default data-size is the size of the buffers (how much data they can fit), not the number of buffers [1]. Buffers are used to store packets. If a packet cannot fit in a single buffer (eg. a jumbo 9K packets), VPP will build a chained list of buffers. In a typical system, you can expect to use 1 buffer per packet with 1500-bytes packets. If you want to increase the number of in-flight packets in VPP, you need to increase the number of buffers VPP can receive at the same time. This is controlled by the size of the rx queues used by drivers for the NIC. For example for DPDK it is controlled by num-rx-desc [2]. Be careful though as as you increase the number of descriptors per rx queue you might run out of buffers globally [3]. Finally, be careful when increasing rx queues size, this impacts performance. You should keep it to the minimum while being big enough to hide your processing latency.
Best ben [1] https://fd.io/docs/vpp/master/gettingstarted/users/configuring/startup.html#default-data-size-number [2] https://fd.io/docs/vpp/master/gettingstarted/users/configuring/startup.html#num-rx-desc-n [3] https://fd.io/docs/vpp/master/gettingstarted/users/configuring/startup.html#buffers-per-numa-number > -----Original Message----- > From: vpp-dev@lists.fd.io <vpp-dev@lists.fd.io> On Behalf Of Feng Gao > Sent: vendredi 16 avril 2021 03:03 > To: Marcos - Mgiga <mar...@mgiga.com.br> > Cc: dmar...@me.com; vpp-dev@lists.fd.io > Subject: Re: [vpp-dev] TX Queue Placement > > Hi Macros, > > please see inline > > On Thu, Apr 15, 2021 at 8:44 PM <mar...@mgiga.com.br> wrote: > > > > Hi Feng, > > > > I do see some drops , as the following output shows: > > > > > > rx bytes 650888808551 > > tx packets 6222947195 > > tx bytes 7707890251578 > > drops 1932271 > > ip4 2291263868 > > rx-miss 9976 > > tx-error 1990 > > > > In order to increase the buffer size, do I have to change the "default > data-size" in buffer section? > > Yes, i think so. the default size is 2048. > > > > > Best Regards > > > > > > Marcos > > -----Mensagem original----- > > De: vpp-dev@lists.fd.io <vpp-dev@lists.fd.io> Em nome de Feng Gao > > Enviada em: quarta-feira, 14 de abril de 2021 23:10 > > Para: Marcos - Mgiga <mar...@mgiga.com.br> > > Cc: dmar...@me.com; vpp-dev@lists.fd.io > > Assunto: Re: [vpp-dev] TX Queue Placement > > > > Hi Macros, > > > > You could use "show interface" to check if there are some drop packets. > > If so, it may mean the rx buffer isn't enough. > > > > And you could use "show node counters" to check if there some drop > packets by other reasons. > > > > On Wed, Apr 14, 2021 at 8:41 PM Marcos - Mgiga <mar...@mgiga.com.br> > wrote: > > > > > > Understood... > > > > > > > > > > > > > > > > > > Usually I use show run to monitor packet drops. Do you have another > suggestion for me to get a more precise information about it? > > > > > > > > > > > > Best Regards > > > > > > > > > > > > De: vpp-dev@lists.fd.io <vpp-dev@lists.fd.io> Em nome de Damjan Marion > > > via lists.fd.io Enviada em: quarta-feira, 14 de abril de 2021 09:06 > > > Para: Marcos - Mgiga <mar...@mgiga.com.br> > > > Cc: vpp-dev@lists.fd.io > > > Assunto: Re: [vpp-dev] TX Queue Placement > > > > > > > > > > > > > > > > > > General rule is that increasing those values are reducing number of rx > > > drops due to > > > > > > VPP being de-scheduled or busy or due to traffic bursts, but also that > degrades performance. > > > > > > > > > > > > I think what we have set as default is reasonable and you should leave > > > it unless you are experiencing some issues… > > > > > > > > > > > > — > > > > > > Damjan > > > > > > > > > > > > > > > > > > On 14.04.2021., at 13:45, Marcos - Mgiga <mar...@mgiga.com.br> wrote: > > > > > > > > > > > > Hello Damjan, > > > > > > > > > > > > Thank you for clarifying... > > > > > > > > > > > > I also have a question about num-rx-desc and num-rt-desc parameters, > > > hope you dont mind to dicuss it in this e-mail > > > > > > > > > > > > I would like to understand what value fits better to my environment, > do have any thoughts about it? > > > > > > > > > > > > Best Regards > > > > > > De: vpp-dev@lists.fd.io <vpp-dev@lists.fd.io> Em nome de Damjan Marion > > > via lists.fd.io Enviada em: quarta-feira, 14 de abril de 2021 08:24 > > > Para: Marcos - Mgiga <mar...@mgiga.com.br> > > > Cc: vpp-dev@lists.fd.io > > > Assunto: Re: [vpp-dev] TX Queue Placement > > > > > > > > > > > > > > > > > > > > > > > > > > > On 14.04.2021., at 13:21, Marcos - Mgiga <mar...@mgiga.com.br> wrote: > > > > > > > > > > > > Hello, > > > > > > > > > > > > I increased VPP rx/tx queues in order to enable RSS on VPP instance. > Since VPP is running on a NUMA system with two threads with 8 cores each, > I would like to pin TX / RX queue to proper NUMA nodes. > > > > > > > > > > > > Using set interface rx-placement I was able to associate rx queue to > desired cores, so I would like to know if is there any possibility to pin > tx queue to a certain workers as well. > > > > > > > > > > > > > > > > > > Not at the moment. tx queues are statically mapped (0 to main thread, > 1 to worker 0, 2 to worker 1, etc.). > > > > > > There are some plans to implement such capability….. > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > >
-=-=-=-=-=-=-=-=-=-=-=- Links: You receive all messages sent to this group. View/Reply Online (#19225): https://lists.fd.io/g/vpp-dev/message/19225 Mute This Topic: https://lists.fd.io/mt/82088483/21656 Group Owner: vpp-dev+ow...@lists.fd.io Unsubscribe: https://lists.fd.io/g/vpp-dev/unsub [arch...@mail-archive.com] -=-=-=-=-=-=-=-=-=-=-=-
