D. Method overhead
Considering that our approach makes use of flow-based
information to classify patterns of traffic within a given time
interval, and that samples of such information are collected
every 3 seconds, there is a remarkable overhead reduction
to the whole detection mechanism when compared to other
approaches based on the KDD-99 dataset. Their additional
overhead is caused by the need to collect every packet sent
to a victim, and then pre-process this information to generate
connection records. Besides, if we consider the worst case
scenario (DDoS flooding attack with a very high flow of
packets for the attack rate), then the overhead tends to be
very high.
In order to compare our method to other KDD-99 dataset
approaches we built another experiment in which a DDoS
flooding traffic was generated with a high rate of attack and
IP header spoofing (worst-case scenario). This experiment
produced a huge number of flow entries that were supported
by the OF switch and stressed our Collector and Extractor
modules as they processed all flow entries.
Table VI shows a comparison of the CPU time to extract
features needed for detection for KDD-99 dataset approaches
and ours. The values for KDD-99 are those presented in [15],
and reported as obtained from experiments run on a system
with 2.66 GHz, dual core CPU, and 3.5 GB of RAM memory.
Our values were obtained from experiments run on a system
with 1.8 GHz, dual core CPU, and 2 GB of RAM memory.
The time interval corresponds to generating 30,000 samples
in both cases.
đang được dịch, vui lòng đợi..