Does anyone know when the switching and router vendors will release their
new models with the Broadcom BCM88370 and BCM88670 chips? It looks like
these chips could be used as a carrier grade router and/or metro E device.
I should dig around for more information around these.
Merchant chips have focused on bandwidth scaling at the expense of key
features available in custom silicon. This has forced me to avoid
certain hardware from even the big vendors.
Bandwidth is not everything... if the approach with this new chip is
different, I'd be interested. Time to hunt...
The BCM88670 (Jericho) is what powers the new Cisco NCS55XX devices. The processor is linerate above around 100 bytes per packet without external TCAM, supports 256K IPv4/64K IPv6 FIB entries (or mixed amounts). These chips are being used for high scale 100G, the initial NCS5508 linecard is a 36x100G QSFP28 one.
Juniper has chosen to use their own silicon for most of their dense 100G platforms, but you’ll see these chips used by pretty much everyone else I imagine at some point in the next year.
Juniper has chosen to use their own silicon for most of their dense 100G platforms, but you’ll see these chips used by pretty much everyone else I imagine at some point in the next year.
Juniper silicon has one big advantage over BCM88670 - it supports 2M FIB entries. This makes PTX1000 (and QFX10002) very attractive platform for SPs.
I was hoping this new Broadcom chip would be able to support enough routes
to hold a full BGP table, and be used for something like cumulus linux. I
have no need for 100G, but 10G and 40G on a platform with deeper buffers
sounds nice.
It does support a path to use an external TCAM if vendors do that, and will support 1M+ entries. It will be more expensive and the datapath will be slower which will impact the performance a bit.
I think you’ll see this make its way into something like a 48x10G/4x100G (or 40G) type platform but we’ll see.
Some points:
1.DNX SDK is significantly different from SGX, adopted by Cumulus and such, yet to be done, and this is not negligible amount of work
2.if you are not interested in capacity but in scale, there’re other BCM chips, perhaps more suitable
3.you don’t have to have all the forwarding entries populated in silicon, as an example - take a look at http://sdn-internet-router-sir.readthedocs.org, code at https://github.com/dbarrosop/sir, one could also leverage approach we have taken in EVPN - decoupling RIB from FIB completely
4.NG silicon will do 1M+ LPM's
Good point, there are many people looking at what I call FIB optimization right now. The key is having the programmability on the device to make it happen. Juniper/Cisco support it using policies to filter RIB->FIB and I believe both also do per-NPU/PFE localized FIBs now. I am not sure if that’s something supported on this new Broadcom chipset. Depends on your network of course and where you are looking to position the router.
Yes. We also have 1M+ FIB support day one too - hence the letter 'R'
denoting the evolution with 3rd generation of its evolution to internet
edge/router use cases.
Not sure what other vendors are doing but I doubt others are yet shipping
large table support.
(there's more to it than just the underlying native silicon)
there’s a phenomenon called “next-hop flattening” which has to do with lookup recursiveness within the silicon.
Unless this is done (and this is big piece of work) not everything supported on Trio or Ezchip can be supported.
In general – Jericho (and its followers) is a great piece of silicon made by clueful folks… watch this space closely
So can this compete routing wise against something like a Juniper MX104 or
Cisco ASR 9001?
Super broad question. Generally they are not targeting same
applications. MX104 (Trio) and ASR9001 (EZchip) are high-touch chips,
where-as Jeiricho is low-touch chip. What it can do, it can do fast
and economically.