RARE user and assistance email list

[Tim Chown <>]

Hi,

Tomek is looking at re-running performance tests for RARE at 100G; this was 
discussed in today’s WB performance call.

The DMC testing, I don’t think Joseph tested the box explicitly as such, but 
we have the throughput figures that the teams obtained to/from the UvA DTN.

I think the idea is a single report from our collective 100G-related testing, 
for which each of these would be one part.

Tim

> On 10 Feb 2022, at 14:00, Ivana Golub <> wrote:
> 
> Hello,
> 
> Can we get a report from 100G RARE testing as a part of the DMC21 in the 
> form of a white paper?
> 
> Thanks 
> Ivana
> --
> Ivana Golub PhD
> ICHB PAN - PSNC
> GÉANT 4-3 Project
> Network Technologies and Services Development Work Package Leader
> 
> 
> On Thu, 10 Feb 2022 at 14:10, Tim Chown <> wrote:
> > On 7 Feb 2022, at 17:57, mc36 <> wrote:
> > 
> > hi,
> > 
> > On 2/7/22 18:39, Tim Chown wrote:
> >> Hi,
> >> Well, our use case is fairly simple.  The Wedge would act as a 100G 
> >> gateway to host four 100G connected servers that we   d use to test 
> >> perfSONAR and various data transfers tools on DTNs.  This would be a 
> >> static setup, with maybe a single v4 and v6 subnet on the server side.   
> >> So BGP seems overkill, but our engineers asked what RARE/FreeRtr 
> >> supports.
> > basically everything is supported that you expect from a cisco/junos 
> > box...
> > if it's bgp, or even bgp-lu, with a single prefix, then that...
> > if it's isis, then we\re part of geant for a year, part of local nren's 
> > ospf and fullbgp,
> > keeping the geant's mdvpn bgp-lu + vpnv4/vpnv6/vpls routes from bmp 
> > flawlessly,
> > so have no concerns, let your engs imagine anything, we already support 
> > it :)
> 
> OK, thanks, but that will be overkill :)
> 
> >> What matters most is the performance, hence our interest in the tests by 
> >> PSNC and on the Data Mover Challenge.  We   ve seen high rates from 
> >> Joseph   s device at UvA, but inconclusive results from PSNC (they 
> >> probably need to be more tightly defined).  We think in a scenario where 
> >> there   s generally a limited number of concurrent flows, the small 
> >> Wedge buffers (16 or 20MB?0 should be OK.
> > so in psnc they did the measures over lr over cwdm which, in terms of 
> > inter-packet times and drops are not a good choice...
> > i remember clearly that initially we had 1 crcing lane out of 4, and 
> > after the swaps, the levels fairly was under warning...
> 
> So what does that mean in practice if we want to use a Wedge running 
> RARE/FreeRtr as an access device for 100G servers?   The results Jospeh saw 
> for his device for the Data Mover Challenge were decent.
> 
> Tomek is preparing new tests for RARE.
> 
> >> There   s no initial need for multicast, PolKa or anything like that.
> > ahh, pick any profile from the available ones depending on what you want 
> > to test...
> > and if your engs complain that this or that table is too small, just drop 
> > a template
> > over the fence with your goals and we'll optimize for that too... it's 
> > about 30m per
> > profile and need to be done once per p4 change (which is rare nowadays)
> 
> OK, I don’t know those requirements yet.  The routing could be static, 
> really, given the simplicity of the topology.
> 
> >> Raul and David are checking QSFPs and cables this week, so we   re 
> >> getting closer to being ready to deploy.
> > sounds goods!
> 
> Hopefully they’ll be at a Tuesday community call before too long.
> 
> Tim
> 
> > 
> > regards,
> > cs
> > 
> > 
> >> Tim
> >>> On 7 Feb 2022, at 15:23, mc36 <> wrote:
> >>> 
> >>> hi,
> >>> exactly, it depends on the exact features you need...
> >>> this is the ultimate source of thruth:
> > >> https://bitbucket.software.geant.org/projects/RARE/repos/rare/browse/profiles
> >>> the .tmpl files are the 'golas' that you have to set yourself...
> >>> it's a two step optimization, first for the tcam, then for the sram...
> >>> this is an ip-only profile with nothing else but the ipv4/ipv6 fib 
> >>> enabled:
> >>> https://bitbucket.software.geant.org/projects/RARE/repos/rare/browse/profiles/profile-rawip.tmpl
> >>> https://bitbucket.software.geant.org/projects/RARE/repos/rare/browse/profiles/profile-rawip.p4
> >>> and here is an other with mpls enabled:
> >>> https://bitbucket.software.geant.org/projects/RARE/repos/rare/browse/profiles/profile-p.tmpl
> >>> https://bitbucket.software.geant.org/projects/RARE/repos/rare/browse/profiles/profile-p.p4
> >>> as flou said, you should set your own features and goals and then it'll 
> >>> optimize for that specific profile...
> >>> one thing to keep in mind that both sram and tcam table increase could 
> >>> span to multiple code stages,
> >>> resulting in lower tables so choose the primary goal according to your 
> >>> needs and let the secondary
> >>> goal span the remaining asic code space...
> >>> regards,
> >>> cs
> >>> 
> >>> 
> >>> 
> >>> 
> >>> 
> >>> On 2/7/22 15:15, Fr  d  ric LOUI wrote:
> >>>> Let me add Csaba as he is tweaking RARE profiles at teh moment.
> >>>> The answer I presume will depends on your use case.
> >>>>> Le 7 f  vr. 2022    14:51, Tim Chown <> a   crit :
> >>>>> 
> >>>>> Hi,
> >>>>> 
> >>>>> What size BGP table can the Wedge take running RARE?
> >>>>> 
> >>>>> i.e. should we suggest our engineers configure static routing?
> >>>>> 
> >>>>> Raul and David (our engineer) will be looking at the Wedge, servers 
> >>>>> and cabling in their current location this week, as the plan to 
> >>>>> deploy 100G in London gets nearer
> >>>>> Thanks,
> >>>>> Tim
>