Daniel, Le 07/03/2017 à 15:48, Daniel Stamatov a écrit : > "I'd say it's a minor one actually, cause honestly who has time to > understand > architectural differences between various generations of NPUs/PFEs and how > those affect the suitability of the platform for a given project, or even > how routers work in general and there're no fancy certificates rewarding > this specific knowledge anyways." > > Huh?
He's right. Most vendors don't disclose enough informations to help us understand how their platforms actually works. We can only hope it will perform up to spec, with no mean to verify it beforehands. It probably doesn't matter to most of us 'til we hit a bug or limitation caused by a serious design flaw. It also causes a serious trust issue. Of course, most experienced engineers won't doubt the PTX1000 and QFX10002 are the exact same platform despite their chipsets beeing called differently. So why is it priced with a 1 to 10 difference ? Another example is how opaque Juniper is regarding factual limits of some hardware built over generic ASICs, such as the QFX5100. I found no trace of actual CAM partionning limits on any publicly available document. The publicly available numbers are misleading. Had to find out wich broadcom chip was inside and match to vendor's specs to finaly know what that model was really capable of doing. F**k it, I'm now buying Edge-Core's instead. -- Jérôme Nicolle _______________________________________________ juniper-nsp mailing list juniper-nsp@puck.nether.net https://puck.nether.net/mailman/listinfo/juniper-nsp