Non-random Test Data risks compression

[aaronmbr]

Original issue 161 created by aaronmbr on 2014-10-06T21:06:02.000Z:

Reading the source, I noticed that the data that NDT sends in S2C and C2s tests is nonrandom.

From the Flash client:
// Prepare the data to send to the server.
for (var i:int = 0; i < NDTConstants.PREDEFINED_BUFFER_SIZE; i++) {
_dataToSend.writeByte(i);
}
https://code.google.com/p/ndt/source/browse/trunk/flash-client/src/TestC2S.as#&nbsp;168

From the C client:
// ....Fill buffer upto NDTConstants.PREDEFNED_BUFFER_SIZE packets
pkts = 0;
k = 0;
for (i = 0; i < (64*KILO_BITS); i++) { // again buffer sizes differ.
// Since the actual transmitted byte count is timed, it doesn't appear
// that it is causing specific problems.
while (!isprint(k & 0x7f))
k++;
buff[i] = (k++ % 0x7f);
}
https://code.google.com/p/ndt/source/browse/trunk/src/test_c2s_clt.c#&nbsp;141

Because we're sending buffers of sequential bytes (and in the case of the C client, only printable ASCII characters), there is a risk that boxes on the network will compress the data and skew measurements.

We should modify the code for data generation so that we use a pseudorandom number generator with a periodicity of at least 64k, then take the last byte of each generated number as the next byte to place in the send buffer.