Click for:

3rd February 2004
Visited Simon to recover Martinet hardware. Jeff ran some high speed traffic to see what the limitations were on the database software.

It was discovered that the Linux box seems to buffer data, and only releases it as the database application requests it via the streams protocol. Steve Edwards:

"...The mkb_cdr program buffers output on all its IP connections individually.
Each connection gets its own 64kbyte buffer. This should be good for quite
a few CDR's. Purpose is to allow for bursts over slow LANs etc..."

We tried several runs at various traffic rates to check the CPU utilisation on the database PC:

PC Spec: AMD K6 500MHz, 128MB RAM.
12 CDRs/sec = 11% CPU
24 CDRs/sec = 20% CPU
48 CDRs/sec = 35% CPU
96 CDRs/sec = 70% CPU
120 CDRs/sec = 100% CPU

Beyond 24 CDRs/sec the MySQL was utilising approx. 4 times the CPU of the insertcdrs application.

Jeff will run tests here (PC Spec: P4 1.5GHz, 384MB RAM) to compare these results - we may be able to extrapolate some kind of graph from this.

Jeff is also working on an Access database to graph average traffic rates and total calls. There are a few problems, but it is giving quite good results so far.

The testing on the "shadow" system is continuing at PTL HQ with the latest development delivery from Simon expected soon..