Performance

CPU, disk and bus I/O are the three main bottlenecks. CoMo activities can be partitioned among multiple CPUs/systems to improve performance:

bus I/O overheads can be reduced by doing packet capture, timestamp and coalescing on an external box (e.g. IXP-based)

subsystems can be run in parallel on SMP machine with little resource conflicts. If the system permits, classifiers can also be run in parallel as packets are stored in readonly memory areas and handled without reference counts.