Wireshark in a Multi-Core Environment Using Hardware Acceleration
Wireshark in a Multi-Core Environment Using Hardware Acceleration
Wireshark in a Multi-Core Environment Using Hardware Acceleration
You also want an ePaper? Increase the reach of your titles
YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.
L<strong>in</strong>e Speed Captur<strong>in</strong>g ReviewNapatech Adapter Feature Benefit Standard Network AdaptersFrame burst buffer<strong>in</strong>g onadapterLong PCI burstsLarge host buffersOS bypass, zero copy ofcaptured packets directly touser application memoryNo data is lost, even when captured databursts exceed the PCI <strong>in</strong>terface speed, orthe PCI <strong>in</strong>terface is temporarily blocked.For NT20X 2 x 10 Gbps can be handleddown to 150 bytes frames.For NT20X 1 x 10 Gbps can be handled atany frame size.Data is lost, when captured burstsexceed the PCI <strong>in</strong>terface speed, or thePCI <strong>in</strong>terface is temporarily blocked.Very high PCI performance can be achieved The PCI performance will depend on thefor all frame sizes. frame size. E.g. the overhead for a 64-byte frame can be as much as 45%,while for a 1-KB frame it will be only 5%.Data can be processed at much higherspeed, and the frame process<strong>in</strong>g overheadis much lower (releas<strong>in</strong>g process<strong>in</strong>g powerto the user application).There is no packet copy<strong>in</strong>g or OS handl<strong>in</strong>goverhead.Merg<strong>in</strong>g of streams Adapters can merge packets received on 2or more ports <strong>in</strong> reception time order,whereby the host CPU is off-loaded.Frames are handled one at a time giv<strong>in</strong>ga large process<strong>in</strong>g overhead result<strong>in</strong>g <strong>in</strong>lower user application speed.A standard OS packet handl<strong>in</strong>g <strong>in</strong>terfaceperforms one or more copy of all framesresult<strong>in</strong>g <strong>in</strong> lower application speed.The sort<strong>in</strong>g of frames <strong>in</strong> time order mustby done by the host CPU, reduc<strong>in</strong>g thepossible host process<strong>in</strong>g performance.16