Re: iSCSI : IscsiSessionStatsEntry MIB OBject contents

[Gwendal Grignou <ggrignou@silverbacksystems.com>]

Thomas,

I think the definition of MIBs fields should be more precise, so that all 
initiators and targets will store the same information, to allow useful 
comparison.

My comments inline :
> Login PDU and Login Response PDU are not counted in iscsiSsnRspPdus or
> iscsiSsnCmdPdus because the session is not instantiated until full feature
> phase (after Login Response).
I don't see why we should not count them.
Within the session object, you have connection objects, each of them with 
"iscsiConnectionAttributesEntry". One of the fields of this sequence is " 
iscsiCxnState", which can be "login",... So as soon as a connection is 
established at TCP level, the connection object (and the session object it 
belongs to) should be instantiated.
> There are separate counters for Login
> attempts - see the ...LoginStatsEntry definitions. 
Those counters do not count the PDUs exchanged between initiator and target, 
but store the result of each login attempts.
> All other PDUs are
> counted, including Data In and Data Out.  I suppose we could discuss
> whether Logout and Logout Response should be counted, since they have
> separate counters in the MIB as well (see the ...LogoutStatsEntry
> definitions).


> As for the Data Octet counters, the easiest interpretation is that it
> refers to all octets in all PDU Data Segments.  I think the count will be
> most useful if it counts only the bytes of data actually read from or
> written to the SCSI device (e.g., it includes only the Data Segments of
> Data Out and Data In PDUs).  However, I suspect that the sum total of all
> other PDU Data Segments will be so small compared to the Data Out and Data
> In that it would not hurt to lump those in.
Well, to have a clear definition would ensure the counters for a given 
session have the same value on both the target and the initiator.

Gwendal.