> On 9 Jan 2019, at 10:24, Gunaseelan Venkatachary - ERS, HCL Tech > <gunaseel...@hcl.com> wrote: > > Hello Team > Wish to reach out to bigger audience for some support in terms of performance > fine tuning, the context being , We are trying to insert a sample node, in > addition to whats already available in VPP, both in RX and TX interfaces, > using feature ARC ( like after device_input node in Rx) . > > We are facing some performance issues in that , since in this case we use > memcpy and byte order operations(using libc apis ), we suspect these 2 may be > causing issues , also found that vpp clib has similar apis too, hence is it > better to use vpp clib apis instead of libc apis to get better performance ?
Likely, try to use clib_memcpy_fast(....) instead. Typically you wan't to avoid function calls in dataplane code, so our implementations are inline functions.... What "show run" says about your node during the heavy load? (don't forget to do "clear run" after traffic is started) > > We observe good amount of rx_miss in show interface command, instead of drop > , any clues ? rx miss typically happens when there is more incoming traffic than what VPP can process... > > Also wish to know are there any known methods (Dos/Donts in VPP) for > performance tuning. For a start try to use our standard node coding pattern we use, like in sample-plugin. -- Damjan
-=-=-=-=-=-=-=-=-=-=-=- Links: You receive all messages sent to this group. View/Reply Online (#11881): https://lists.fd.io/g/vpp-dev/message/11881 Mute This Topic: https://lists.fd.io/mt/28982608/21656 Group Owner: vpp-dev+ow...@lists.fd.io Unsubscribe: https://lists.fd.io/g/vpp-dev/unsub [arch...@mail-archive.com] -=-=-=-=-=-=-=-=-=-=-=-
