RARE user and assistance email list

[Marcos Felipe Schwarz <>]

Hi Tim, there is also Edgecore 9516-32D (DCS810) which is a Tofino 2 device 
and has 64MB and 400G ports.

Regards,

Marcos Schwarz

-------- Forwarded Message --------
Subject: Re: [PRP-List] PRP/NRP Zoom from 10am to 11am Pacific on Thursday 
March 3
Date: Thu, 10 Mar 2022 12:02:46 +0000
From: Tim Chown <>
To: Tomasz Szewczyk <>
CC: Fr  d  ric LOUI <>, Xavier Jeannin 
<>, mc36 <>, Ivana Golub 
<>

Hi,

We had a useful chat in today   s WB-PERF call on testing, and focus on use 
cases.

For info, there   s discussion of buffer size impact at the usually good 
ESnet fasterdata site - 
https://fasterdata.es.net/network-tuning/router-switch-buffer-size-issues/

That links to a table of devices and buffer sizes - 
https://people.ucsc.edu/~warner/buffer.html
That lists the wedge at 16MB, as per the data sheet.  I see an Edgecore 
device with 64MB, but that is a Broadcom Tomahawk 3 device with 400G ports :)

Tim

> On 9 Mar 2022, at 06:46, Tomasz Szewczyk <> wrote:
> 
> Hi Frederic,
> In general - our observation was that Broadcom based boxes share 
> same/similar buffer space. So in other words results are comparable.
> Juniper MX as Service/Edge platform is a bit different story. It is 
> prepared to provide HQoS so the buffer space is much larger.
> Very interesting topic to myself is the question raised by Tim - how the 
> box will operate in real network/with native traffic. Just remember - our 
> tests generate artificial conditions.
> So at the moment we can compare low level parameters of the box (good 
> we can verify them) - but other conclusions need some extra assumptions or 
> tests :-) I hope to join the VC today.
> 
> br
> Tomek
> 
> W dniu 08.03.2022 o 17:59, Fr  d  ric LOUI pisze:
>> Hi Tomek,
>> 
>> Many thanks for these tests.
>> I have a question:
>> How does these tests are handled with platform from vendor you had the 
>> chance to have as DUT ? In other words, how does the WEDGE with 
>> RARE/freeRtr compare to a Juniper/Cisco/Nokia/DELL/Cumulus/Broacom routers 
>> you had the chance to test ?
>> 
>> What is your conclusion WRT the RARE/freeRtr platform @ WEDGE: would you 
>> recommend ? Or not ? For which use cases ?
>> 
>> I also would be curious to have the same test with RARE/freeRtr with DPDK.
>> (Granted the fact that the performance will depend on DPDK PMD 
>> maturity)
>> 
>> Anyeway, great work !
>> Frederic
>> 
>>> Le 8 mars 2022    14:13, Tim Chown <> a   crit :
>>> 
>>> Thanks for clarifying Tomek, that makes more sense, though it does raise 
>>> the question of    native    tests.
>>> 
>>> Tim
>>> 
>>>> On 4 Mar 2022, at 11:24, Tomasz Szewczyk <> wrote:
>>>> 
>>>> Hi,
>>>> 
>>>> The setup is like this:
>>>> <ViCfBtcuAvuVG0ov.png>
>>>> However please consider "Egress buffer" as virtual entity. The 
>>>> measurement is independent from buffering method (you can have VOQ as 
>>>> well). The diagram shows just the general idea.
>>>> The second assumption is equal speed of all interfaces - what makes 
>>>> preparation and data interpretation much easier.
>>>> The current run was made of course with 100G interfaces.
>>>> So we do not see packet drops when burst is about 8000 packets. It means 
>>>> the device handles properly the busty traffic for such conditions.
>>>> Please note it is NOT "native" traffic - the whole setup is built 
>>>> in order to estimate buffer size. In order to verify platform 
>>>> operation in native environment/traffic some other tests should be 
>>>> used :-) br Tomek
>>>> 
>>>> 
>>>> W dniu 04.03.2022 o 12:12, Tim Chown pisze:
>>>>> Hi,
>>>>> 
>>>>> Sorry, I didn   t see from the report that there are two ingress and 
>>>>> one egress report, I   m not used to looking at reports like this.  
>>>>> Topology helps.
>>>>> 
>>>>> So does this show that if these are 100G links, and you have 10G on two 
>>>>> ingress ports, there   s drops rather than a clean 20G aggregate on the 
>>>>> egress?  I   d not expect any drops in this scenario, but that may be 
>>>>> naive :)
>>>>> 
>>>>> Tim
>>>>> 
>>>>> 
>>>>>> On 4 Mar 2022, at 11:07, Tomasz Szewczyk <>
>>>>>>  wrote:
>>>>>> 
>>>>>> Hi,
>>>>>> You can se that in 6th step we noticed a number of dropped 
>>>>>> packets. This mean burst size was 1+5x1000 for every ingress 
>>>>>> interface. So we can say that the buffer size is ~8000 packets 
>>>>>> (IMIX size distribution) This is also independent from load (you 
>>>>>> can see similar effect for different load values) br Tomek
>>>>>> 
>>>>>> W dniu 04.03.2022 o 11:56, Tim Chown pisze:
>>>>>> 
>>>>>>> Hi,
>>>>>>> 
>>>>>>> Thanks for the initial test results.  I   m not sure how to interpret 
>>>>>>> them.  I can see there are losses at certain points, but it   s 
>>>>>>> unclear to me what the tests are testing :)
>>>>>>> 
>>>>>>> Tim
>>>>>>> 
>>>>>>> 
>>>>>>>> On 4 Mar 2022, at 09:51, Tomasz Szewczyk <>
>>>>>>>>  wrote:
>>>>>>>> 
>>>>>>>> Hi,
>>>>>>>> I just finished first approach. Please take a look at attchached 
>>>>>>>> report. It was made with Burst size step 1000. Unfortunately this is 
>>>>>>>> not included in tables. First value for each load size is burst 1, 
>>>>>>>> next is 1000 and so on.
>>>>>>>> br
>>>>>>>> Tomek
>>>>>>>> 
>>>>>>>> W dniu 04.03.2022 o 10:44, Xavier Jeannin pisze:
>>>>>>>> 
>>>>>>>>> Hello
>>>>>>>>> 
>>>>>>>>> Le 04/03/2022    10:35, Tim Chown a   crit :
>>>>>>>>> 
>>>>>>>>>> The Internet2 people are about to play with 400G Edgecore!
>>>>>>>>>> 
>>>>>>>>>> Something to consider?  Would Tomek be able to test it with 
>>>>>>>>>> RARE/FreeRtr? :)
>>>>>>>>>> 
>>>>>>>>> FYI, Tomasz is right now restarted the 100G test of 
>>>>>>>>> RARE/freeRtr switch in Pozna  . Csaba kindly helped him 
>>>>>>>>> yesterday to solve an issue related to a laser Regards Xavier
>>>>>>>>> 
>>>>>>>>> 
>>>>>>>>>>> Begin forwarded message:
>>>>>>>>>>> 
>>>>>>>>>>> From: Thomas DeFanti
>>>>>>>>>>> <>
>>>>>>>>>>> 
>>>>>>>>>>> Subject: [PRP-List] Re: PRP/NRP Zoom from 10am to 11am 
>>>>>>>>>>> Pacific on Thursday March 3
>>>>>>>>>>> Date: 3 March 2022 at 18:47:54 GMT
>>>>>>>>>>> To:
>>>>>>>>>>> , Tom DeFanti <>, 
>>>>>>>>>>> 
>>>>>>>>>>> 
>>>>>>>>>>> Reply-To:
>>>>>>>>>>> 
>>>>>>>>>> <snip>
>>>>>>>>>> 
>>>>>>>>>>>  Stordis says a 400G switch is coming.  Will have 400G edgecore 
>>>>>>>>>>> very soon.
>>>>>>>>>>> 
>>>>>>>>>> Tim
>>>>>>>>>> 
>>>>>>>>> --
>>>>>>>>> Xavier JEANNIN
>>>>>>>>> Network Dpt.
>>>>>>>>> GN4-3 Network technology evolution (WP6 T1) task leader T  l. 
>>>>>>>>> : +33 1 5394 2042 Mob. : +33 6 3017 0963
>>>>>>>>> 
>>>>>>>>> 
>>>>>>>>> www.renater.fr
>>>>>>>> --
>>>>>>>> Tomasz Szewczyk
>>>>>>>> Pozna  skie Centrum Superkomputerowo Sieciowe :: Pion 
>>>>>>>> Technologii Sieciowych Poznan Supercomputing and Networking 
>>>>>>>> Center :: Network Technology Division
>>>>>>>> tel: +48 61 858 2040
>>>>>>>> 
>>>>>>>> 
>>>>>>>> 
>>>>>> --
>>>>>> Tomasz Szewczyk
>>>>>> Pozna  skie Centrum Superkomputerowo Sieciowe :: Pion Technologii 
>>>>>> Sieciowych Poznan Supercomputing and Networking Center :: Network 
>>>>>> Technology Division
>>>>>> tel: +48 61 858 2040
>>>>>> 
>>>>>> 
>>>> --
>>>> Tomasz Szewczyk
>>>> Pozna  skie Centrum Superkomputerowo Sieciowe :: Pion Technologii 
>>>> Sieciowych Poznan Supercomputing and Networking Center :: Network 
>>>> Technology Division
>>>> tel: +48 61 858 2040
>>>> 
> 
> --
> Tomasz Szewczyk
> Pozna  skie Centrum Superkomputerowo Sieciowe :: Pion Technologii 
> Sieciowych Poznan Supercomputing and Networking Center :: Network 
> Technology Division
> tel: +48 61 858 2040
> 








-=-=-=-=-=-=-=-=-=-=-=-
Groups.io Links: You receive all messages sent to this group.
View/Reply Online (#247): https://groups.io/g/freertr/message/247
Mute This Topic: https://groups.io/mt/89686813/6413194
Group Owner: 
Unsubscribe: https://groups.io/g/freertr/unsub []
-=-=-=-=-=-=-=-=-=-=-=-