[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
RE: iSCSI: SCSI Cmd PDU larger than 48 bytes
This is a reasonable question. Here is my opinion.
Firstly the occurrence of such commands in most iSCSI sessions is not a random function with predictable distribution.
For a target, the likelihood of a bi-directional command or a CDB larger than 16 bytes is near zero unless the target supports such commands. Many of today's targets will support (understand and execute) neither bi-directional nor large CDB commands.
For an initiator (HBA), the probability is zero until support is added to upper layer (device class) software to be able to send such commands. Further the OS's interface to the HBA must comprehend and permit such requests to be passed from the device class drivers to the HBA drivers. The HBA might not know whether device drivers will send such commands until they do so.
Still these can not be effectively used unless the same customer has both. For a bridge device which might be translating between iSCSI and Fibre Channel, again there must be support in both end points. Otherwise the command will not be generated, or when generated and delivered, will be rejected. Hopefully after the first such rejected command, additional similar commands will be very few and far between.
There are SCSI commands defined today which require long CDBs or bi-directional data transfers and there are more commands being proposed which would require one or both. At least one major vendor has already built devices which use them. They may be widely used someday. They are not yet "main-stream" or "common place".
If a customer has a device which supports or depends on this functionality and an initiator which can use them, iSCSI must be able to deliver them. In sessions with such a device, a high percentage of the commands could fall into these categories.
Last updated: Mon Jun 10 11:18:47 2002
10623 messages in chronological order