[187512] in North American Network Operators' Group
=?utf-8?Q?Re=3A_Low_density_Juniper_=28or_alternative=29_Edge?=
daemon@ATHENA.MIT.EDU (Shawn L)
Wed Feb 3 13:20:44 2016
X-Original-To: nanog@nanog.org
Date: Wed, 3 Feb 2016 10:12:42 -0500 (EST)
From: "Shawn L" <shawnl@up.net>
To: "NANOG" <nanog@nanog.org>
In-Reply-To: <CAMDdSzM=DdUfuvecCnX01-KGcHE=a0D9dCtcdgBpEnZ3FOqf7Q@mail.gmail.com>
Errors-To: nanog-bounces@nanog.org
=0AWe use the Accedian Metro Nid in places. They work well, but are layer =
2 only -- at least the ones we got. =0A =0A =0A-----Original Message-----=
=0AFrom: "Colton Conor" <colton.conor@gmail.com>=0ASent: Wednesday, Februar=
y 3, 2016 9:34am=0ATo: "Nick Hilliard" <nick@foobar.org>=0ACc: "NANOG" <nan=
og@nanog.org>=0ASubject: Re: Low density Juniper (or alternative) Edge=0A=
=0A=0A=0AI see Cisco and Juniper mentioned here, but what about all the sma=
rt NID=0Acompanies out there? I found these of MEF list:=0A=0AAccedian, Alt=
era, BTI Systems, Ciena (Nasdaq: CIEN=0A<http://www.fiercetelecom.com/tags/=
ciena>), Cisco (Nasdaq: CSCO=0A<http://www.fiercetelecom.com/tags/cisco>), =
Cyan, FibroLAN, Huawei,=0AInfinera (Nasdaq: INFN <http://www.fiercetelecom.=
com/tags/infinera>),=0AJuniper Networks (NYSE: JNPR=0A<http://www.fiercetel=
ecom.com/tags/juniper-networks>), MRV, Omnitron,=0AOverture, PT Inovacao, P=
ulsecom, RAD Data Communications, Telco Systems,=0ATellabs (Nasdaq: TLAB <h=
ttp://www.fiercetelecom.com/tags/tellabs>),=0ATransition Networks and Trans=
mode.=0A=0ASome of these guys focus what seems like exclusively on ethernet=
NID=0Adevices, and most all are MEF certified. Does anyone use the above v=
endors=0ANIDs?=0A=0A=0A=0AOn Wed, Feb 3, 2016 at 1:58 AM, Nick Hilliard <ni=
ck@foobar.org> wrote:=0A=0A> David Bass wrote:=0A> > Looking to see what ot=
hers are using out there as an alternative to a=0A> > Cisco ME3600X? Also, =
what other vendors out there are playing in this=0A> > space?=0A> >=0A> > N=
eed a full MPLS stack.=0A>=0A> Before choosing a box, you need to figure ou=
t:=0A>=0A> - how many ports you need, and of what speed=0A> - how much you'=
re prepared to pay=0A> - how much rack real estate you're ok about dedicati=
ng per box=0A> - what sort of mpls features you need (vpls / l2vpn-pw / l3v=
pn / 6pe /=0A> 6vpe, etc)=0A> - whether rich qos is a requirement=0A> - whe=
ther you're ever going to need good quality LAG / ECMP support on=0A> the p=
latform=0A> - what vendor software you're happy to work with=0A> - whether =
you're ok with per port licensing=0A>=0A> Typically the features that fall =
by the wayside first are: reasonable=0A> port buffers, qos knobs and decent=
lag/ecmp hashing support for mpls=0A> packets. The qos/port buffers tend t=
o be more of a problem on the 10G=0A> platforms, but you didn't state wheth=
er you were interested in 1G or=0A> 10G, or how many ports you were looking=
for per box.=0A>=0A> E.g. the production evolution for the me3600 is the a=
sr920, which is=0A> better is most aspects except for shared buffer space. =
This means that=0A> the me3600 has better qos support, if deeper buffers ar=
e what's=0A> important. OTOH, if you need to do fine-grained qos based on A=
CLs or=0A> ports, then this platform isn't for you.=0A>=0A> Most smaller mp=
ls boxes don't load balance well over LAGs or ECMP=0A> because they lack th=
e ability to inspect deep into the packet to get=0A> enough flow-aware entr=
opy together to build a reasonable hash. If all=0A> your PE devices support=
flow-aware transport (rfc6391), you're fine, but=0A> very few smaller mpls=
boxes support this feature.=0A>=0A> If 10G is a requirement, then you need=
to make a choice between one of=0A> the merchant chipsets (e.g. broadcom t=
rident range) and vendor specific=0A> chipsets. Many of the larger vendors =
support the merchant chipsets=0A> these days for 10G access, but feature su=
pport can be varied. E.g. some=0A> devices don't support vpls and never wil=
l. Some are a bit behind on=0A> product development and don't yet support f=
eatures like l3vpn or 6PE or=0A> 6VPE, even though they are roadmapped.=0A>=
=0A> Nick=0A>
