Really brilliant idea! Found it here: http://www.net.in.tum.de/fileadmin/bibtex/publications/papers/MoonGen_draft2.pdf
Quote:
Measuring the CPU load caused by a DPDK-based application is difficult because DPDK recommends a busy-wait loop, i.e. the CPU load is always 100% for a typical application. MoonGen and other DPDKbased generators like Pktgen-DPDK are no exception to this. The bottleneck for packet transmission is usually not the CPU but the line rate of the network, so just measuring the achieved rate provides no insight. We therefore decrease the clock frequency of our CPU such that the processing power becomes the bottleneck. The performance can then be quantified as CPU cycles per packet. The same approach was used by Rizzo to evaluate the performance of netmap.
Quote:
Measuring the CPU load caused by a DPDK-based application is difficult because DPDK recommends a busy-wait loop, i.e. the CPU load is always 100% for a typical application. MoonGen and other DPDKbased generators like Pktgen-DPDK are no exception to this. The bottleneck for packet transmission is usually not the CPU but the line rate of the network, so just measuring the achieved rate provides no insight. We therefore decrease the clock frequency of our CPU such that the processing power becomes the bottleneck. The performance can then be quantified as CPU cycles per packet. The same approach was used by Rizzo to evaluate the performance of netmap.
No comments :
Post a Comment
Note: only a member of this blog may post a comment.