7/95 FC-AL Direct Attach Disk Adhoc Minutes
kc at core.rose.hp.com
Tue Jul 11 17:56:31 PDT 1995
From: Kurt Chan , HP
To: FC, SCSI Reflectors
Subj: 7/95 FC-AL Direct Attach Disk Adhoc Minutes
FC-AL Direct Attach Disk Adhoc Meeting Minutes
Colorado Springs, CO
Dal Allan ENDL dal.allan at mcimail.com
Charles Binford Symbios Logic charles.binford at symbios.com
Kurt Chan HP Roseville kc at core.rose.hp.com
Mike Chenery Fujutsu mchenery at fcpa.fujitsu.com
Jim Coomes Seagate jim_coomes at notes.seagate.com
Tom Coughlan Digital coughlan at star.enet.dec.com
Edward Fong Amdahl edward_fong at spg.amdahl.com
Giles Frazier IBM gfrazier at ausvm6.vnet.ibm.com
Gene Freeman Symbios Logic gene.freeman at symbios.com
Ed Gardner Quantum gardner at acm.org
Doug Hagerman Digital hagerman at starch.enet.dec.com
Norm Harris Adaptec nharris at eng.adaptec.com
Bill Hutchison HP hutch at boi.hp.com
Skip Jones Qlogic sk_jones at qlc.com
Jean Kodama QLogic j_kodama at qlc.com
Mark Lewis Digital lewis at ssdevo.enet.dec.com
Gene Milligan Seagate gene_milligan at notes.seagate.com
Charles Monia Digital monia at shr.dec.com
Al Pease Buslogic alp at buslogic.com
Jim Pherson Digital pherson at ssdevo.enet.dec.com
Lansing Sloan LLNL ljsloan at llnl.gov
Bob Snively Sun bob.snively at sun.com
Arlan Stone Unisys arlan.stone at mv.unisys.com
Horst Truestedt IBM truested at vnet.ibm.com
Michael Walker Digital walker at peaks.enet.dec.com
Stewart Wyatt HP DMD stewart at hpbs3928.boi.hp.com
Anthony Yang Hitachi yang_a at halsp.hitachi.com
There are no future FC-AL direct attach disk adhoc meetings scheduled.
The next two FC-AL2 meetings will be held Friday mornings of X3T11 weeks:
Fri 8/11, 830am Tarrytown, NY
Fri 10/6, 830am Toronto
The Monday slot during X3T10 weeks will no longer be used for either
profile or FC-AL meetings.
2. DOCUMENT LOGISTICS
Revision 1.90 of the Direct Attach profile was distributed, and revision
2.00 will be the version forwarded to X3T11 as a Technical Report (TR).
The postscript for the profile is available via anonymous ftp on
ftp.symbios.com unden pub/standards/io/fc/prv_*.ps. Dal distributed
copies of an X3T11 project proposal for creating the TR.
Any future discussions kegarding the profile will take place on the
disk_attach reflector (disk_attach at dt.wdc.com). To subscribe or
unsubscribe, send email to majordomo at dt.wdc.com with a blank subject line
and one of the following text lines:
subscribe disk_attach your_email_address
unsubscribe disk_attach your_email_address
When posting to the reflector, please include a header which includes a
"From:" and "To: disk_attach" field so those of us who subscribe to
multiple reflectors know which reflector the posting is associated with.
3. TIMERS AND TIMEOUTS
Version 1.90 reevaluates the timers needed for interoperability on private
loops. There are three "interface" times defined:
A more liberal AL_TIME (called PL_TIME) was agreed upon which assumes
an *average* inter-node cable segment length of 500m rather than 10km.
This gives a total loop length of 67km (42 miles) for a 134-node loop
rather than the unrealistic worst-case of 1,340km (837 miles) assumed
by FC-AL. As a result, PL_TIME is 1ms whereas AL_TIME is 15ms.
LIS_HOLD_TIME was defined as the maximum amount of time a node shall
hold a Loop Initialization Sequence before forwarding it to the next
node. The value chosen is 1ms.
LM_TOV is the minimum amount of time the Loop Master waits for a Loop
Initialization Sequence to be returned before assuming the loop is not
functioning. The value includes at least 126 AL_TIMEs and 126
LIS_HOLD_TIMEs, and is chosen to be 300ms.
There are 4 initiator times (in addition to a ULP timer): AW_TIME,
SRR_TIME, E_D_TOV, and R_A_TOV.
AW_TIME is the minimum period of time an NL_Port shall wait to win
arbitration before it may assume a malfunction or excessive unfairness.
The value is 1 second, which takes into account 126 nodes all
transmitting 64k bytes per loop tenancy at 266Mb/s, and each node
unfairly arbitrating once in addition to its initial arbitration.
SRR_TIME is the maximum time an Initiator assumes the Target will take
to respond to a transfer of Sequence Initiative before invoking error
recovery (after also taking into account physical propagation delay and
arbitration time). The value may be command-dependent,
workload-dependent, and/or device-dependent. Initiators may choose not
to define this time specifically, but group it into a ULP timer.
E_D_TOV is the minimum time an Initiator shall wait between the time it
transmits a frame requesting transfer of Sequence Initiative and the
time it receives a frame within the same Exchange which acknowledges
the transfer of Sequence Initiative before invoking error recovery. It
includes at least one AW_TIME, one SRR_TIME, and two PL_TIMEs. The
value chosen is 2 seconds. Initiators may elect not to define this
specifically, but group it into a ULP timer.
R_A_TOV is equal to zero since private loops cannot misorder frames.
Following receipt of the ACC to ABTS, Initiators may immediately
A ULP timeout value is the time between the ULP dispatches a SCSI
command to the interface and the time the ULP receives SCSI status.
It shall be greater than or equal to E_D_TOV.
There is one Target timer:
R_R_TOV is a resource recovery timer used by multi-initiator targets
to reclaim resources which the target has dedicated to initiators
which have since "failed." It is the minimum time a target shall
a) ADISC/PDISC from an initiator which has an open task, following
receipt of LIP, or
b) an initiator response to a transfer of Sequence Initiative
before implicitly logging out that initiator, unless an ACA condition
is in effect for that initiator (in which case R_R_TOV is infinite).
The value chosen is 2 seconds.
4. EXCHANGE AUTHENTICATION
Rather than require Targets to originate ADISC or PDISC upon expiration
of RR_TOV, the Targets will instead be required to:
a) suspend all open tasks until the initiators with open tasks complete
the ADISC/PDISC authentication protocol, and
b) implicitly log out any initiators which do not send an ADISC or
PDISC no earlier than RR_TOV after the target receives a LIP.
5. LOGIN VALIDATION AND IMPLICIT LOGOUT
The examples for login validation were clarified to require that when an
ADISC or PDISC is sent to a target, all three identifiers (Port name,
Node name, N_Port ID) must match a logged in initiator in the target or
else LOGO will be returned to the initiator in response to the ADISC or
If a target supports 'N' logins and receives an N+1'th PLOGI request,
a) if there were no open tasks with the least recently used initiator,
that initiator may be implicitly logged out.
b) if there are open tasks with all 'N' initiators, the login
request shall be returned with LS_RJT and reason code "unable to
perform command request."
6. FCP_RSP PAYLOAD
Single-lun targets are restricted to 128-byte FCP_RSP payloads. Hosts
which ONLY support single-lun direct access targets may rely on no
FCP_RSP payload exceeding this value. However, hosts which support
other devices or device types (e.g., controller devices) must assume
those devices can return the full 255 bytes permitted by SCSI-3.
To remain consistent with the FCSI profiles, initiators are prohibited
from requiring targets to implement ACA. Prior to this, targets which
did not implement ACA and Initiators which required it could not
interoperate. Other profile may have additional requirements regarding
host notification of configuration or logical block size changes in
order to preserve data integrity without the use of ACA (see FCSI SCSI
8. MULTI-LUN DISK
Some high-capacity (8+Gbyte) drives must use multiple LUNs in order to
make full use of their capacity on some operating systems. Charles
Binford asked the question of where the LUN should be specified in the
FCP_LUN field of the FCP_CMND payload. The consensus was to use the
"first level addressing" field in the SCC command set, and to report
"peripheral device addressing method" (00b) per Table 5 in SCC.
More information about the T10