour. However, when
multiple PCs send in lines, the data loss may arise with 50 PCs active.
The difference is the number of active network connections.

As I don't believe the data is being lost on the network (I have
monitored this and have not seen losses so far), it is most likely being
lost through the networking code/file system combination, and probably
the former.

I was wondering if anyone had a better method for collecting this
asynchronous auditing information, one which did not lose data.


More information about the dba-Tech mailing list