RARE user and assistance email list

[Alexander Gall <>]

On Fri, 5 Aug 2022 19:13:36 +0200, mc36 <> said:

> not yet... you can try to achieve this, should not be a big deal to pass 
> 0..23 to pipe0 and 24..32 to the rest to pipe1,
> but how the heck would you decide that you have to match against to the 
> pipe1 without doing so? :)
> yet an other issue is that crossing the pipe boundary means recirculation, 
> which is, a looped back port,
> that is, you limit the whole box to somewhat 100gbps of performance with 
> such packets....

Right, the recirculation ports become a bottleneck if you still want
to use all 32 ports for forwarding.  Another problem would be that the
setup would have to be symmetric, i.e. packets arriving on pipe1 would
have to be recirculated to pipe0 for the second lookup and I'm not
sure if this symmetry could actually be implemented with a split table.

I think the reasonable approach would be to sacrifice half of the
ports and use them all for recirculation, i.e. each port on pipe0
would have its dedicated recirculation port on pipe1, namely one of
the front-plate ports configured in loopback mode (the 32QS with two
additional pipes is providing this already with no loss of external
ports and would probably be better suited than the 32X if one needs
more than 16 ports).

It's also not immediately clear to me whether splitting a table into
two independent tables is a trivial thing. This also seems to depend
on whether one uses a normal TCAM or an algorithmic TCAM
(i.e. ALPM). But I'm not an expert on this and maybe that's all
well-known stuff and no problem at all :)

-- 
Alex

> br,
> cs



> On 8/5/22 18:06, Aaron A. Glenn wrote:
>> yes I saw your auto reply and thought the timing excellent :-)
>> 
>> hah of course, the mailing list! I had not even thought of it. thanks!! 
>> hope your holiday is superb!
>> 
>> cheers from sunny los angeles,
>> Aaron
>> 
>> On Fri, Aug 5, 2022, at 09:01, Frederic LOUI wrote:
>>> Hi Aaron !
>>> I'm on leave right now. I'm relaying your very interesting subject to
>>> the mailing list where you can get proper timely support     
>>> 
>>> I'm pretty sure you'll have very interesting feedback !
>>> 
>>> All the best
>>> Frederic
>>> 
>>> Le 5 ao  t 2022 17:27, "Aaron A. Glenn" <> a   crit :
>>>> Bonjour Fr  d  ric!
>>>> 
>>>> Have you or anyone in the project considered or suggested exploring a 
>>>> "folding pipeline" architecture in RARE for expanded LPM memory?
>>>> 
>>>> I have a (dual pipeline, Ufispace S9180) Tofino, at last! I've been 
>>>> commissioned to evaluate the platform as a "basic" provider edge / 
>>>> peering router for Packet Clearing House (pch.net). Table memory is a 
>>>> priority, as you might imagine. With the FreeRTR control plane being 
>>>> extraordinarily feature complete, and the continuous testing from the 
>>>> RARE project...I'd like to validate this for recommendation to 
>>>> production -- but I need more routes!
>>>> 
>>>> I am stuck in the US at present but return to CET in two weeks. Perhaps, 
>>>> schedule permitting, we could catch up? Email always works too, in these 
>>>> busy times :-)
>>>> 
>>>> Best,
>>>> 
>>>> -- 
>>>> Aaron A. Glenn,
>>>> Managing Director
>>>> 
>>>> Predicted Paths BV
>>>> Keizersgracht 62-64,
>>>> NL-1015 CS Amsterdam
>>>> Tel / Fax +31 85 208 2733
>>>> NL861040715B01 / KvK 77542223
>>>> 
>>