Відео

are you in cli? if not it won't work

What is the ip forwarding capacity of a single IOM? Does MPLS switching VS IP routing affect capacity?

@@playfulyogi5639 Nokia's current fp5 linecards state a capacity of 18Tb/s per slot this would consist of 3 FP5 route processors. Technically yes MPLS header overheads reduce payload capacity since by definition you burn 32 bits per label and there is a hard limit inside the box for ethernet mtu's of 9216. The linecards should also have a tested a packets per second rate (pps) that is more agnostic of header overhead since as long as the packet can be processed in one pipeline pass and does not have to be recirculated there is no pps difference between an mpls label lookup to forward vs an ip lookup to forward, only if multiple different lookups or de-encaps / re-encaps have to be done forcing the asic to recycle the packet. Cisco has a fantastic video about ASIC packet forwarding rates and comparing pps vs packet size in RFC 2544. I don't know if I can directly link videos so here are the watch urls: wt-O_e2WWM4 2RKZ9mAwP7w

​ @dabb11 ​ thank you for the explanation! I feel like we need to make a 2023 follow-up video for FP5 and 7750 SR-s that explains its technical capabilities, especially QoS features for carrier-grade services. Your comments are a great starting point. A new video could also go into some of the integrations with 7750 + NFM-P (formerly 5620 SAM) as well as gRPC/gNMI programmability, maybe presenting a case study?

ASIC design is generally closely protected intellectual property so you won't find much from Nokia beyond brochure buzzwords but if you just want to know more about the design in general watch some of the Broadcom Jericho 2 talks and the Cisco 8000 Tech Field Day talk where they talk about the Cisco One ASIC. All the netconf or API programmability has specifically to do with the control plane which is handled by a general purpose x86 computer (Juniper call it the route processor Cisco calls it the Route Switch Processor I am unsure what Nokia calls it). What is programmed in the ASIC can be a lot but it's essentially only instructions of where and how to forward. The point of the ASIC isn't to sit and think about packets it's to get them out of the box as quickly as possible, obviously stuff like QoS dictates how specific traffic should be forwarded vs other traffic but you really just want to process and forward the packet. That is all to say when the x86 general purpose computer running all the routing protocols and gRPC/gNMI daemons and whatnot that you program and configure, when it figures out how traffic should flow it programs into the ASIC's memory instructions for certain packet markers so either a DSCP (QoS) field or maybe an IP header (ACL) just as examples. These instructions are just entries in a table that resolve to a pointer to forward to a specific interface, queue, or ask the exit interface if it can buffer or send, or re-encapsulate or de-encapsulate the packet. These are all very agnostic of the control plane protocols that you interact with through gRPC. The ASIC doesn't speak those languages it speaks JTAG or P4 or some machine other language for that circuit. The changes in ASIC's where it matters the most are the Serializer Deserializers. They are a standard and the current fastest standard is 112Gb/s with 224Gb/s still in the works. This is one of the main things that speed up network ASIC's over time, the challenge with designing these chips usually comes down to how many lookups and how many (de)encapsulations you can do in one pipeline depending on the ASIC model so Compute To Completion or otherwise, as well as maintaining signal integrity with the new emissions from the higher speed SerDes. A bit of a tangent but you can tell what speed SerDes a chip uses if you can find the specs for its (almost always) ethernet interfaces where it will say something like 400Gb/s Ethernet - 4x100Gb/s optically - 8 x 50 Gb/s Electrically. This means it would be using 56 Gb/s SerDes so you can always look and see for yourself if Nokia hides these stats anywhere they are almost always posted somewhere but manufacturer websites are particularly terrible and unhelpful for actually posting their real specs. For example: most Carrier grade Network Asics have supported all the same lookups and encapsulations that are used now. I didn't mean that all ASICS did and still do since a large problem for a very long time was for example VXLAN l3 gateway where Trident 2 (?) boxes didn't support this which was actually a big driver behind Juniper making their Q5 silicon. A more specific case: the silicon used in Ciscos NCS 500 boxes can only push 4 labels at once, 2 being transport 2 being service with there being a way to push 3 transport and 1 service label. If you need to push more than 3 transport labels the ASIC recirculates the packet by having a special on chip interface and a pointer in the forwarding table to itself so it can push more labels in the compute pipleine again but this effectively burns double the bandwidth for that packet so if you have to do this for every packet you will literally half your pps and throughput. It's a long post and I am probably not summarizing the concepts well enough but it's a hard topic to holistically cover in a single UA-cam comment.

I don't mean to be reductionist here either, Nokia FP5 does have a fair amount of new features over FP4 like linerate AnySEC which is like MACSEC but not interface based and not just layer 2.

These are virtual routers you need physical router to connect a cable

I am glad it helped 😊

its in the description also test-for-coronavirus.service.gov.uk/register-kit

check with a different usb cable or port i had exactly same issue

That's how it is supposed to be. The dongle is the device.

boot camp

Mac has always suhced.

>2013 I don't think he cares

it's dynamic

Faisal Khan By definition, Dynamic, yes. But VMs DHCP lease times are really long. 4 days I think.

LOL Open challenge for every one to hack it