RARE user and assistance email list

[Frédéric LOUI <>]

Hi Cristina,

It has been recently decided by GÉANT project that GÉANT P4 lab will become a 
service in the long term.
Service definition/Processes are work in progress. Things are getting shape 
slowly though but surely :)

So policy with respect to other P4 switch belonging to other organisation is 
still a « unknown » topic.
Even if there is gentlemen agreement that these equipment can be accessed and 
monitored remotely.

I would therefore just recommend to consider nodes belonging to GÉANT 
Association only.
(i.e waiting for confirmation on how to best expose access to other P4 switch 
that does not belongs to GÉANT.)

For example, how to expose access to UFES P4 switch (grant for INTEL) ? 
Would you be OK to let other organisation from R&E say one NREN into GÉANT to 
access it in order to do experiment ?

This sort of policy document have not been written yet.

> - In the config you provided, PAR0101 is a DPDK node and the others are all 
> Tofino nodes?
Basically in the GÉANT P4 lab:

4 P4 switches:
* AMS0001/POZ0001/BUD0001 and are P4 nodes belonging to GÉANT Association.

2 DPDK nodes: (which area acting as RARE/freeRtr but are effectively 2 GÉANT 
BMS, PAR-BMS4 and PRA-BMS9)
* PAR0101 and PRA0101 are DPDK nodes respectively in PARIS and PRAGUE

3 DTNs:
* AMS-BMS6, HAM-BMS7,PRA-BMS8 

> - In the booking site (https://p4-bkd-srv.rare.nmaas.eu), we can only 
> reserve the 4 Tofino nodes (FRA,AMS,BUD,POZ), which are accessed via ssh. 
> What about the DPDK nodes? How to access them and where are their 
> connection points in the topology?

As Jordi left recently, the Booking system could not have been updated with 
the equipment above. 
This is WIP. 

> - Regarding the dpdk freerouters, are they already configured and ready to 
> use or we have to setup the virtual interfaces and other installation steps?
DPDK routers are configured as a RARE/freeRtr P4 switch node. With specific 
role for PAR0101 and PRA0101. PAR0101 having a leg in public Internet.

> - Regarding the traffic generator, we will start with ping and iperf, but 
> then we will need something more powerful. We participated in the dev 
> sessions when Carmem demonstrated TREX in her setup. Do you think we could 
> reproduce that in the P4 lab?
AMS-BMS6, HAM-BMS7 have Trex installed so feel free to use them. PRA-BMS8 is 
acting more as a multicast traffic source generator but you can install DPDK 
if needed.

> - Does freeRtr support any way to balance the load through different 
> tunnels (as in WCMP, weighted cost multiple paths)?

I’ll defer this question to Csaba on when this will be available. @mc36 ?

> - If we want to access other nodes that are not currently in the booking 
> site (e.g., CHI or GVA), how is it possible?

For now as it is unclear how to integrate all of these persistent testbed 
domain I’d (for now) only consider GÉANT nodes.

Don’t get me wrong, It does not mean that you cannot use GVA0001. For example 
GVA can basically be considered taking part of GP4L except that it belongs 
officially to SWITCH NREN, I’m talking under SWITCH control, SWITCH is 
willing to put GVA0001 at disposal of GP4L but as it is not yet written 
anywhere, I would rather have this policy document beforehand.

But I’m pretty sure for example, that switches in CERN or CALTECH might have 
different policy and access to these 2 switches might not be as easy as it is 
in GP4L.

> I don't know if you still are scheduling the freeRtr dev sessions on 
> Tuesdays, but, if possible, it would be a good opportunity to discuss this 
> experiment.

I cannot attend the regular VC today. (Meeting schedule clash) 
But as it is not freeRtr specific, we can tackle this in a side meeting.

The bad news is that I’m unavailable this week and next week (business trip). 
Therefore, if it is not too late we can have this VC monday after the GNA-G 
DiS meeting monday 17h->18h CEST. (Mainly from people who will access teh Lab)
Or another day during this week.

Please let me know your thoughts,
Frederic

> Le 10 mai 2022 à 01:56, Cristina Klippel Dominicini 
> <> a écrit :
> 
> 
> Hi Csaba,
> 
> Thanks for the files and explanation!
> 
> We still have some doubts:
> - In the config you provided, PAR0101 is a DPDK node and the others are all 
> Tofino nodes?
> - In the booking site (https://p4-bkd-srv.rare.nmaas.eu), we can only 
> reserve the 4 Tofino nodes (FRA,AMS,BUD,POZ), which are accessed via ssh. 
> What about the DPDK nodes? How to access them and where are their 
> connection points in the topology?
> - Regarding the dpdk freerouters, are they already configured and ready to 
> use or we have to setup the virtual interfaces and other installation steps?
> "you can always initiate a ping flood from these, or, the dpdk freerouters' 
> linux could connect to the testbed if you add a veth pair, whose side1 is 
> used by the linux to do the iperfing, and side2 is given to freerouter to 
> handle the packets... (the famous eth20001 interface of the default 
> freerouter installation to a vm, for example)"
> - Regarding the traffic generator, we will start with ping and iperf, but 
> then we will need something more powerful. We participated in the dev 
> sessions when Carmem demonstrated TREX in her setup. Do you think we could 
> reproduce that in the P4 lab?
> - Does freeRtr support any way to balance the load through different 
> tunnels (as in WCMP, weighted cost multiple paths)?
> - If we want to access other nodes that are not currently in the booking 
> site (e.g., CHI or GVA), how is it possible?
> 
> I don't know if you still are scheduling the freeRtr dev sessions on 
> Tuesdays, but, if possible, it would be a good opportunity to discuss this 
> experiment.
> 
> Best regards,
> Cristina
> 
> ________________________________________
> De: mc36 <>
> Enviado: segunda-feira, 9 de maio de 2022 13:55
> Para: Cristina Klippel Dominicini; Frederic LOUI
> Cc: Marcos Schwarz; Everson Scherrer Borges; Magnos Martinello; Rafael 
> Silva Guimarães; 
> Assunto: Re: Experiments PolKA vs. Segment Routing
> 
> hi,
> i go inline....
> br,
> cs
> 
> On 5/9/22 18:38, Cristina Klippel Dominicini wrote:
>> 
>> The main idea is to compare PolKA with Segment Routing in RARE, since both 
>> are source routing approaches supported by freeRtr. Thus, the first (and 
>> fundamental) question is: if you
>> can collaborate with us to produce this experiment, and if you see this as 
>> a nice showcase of RARE/freeRtr. Everson, Marcos, me and the rest of the 
>> team will be responsible for
>> producing the experiment, but we will for sure need your guidance and 
>> knowledge to properly explore RARE/freeRtr.
>> 
> yess!
> 
> 
>> 
>> As an initial step towards the big plan, we want to reproduce the emulated 
>> PolKA experiment AS IS in the P4 Lab, and we already have some initial 
>> questions:
>> 1) Is there any initial documentation on how to setup a basic experiment 
>> in the RARE P4 Lab with freeRtr (access, instalation, ...)?
> right now, the testbed is running rare in a mixed tofino & dpdk node 
> environment...
> attaching you the current configs plus the script to bring it up locally on 
> your
> pc in a similar way as everson does with his topology... as you can see in 
> the
> configs, both polka and sr are both up and running, the only thing yo need 
> to
> do is to add the tunnels and compare the results...
> 
>> 2) At the moment, the node configurations (port numbers, addresses) of the 
>> emulated experiment is not resembling the real testbed. Is there any 
>> documentation with this configuration?
> as you can see in the script, the configs are archived at nmaas, accessible 
> by eduid...
> (in case you dont have access to it, also attaching the current config 
> archive...)
> 
>> 3) In the emulated experiment, we have four core nodes (Frankfurt, 
>> Amsterdam, Poznan and Budapest) running PolKA core and two edges that 
>> setup the tunnels. For an experiment in the
>> physical setup, do you have any advice on how to design the edge nodes? In 
>> the first version of the experiment, should we use P4 nodes only for the 
>> core and DTNs as edges with
>> freeRtr? How do you normally setup edges in your experiments? It is not 
>> clear for us at the moment how exactly to deploy edge freeRtr nodes in the 
>> P4 Lab.
> so the lab have 4 tofino switches, and 2 dtn nodes running freerouters...
> you can always initiate a ping flood from these, or, the dpdk freerouters'
> linux could connect to the testbed if you add a veth pair, whose side1 is
> used by the linux to do the iperfing, and side2 is given to freerouter to
> handle the packets... (the famous eth20001 interface of the default
> freerouter installation to a vm, for example)
> 
>> 4) How do you currently generate traffic in the P4 Lab?
>> 
> ping flood from the dtns usually enough for me.... :)
> 
>> If possible, it would be very nice if we could schedule some hands-on 
>> sessions (as in the freeRtr dev meetings) to discuss and test this 
>> experiment with you.
> good idea... then it could be the topic for today...
> 
> ________________________________
> 
> Esta mensagem (incluindo anexos) contém informação confidencial destinada a 
> um usuário específico e seu conteúdo é protegido por lei. Se você não é o 
> destinatário correto deve apagar esta mensagem.
> 
> O emitente desta mensagem é responsável por seu conteúdo e endereçamento.
> Cabe ao destinatário cuidar quanto ao tratamento adequado. A divulgação, 
> reprodução e/ou distribuição sem a devida autorização ou qualquer outra 
> ação sem conformidade com as normas internas do Ifes são proibidas e 
> passíveis de sanção disciplinar, cível e criminal.
>