[netsniff-ng] Re: [PATCH 0/4] Introduce new pcap io API for pcap r/w accesses
On 2017-01-12 at 13:59:05 +0100, Vadim Kochan wrote: > On Wed, Dec 14, 2016 at 11:33 PM, Vadim Kochan wrote: > > On Wed, Dec 14, 2016 at 11:26 PM, Tobias Klauser > > wrote: > >> On 2016-12-12 at 22:09:52 +0100, Vadim Kochan wrote: > >>> Add new pcap io API to make pcap read/write accesses more > >>> simpler and generic. Added pcap_io & pcap_packet struct's to > >>> keep some internal pcap state like magic, link type & packet header > >>> instead of to pass them like parameters and keep it all within > >>> netsniff-ng.c. > >>> > >>> Also such approach might be used to unify sniffing from ring buffer via > >>> pcap io API similary as it is done with regular files. > >>> > >>> Some fast-path sensitive or setter/getter functions were inlined in > >>> pcap_io.h. > >> > >> This series touches very sensitive fast-path code. Before even reviewing > >> it, I'd like to see performance measurments proofing that the your > >> patches don't introduce any performance regressions and that inlining of > >> these functions indeed does the job. > >> > >> Thanks! > > > > Hm, yeah, I need to think how to perform this on my laptop ... > > I will try to measure time execution of critical code via clock() > syscall for both versions if it will be OK for you, if yes - > would be it OK to add some #IFDEF's (ahhh not too nice probably) to > have such calculation persistent in the code ? I think it should also be possible to test basic performance by just invoking netsniff-ng under perf and then comparing profiles. Of course you're also free to instrument the code for performance measurments locally, but I'd rather not put any instrumentation code into the upstream repo. Thanks Tobias -- You received this message because you are subscribed to the Google Groups "netsniff-ng" group. To unsubscribe from this group and stop receiving emails from it, send an email to netsniff-ng+unsubscr...@googlegroups.com. For more options, visit https://groups.google.com/d/optout.
[netsniff-ng] Re: [PATCH 0/4] Introduce new pcap io API for pcap r/w accesses
On Wed, Dec 14, 2016 at 11:33 PM, Vadim Kochan wrote: > On Wed, Dec 14, 2016 at 11:26 PM, Tobias Klauser wrote: >> On 2016-12-12 at 22:09:52 +0100, Vadim Kochan wrote: >>> Add new pcap io API to make pcap read/write accesses more >>> simpler and generic. Added pcap_io & pcap_packet struct's to >>> keep some internal pcap state like magic, link type & packet header >>> instead of to pass them like parameters and keep it all within >>> netsniff-ng.c. >>> >>> Also such approach might be used to unify sniffing from ring buffer via >>> pcap io API similary as it is done with regular files. >>> >>> Some fast-path sensitive or setter/getter functions were inlined in >>> pcap_io.h. >> >> This series touches very sensitive fast-path code. Before even reviewing >> it, I'd like to see performance measurments proofing that the your >> patches don't introduce any performance regressions and that inlining of >> these functions indeed does the job. >> >> Thanks! > > Hm, yeah, I need to think how to perform this on my laptop ... I will try to measure time execution of critical code via clock() syscall for both versions if it will be OK for you, if yes - would be it OK to add some #IFDEF's (ahhh not too nice probably) to have such calculation persistent in the code ? Regards, Vadim Kochan -- You received this message because you are subscribed to the Google Groups "netsniff-ng" group. To unsubscribe from this group and stop receiving emails from it, send an email to netsniff-ng+unsubscr...@googlegroups.com. For more options, visit https://groups.google.com/d/optout.
[netsniff-ng] Re: [PATCH 0/4] Introduce new pcap io API for pcap r/w accesses
On Wed, Dec 14, 2016 at 11:26 PM, Tobias Klauser wrote: > On 2016-12-12 at 22:09:52 +0100, Vadim Kochan wrote: >> Add new pcap io API to make pcap read/write accesses more >> simpler and generic. Added pcap_io & pcap_packet struct's to >> keep some internal pcap state like magic, link type & packet header >> instead of to pass them like parameters and keep it all within netsniff-ng.c. >> >> Also such approach might be used to unify sniffing from ring buffer via >> pcap io API similary as it is done with regular files. >> >> Some fast-path sensitive or setter/getter functions were inlined in >> pcap_io.h. > > This series touches very sensitive fast-path code. Before even reviewing > it, I'd like to see performance measurments proofing that the your > patches don't introduce any performance regressions and that inlining of > these functions indeed does the job. > > Thanks! Hm, yeah, I need to think how to perform this on my laptop ... -- You received this message because you are subscribed to the Google Groups "netsniff-ng" group. To unsubscribe from this group and stop receiving emails from it, send an email to netsniff-ng+unsubscr...@googlegroups.com. For more options, visit https://groups.google.com/d/optout.
[netsniff-ng] Re: [PATCH 0/4] Introduce new pcap io API for pcap r/w accesses
On 2016-12-12 at 22:09:52 +0100, Vadim Kochan wrote: > Add new pcap io API to make pcap read/write accesses more > simpler and generic. Added pcap_io & pcap_packet struct's to > keep some internal pcap state like magic, link type & packet header > instead of to pass them like parameters and keep it all within netsniff-ng.c. > > Also such approach might be used to unify sniffing from ring buffer via > pcap io API similary as it is done with regular files. > > Some fast-path sensitive or setter/getter functions were inlined in pcap_io.h. This series touches very sensitive fast-path code. Before even reviewing it, I'd like to see performance measurments proofing that the your patches don't introduce any performance regressions and that inlining of these functions indeed does the job. Thanks! -- You received this message because you are subscribed to the Google Groups "netsniff-ng" group. To unsubscribe from this group and stop receiving emails from it, send an email to netsniff-ng+unsubscr...@googlegroups.com. For more options, visit https://groups.google.com/d/optout.