[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
[PSAMP] sequence number meaning in current drafts
I think that there is some inconsistency about the meaning of sequence number
among the architecture, protocol and info model drafts of the PSAMP WG.
In summary, the problem is whether this number should incrementally count the
sampled packets or the observed packets.
Note that, at the collector, only the count of sampled packets can be used to
detect losses between the exporter and the collector (e.g. a gap in the
sequence….5,6,7,9,10… would indicate that the eight sample was
The count of observed packets cannot on the contrary be used to detect losses
between the exporter and the collector, except in the case of systematic
count-based sampling (e.g. a gap in the sequence …6,12,18,30,36 indicates
that the sample associated to packet #24 was lost, but with time based or
probabilistic samples there wouldn’t be a regular structure in the
The count of observed packets is on the contrary useful to directly estimate,
at the collector, the sampling rate without the need of any additional
information (this is clearly explained in the architecture draft, sec. 4.3).
Now, the info model draft says (6.2.2) that the sequenceNumber is used to
specify the sequence number of a sample packet to record loss of packets while
exporting data flow records and then clearly refers to sampled packets.
The architecture draft, when describing Input Sequence Numbers (sec. 4.3)
clearly refers to observed packets, and a few paragraphs further (5.1)
specifies that the input sequence numbers MUST be contained in packet reports.
Then, in sec. 7.5, second paragraph, states that An additional sequence number
for dispatched export packet enables the collector to determine the degree of
loss in transmission. This clearly refers to sampled packets, but it
doesn’t specify further if/how this information is to be exported.
The protocol draft, in section 7.1, references both section 6.2.2 of the info
model draft and section 5.1 of the architecture draft, which we saw are
inconsistent. Furthermore, it suggests an additional way to detect losses
between the exporter and the collector by periodically exporting (in an option
record) the total number of the exported packets. It’s not clear if this
method is additional/parallel to a loss detection method based on sequence
numbers for samples (and not for export packets).
I hope that this summary can help in speeding the discussion and fixing the
to unsubscribe send a message to email@example.com with
the word 'unsubscribe' in a single line as the message text body.