|
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index] Re: iscsi : default iscsi mode page settings.As a futher complexity nit.... If the iSCSI layer must snoop mode select commands - it must snoop the entire i/o from end to end (cmd, data and response). If the target or attached SCSI layer rejects the mode select, it must not change operational parameters until the accepted status is received. Under the appropriate check condition, the target may actually accept the change, but rounded the values - and the iSCSI layer then is dependent on another Mode Select being issued to make sure it has the right value... Nasty for normal target/initiator implementations, very ugly for bridging functions. I am firmly in the camp that iscsi transport parameters be maintained by iscsi mechanisms - thus it should be a login key. Although I won't win this one - I further believe that these things should not be reflected in SCSI mode pages at all, for all the application/compatibility/multi-initiator/snooping reasons pointed out. Besides, iSCSI should not be as dependent upon SCSI pass through (thus, I assume, the use of mode pages) for device management as previous transports were (iSCSI should have SNMP support present). It'd be nice to have iSCSI put a stake in the ground and clean up this issue rather than propagate it. -- James "Sanjeev Bhagat (TRIPACE/Zoetermeer)" wrote: > Julo, John J&J :), > > So I see that you basically come to the point that FirstBurstSize and > MaxBurstSize field be put into the login keys. If you agree this, and if i > have to see the SCSI mode page parameters, the only odd field left out is > EMDP (enable modify data pointer). This should also be included in > login/text keys. Why should this be left out alone? The inititator will need > to make an extra mode sense command to know this. We should either keep all > three parameters readable through mode sense command or login /text keys or > both. > > By the way in my view there is no SCSI (CDB) command to set transport mode > page parameters. Also a zero copy buffer operation to SCSI is only possible > if there is no missing data in the buffer. > > Regards, > Sanjeev >
Home Last updated: Wed Sep 26 16:17:18 2001 6773 messages in chronological order |