Electrical computers and digital processing systems: multicomput – Computer network managing
Reexamination Certificate
1998-09-15
2003-11-04
Wiley, David (Department: 2142)
Electrical computers and digital processing systems: multicomput
Computer network managing
C709S221000, C709S227000, C714S002000, C714S004110
Reexamination Certificate
active
06643693
ABSTRACT:
TECHNICAL FIELD OF THE INVENTION
The present invention relates generally to data and information communications systems and their operation and more particularly to communications networks including a Fibre Channel network. Even more particularly, the present invention relates to a Fibre Channel network system and method of operation having the ability to maintain the target and initiator states across address changes for improving the management of inputs and outputs to improve network performance, eliminate communication failures, and avoid the loss of data that may occur as a result of a network link state change.
BACKGROUND OF THE INVENTION
In a Fibre Channel network, the network address assigned to initiators and targets may be dynamic. After a link state change on the network, the network addresses assigned to existing nodes in the network can change. This behavior executes a problem for SCSI initiators and targets in managing I/O (inputs/outputs) in progress between a specific initiator-target pair across link state changes.
With known methods for managing I/Os in a Fibre Channel network, when an existing I/O is in progress and either an initiator or target detects a link state change, the initiator must authenticate that it is still talking to the same target before the I/O can continue. This authentication is detailed in The Fibre Channel PLDA (Private Loop Direct Attach) Profile. This profile is an extension to the SCSI-3 Standard FCP (Fibre Channel Protocol), which defines SCSI implementation over Fibre Channel. The PLDA profile states that after a link state change from a link down to link up condition, the initiator must send an ADISC (Address Discovery) or PDISC (Port Discovery) ELS (Extended Link Service) command to every target with which it has I/O's in progress. The ADISC and PDISC commands are used to exchange node name and port name information. The PLDA profile further requires that if the target detects that the network address of an initiator has changed, then the target should terminate the existing I/Os from that initiator. After this step, the initiator performs a recovery process and in some other way restarts the I/O transmission.
The termination of existing I/Os because of a network address change is disruptive. This termination reduces the overall performance of the network, because it unnecessarily slows system or network traffic. In some applications, such as where a tape drive serves as the target for a SCSI initiator, a general application failure may result. Moreover, loss of data results in many instances where this I/O termination process must occur.
In Fibre Channel, initiators talk to targets using network address (PID's). These are called AL-PA's, Arbitrated Loop Physical Address, in an arbitrated Loop environment and PID's in other topology. Each target and each initiator is assigned a PID. In essence, targets and initiators only know one another in the Fibre Channel network by their network address.
In the event of a link state change, such as when a link goes down and then comes back up, or when a new node is added into the network, the network address of either or both the initiator or target can change. The initiator may be, for example, a host personal computer that initiates a SCSI command. A target may be some type of SCSI device such as a CD-ROM, a tape drive, or other storage device. If a state change occurs in the network while the I/Os are still in the process of being transmitted, a problem may arise upon the resumption of transmission. If a link state change occurs an address change could result. Because the initiator and target conventionally only use the addresses for identification, a confused state can arise in the I/O transmission. This confused state could have numerous manifestations, not the least of which would be the generation of corrupted data, loss of security, or generally low performance of the network.
The PLDA (Private Loop Direct Attach) Profile addresses this problem by assigning to each target and each initiator in a Fibre Channel network loop a unique node name and a unique port name, each of which are represented as 64-bit identifiers. The node name for the initiator or target defines the node for the particular device, while the port name defines the particular port on the identified node. Note, however, that a given node may possess numerous ports. Accordingly, the port name uniquely identifies the desired port on the given node.
The initiator and target exchange the node name and port name during initial communication, as defined in the FC-PH (Fibre Channel Physical) Standard. The node name and port name, when combined with the PID, uniquely identify a node. In the PLDA profile, after a link state change, the initiators send either an ADISC (Address Discovery) or PDISC (Port Discovery) command. The ADISC or PDISC command communicates between the target and initiator, the port name, node name, and PID for identification. The target and initiator compare the node name, port name, and PID, provided during initial communication to the node name, port name, and PID, provided by the ADISC or PDISC command. If they are the same, then the same target and same initiator are communicating with one another both before and after the state change. If not, then the addresses have changed.
According to the PLDA profile, the set of data including the PID, the node name, and the port name is called a “triplet.” Thus, a triplet in a Fibre Channel network uniquely identifies a network node. In a Fibre Channel network, the PID is the address from which the ADISC or PDISC is received. The ADISC or PDISC, therefore, includes the port name and node name within their data payload.
The initiator sends out the ADISC or PDISC. In response, the target sends to the initiator an “accept” command, which operates as essentially a “handshake” that includes the same information as the ADISC or PDISC command. In addition, the target sends to the initiator the target's PID address together with the target's port name and node name.
The handshaking occurs prior to I/O continuation between the target and initiator following a state change. Thus, every time a link goes down and then comes back up later this form of authentication occurs. Actually, authentication occurs continually throughout the communication of I/Os in the Fibre Channel arbitrated loop.
According to the PLDA profile, when I/Os are in progress and a target's PID changes, the target is to abort all I/Os. Then, the initiator must perform error recovery on these aborted I/O's.
Aborting I/O's in accordance with the authentication process defined by the PLDA profile results in performance problems. In the case of SCSI tape operations the authentication process may result in application failures from which the associated host may not be able to recover. That is, in the example of a SCSI tape it may be impossible for the host to identify the physical location of the tape head or the sequentially-fed tape. Other problems arise from the present method of target authentication following a state change in the Fibre Channel network because the process is in essence a catastrophic procedure for addressing a rather common Fibre Channel network occurrence.
In a multi-initiator network, the current standard for PLDA that requires target authentication following a failure such as a tape failure cannot operate. In the event that the I/O are interrupted, all initiators on the multi-initiator network must take the time and cause the necessary delays associated with the target authentication process. Accordingly, there is a need for a method and system that addresses the performance limitations arising in the known systems and methods for identifying targets and initiators following a state change in a Fibre Channel network loop.
A need exists for a method and system that overcomes the data loss problems occurring in known target authentication methods and systems during a state change in a Fibre Channel network.
Still, a need exists for a method
Arroyo Keith M.
Reynolds Robert A.
Crossroads Systems Inc.
Gray Cary Ware & Freidenrich LLP
Vaughn, Jr. William C.
Wiley David
LandOfFree
Method and system for managing I/O transmissions in a fibre... does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Method and system for managing I/O transmissions in a fibre..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Method and system for managing I/O transmissions in a fibre... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3135553