Network traces are a valuable source of information for modeling and analysis of network behavior and for the evaluation of network protocols. These crucial activities should be supported with reliable traffic traces. Reliable packet capturing facilities should be devoted to avoid losing packets and at the very last, report when packets have been lost with the highest accuracy possible. This paper describes a methodological approach to maximize the accuracy in the packet loss monitoring. The approach lies on the monitoring of appropriate statistical indicators directly from Ethernet hardware. These indicators become the enablers for the collection of metadata traces in parallel with the standard trace collections. The resulting meta-traces are proposed as a mechanism to monitor packet loss accuracy at the packet level as they provide the means to identify the exact location and amount of the packet loss in the collected traces files. We describe an evaluation use case of our approach with commodity hard- and software. The conclusion drawn from our experimental set up configuration is that it is possible to maximize the accuracy of the packet loss monitoring for reliable trace collections at affordable costs.
[1]
Fulvio Risso,et al.
Profiling and optimization of software-based network-analysis applications
,
2003,
Proceedings. 15th Symposium on Computer Architecture and High Performance Computing.
[2]
Luca Deri,et al.
High-Speed Dynamic Packet Filtering
,
2007,
Journal of Network and Systems Management.
[3]
Steven McCanne,et al.
BPF+: exploiting global data-flow optimization in a generalized packet filter architecture
,
1999,
SIGCOMM '99.
[4]
Fulvio Risso,et al.
Optimizing packet capture on symmetric multiprocessing machines
,
2003,
Proceedings. 15th Symposium on Computer Architecture and High Performance Computing.
[5]
Patrik Arlos,et al.
On the Quality of Computer Network Measurements
,
2005
.