[email protected] wrote: > On (01/19/09 11:55), Jason King wrote: > >> Well I guess I should step back and ask how common is it for hw to >> support lso, lro, etc. for things other than IP? My thinking was that >> while a property of the physical NIC, most seem (on initial appearance >> at least) are specific to TCP/IP, thus might be more appropriate to >> group with other tcp/ip administration 'stuff' while things like >> duplex, speed, are always relevant regardless of what's running on top >> of the NIC. >> > > This is getting into the realm of philosophy, but the rationale > was not so much to have ipadm, ipxadm etc. sitting atop dladm, > but rather to have each 'foo'adm target configurations of > parameters at the 'foo'layer. In this case, LSO and Jumbo MTU support > are both data-link properties which is why I feel that dladm > is the more appropriate tool. >
I just did a tests on an e1000g machine, and found m_getcapab() was called when unplumbing/plumbing. So the current drivers should at least support HW checksum, LSO on/off through unplumbing/plumbing. I agree with you. From a long term's viewpoint, I think dladm is the way to go since 1. LSO and HW checksum can be disabled/enabled at runtime. 2. dladm provides more consistent interfaces for these properties. What's your oponion in providing standard interfaces for LSO and HW checksum or private ones in dladm? Miles Xu > --Sowmini > _______________________________________________ > networking-discuss mailing list > [email protected] > _______________________________________________ networking-discuss mailing list [email protected]
