dai wen jun wrote: > Hi,everyone. > > I got something from > http://www.opensolaris.org/os/community/advocacy/events/current_tech_days/frankfurt/osol-netadmin.pdf. > 802.3ad Link Aggregation support is even worse: >> Some links are aggregated with dladm(1M) >> Others are aggregated with the unbundled nettr(1M) >> Many cannot be aggregated at all!
That hasn't been true in a long time. The presentation you're looking at is dated December 2007, and the projects that fixed the problem were well underway when that presentation was given. The point of it was to describe the motivation for why we were fixing things -- not to provide a disincentive for using OpenSolaris, or to describe an inescapable state of decrepitude. > And I test OpenSolaris b118 with four nic card as a aggregation.And the > swith is configured as a active lacp trunk group.Last I got two 100MB and two > 50MB bandwidth. I trusted my configure was right.Can somebody tell me the > aggregation status now. > My nic card is Intel PRO/e1000g and switch is BLADE G8000. > Thank you very much. Your best bets, if you're having trouble with driver performance, would be to contact the driver group ([email protected]) and/or filing a bug against the driver in question. Given the numbers you're quoting, I don't suspect that you have a problem with aggregation specifically. Note, of course, that aggregations do not and cannot (per the standards) spread out the traffic from a single flow, which means that if you aggregate four 1Gbps ports, your maximum throughput for a single flow will still be 1Gbps (or, equivalently, slightly over 100MBps). What increases is the potential combined rates for multiple flows. And more flows are better, as the algorithms used to spread out the traffic are statistical in nature. Note also that Ethernets and telecom circuits use base 10 for speeds, but that data processing generally uses powers of 2 for data quantity and rates. This means that -- ignoring protocol overhead completely -- a 1Gbps rate would actually be 119.2MBps and not the 125MBps you might expect by dividing 1000Mbps by 8. (And Ethernet and TCP/IP protocol overhead will generally steal about 4.5% from that, so figure 113Mbps or so as the real maximum.) -- James Carlson 42.703N 71.076W <[email protected]> _______________________________________________ networking-discuss mailing list [email protected]
