On (2012-10-24 01:12 +0400), Tima Maryin wrote: > Check KB24339 also > (Hash key computation on MPC cards)
Is this accurate? ----- Payload recognition uses the following criteria to estimate the nature of the payload: IPv4 assumes the first nibble, which follows the bottom label, is 0x4. If this determination is made, then the fields in the IPv4 section are used to load-balance the payload. ------ If this is true and you have RFC4448 pseudowire without control word, and your DMAC starts with 4, you're going hash it as IPv4, which will lead to reordering, which will cause TCP performance to sink. [ytti@sci ~]% egrep "^4.-" oui.txt|wc -l 257 257 OUI starting with 4 assigned, including Juniper, Cisco, ALU, HP, Intel, Apple, Samsung, Sony, RIM, Nokia, Nintendo, Motorola, Nortel, IBM, ZTE, Netgear. So there is non-zero chance of having this situation. I hope the BB is wrong or I'm misunderstanding it. -- ++ytti _______________________________________________ juniper-nsp mailing list juniper-nsp@puck.nether.net https://puck.nether.net/mailman/listinfo/juniper-nsp