PTX1k is based on the paradise chipset and uses only LPM. The lookup memory is HMC based and is entirely on-chip. ( that’s at least the fpc3/ptx5k behavior, i’m relatively sure it’s the same for ptx1k without having checked it explicitly)
The issue i have is not the black art part…the thing with the current implementation in EOS is that it’s highly dependent on the prefix distribution and you have to monitor in order to not shoot you in your feet… And given the current capacity limits of Jericho you probably will see issues more in the 700-800k prefix range than in the claimed 1,2m FIB entries. Assuming the current FIB size in the DFZ, * I * would want to know the exact limits and have some reasonable margins. Will that change? sure… but today it is like it is and *I* would not bet my job on it… but again, your milage may vary... /hannes > On 17 Dec 2016, at 19:46, Saku Ytti <s...@ytti.fi> wrote: > > On 17 December 2016 at 19:18, Hannes Viertel <hvier...@bastelbude.net> wrote: > > Hey, > >> the PTX1k with the routing license can run both easily… i could give >> another couple of examples where both Juniper and Cisco charge you a premium >> but in return you get a grown-up OS with a relative complete feature list… ( >> and yes sure you hit PR’s here and there.., but i believe we all agree there >> is no bug free sw in the market ) >> >> but again, if the black art of flex route magic works for you and you don’t >> plan to have more than 2m-3m RIB routes while doing no sophisticated mpls >> scenarios like PCE initiated RSVP-TE i think it can be feasible. If you want >> to run it as a peering router on a large IXP with traffic engineering in >> your core * i * would place a ptx1k there. > > NCS5k (non SE) does same 'black art'. And I'm not sure why it's > brought up to be so big issue, I'm sure vendors have always tried to > make software which most efficiently uses available HW. Heck, we're > still figuring out novel ways to do stuff in C64. > > The fact that Arista and NCS5k use exact lookups and LPM lookups in > parallel to scale FIB size IMHO shouldn't be held against (or for) > them. Do we know how PTX1k works? I don't, but on that basis who am I > to say NCS/Arista work in 'black art' way, when I have no idea how > PTX1k works. > > If I were to guess, I'd guess PTX1k does only exact match lookups and > uses on-chip bloom filters to know which external mem exact match > table(s) to consult. If so, then it's also doing pretty similar thing > as Arista/NCS5k, only arguably in far more complex or 'black artsy' > way. > > > -- > ++ytti _______________________________________________ juniper-nsp mailing list juniper-nsp@puck.nether.net https://puck.nether.net/mailman/listinfo/juniper-nsp
