Error detection/correction and fault detection/recovery – Data processing system error or fault handling – Reliability and availability
Reexamination Certificate
2000-11-06
2004-04-20
Beausoliel, Robert (Department: 2184)
Error detection/correction and fault detection/recovery
Data processing system error or fault handling
Reliability and availability
C709S239000
Reexamination Certificate
active
06725393
ABSTRACT:
FIELD OF THE INVENTION
The invention relates to the field of high-reliability computing on storage area networks. In particular the inventions relates to systems and methods of maintaining mirrored datasets when a storage area network suffers sufficient disruption that a particular copy of a mirrored dataset can not be seen for direct writes by one, but not all, compute nodes of the network.
BACKGROUND OF THE INVENTION
In the field of high-reliability computing, it is often desirable to maintain redundant data. Redundant data can provide some protection against failure of a storage device. For example, a RAID (Redundant Array of Independent Disks) system can often be configured to keep full duplicate copies of data on separate disk drives. Should a failure occur that affects one, but not both, of these duplicate, or “mirrored” datasets, data will not be lost. Continued operation may also be possible using the surviving dataset. Other configurations are known, for example, in RAID-5 operation data and parity-recovery information may be striped across a number of drives, failure of any one drive will not result in data loss.
Mirrored datasets are not limited to duplicate datasets maintained by RAID systems. For example, it may be desirable to maintain a primary copy of a mirrored dataset at a different geographical location than the secondary copy. Such remotely located mirrored datasets can provide protection against data loss in the event of flood, fire, lightning strike, or other disaster involving the location of one copy of the dataset.
A mirrored dataset ideally has at least two copies of all information written to the dataset. Whenever a write occurs, that write must be made to both copies for full redundancy to be maintained. If only one copy is written then redundancy protection is lost until repairs can be made and the datasets synchronized. Synchronization of datasets can be a time consuming task; it is desirable that need for synchronization be minimized. On the other hand, reading of data from a mirrored dataset can occur from any copy if the dataset is synchronized, or if the data read is known not to have been altered since the last synchronization of the data.
Storage Area Networks (SANs) are characterized as high-speed networks primarily conveying data between storage nodes and compute nodes, often utilizing separate network hardware from that used for general-purpose network functions. Storage nodes are machines that primarily serve storage to other nodes of the network, while compute nodes are typically computers that use storage provided by storage nodes. Compute nodes may, and often do, have additional storage devices directly attached to them.
SANs are often implemented with fibre-channel hardware, which may be of the arbitrated loop or switched-fabric type. Storage area networks may be operated in a “clustering” environment, where multiple compute nodes have access to at least some common data, the common data may in turn be stored with redundancy. SANs having multiple processors accessing a common database stored with redundancy, are often used for transaction processing systems.
SANs are also known that use non-fibre-channel interconnect hardware.
Most modern computer networks, including fibre-channel storage area networks, are packet oriented. In these networks, data transmitted between machines is divided into chunks of size no greater than a predetermined maximum. Each chunk is packaged with a header and a trailer into a packet for transmission. In Fibre-Channel networks, packets are known as Frames.
A network interface for connection of a machine, to a Fibre Channel fabric is known as an N_port, and a machine attached to a Fibre-Channel network is known as a node. Nodes may be computers, or may be storage devices such as RAID systems. An NL_port is an N-port that supports additional arbitration required so that it may be connected either to a Fibre Channel fabric or to a Fibre Channel Arbitrated Loop. An L_port is a network interface for connection of a machine to a Fibre Channel Arbitrated Loop. Typically, an N_port, NL_port, or L_Port originates or receives data frames. Each port incorporates such hardware and firmware as is required to transmit and receive frames on the network coupled to a processor and at least one memory system. Ports may incorporate a processor and memory of their own, those that don't utilize memory and processor of their node. Received frames are stored into memory, and transmitted frames are read from memory. Such ports generally do not re-address, switch, or reroute frames.
SANS often have redundant network interconnect. This may be provided to increase performance by providing high bandwidth between the multiple nodes of the network; to provide for operation despite some potential failures of network components such as hubs, switches, links, or ports; or both.
Description of the Problem
It is possible for some network interconnect components of a SAN to fail while other components continue to operate. This can disrupt some paths between nodes of the network.
There are possible network configurations where a first compute node of the SAN can lose its direct path to a first storage node; while the first compute node has a path to a second storage node of the network, and a second compute node still has a path to the first storage node. If data is mirrored on the primary and secondary storage nodes, the first processor has difficulty updating the copy on the primary storage node, although it can read data from the copy on the secondary node and update that copy.
When failures of this type occur, typical SAN-based systems are left with two alternatives: First, the first processor may be shut down, forcing the second processor to handle all load, but permitting maintenance of the mirrored data. This is undesirable because there may be significant loss of throughput with the first processor off-line. Second, the first storage node may be shut down, permitting the processors to share the load, but causing the mirrored datasets to lose synchronization. This is undesirable because synchronization of the datasets is required before the first storage node can be brought back on-line, and because there is a risk of data loss should the second storage node fail before synchronization is completed.
Solution to the Problem
A modified NL_Port (M_Port) has capability to automatically maintain a mirrored dataset on a pair of storage nodes. A second M_Port can perform a write operation to a copy of a mirrored dataset on behalf of a first M_Port should the second M_Port be able to communicate with the first M_port, and the first M_Port be unable to reach that copy of the mirrored dataset. This write by the second M_Port in behalf of the first M_Port is known herein as a surrogate write.
In a first embodiment, surrogate writes are performed by port hardware, without need to involve a node processor in the surrogate write.
In another embodiment of the invention, surrogate writes are performed by a SAN node, thereby enabling surrogate writes when surrogate write requests are received on a port other than that in communication with the target of the surrogate writes.
The invention is applicable to Storage Area Networks (SANs) in general, and is of particular utility for Web-page serving and transaction processing systems.
REFERENCES:
patent: 5155845 (1992-10-01), Beal et al.
patent: 5216619 (1993-06-01), Dunbar et al.
patent: 5544347 (1996-08-01), Yanai et al.
patent: 5633999 (1997-05-01), Clowes et al.
patent: 5668943 (1997-09-01), Attanasio et al.
patent: 6073209 (2000-06-01), Bergsten
patent: 6101508 (2000-08-01), Wolff
patent: 6128750 (2000-10-01), Espy et al.
patent: 6389432 (2002-05-01), Pothapragada et al.
patent: 6397356 (2002-05-01), Yonezawa
patent: 6400730 (2002-06-01), Latif et al.
patent: 6421711 (2002-07-01), Blumenau et al.
patent: 6535518 (2003-03-01), Hu et al.
patent: 6564336 (2003-05-01), Majkowski
patent: 6643795 (2003-11-01), Sicola et al.
patent: 6654769 (2003-11-01), Ito et al.
patent: 6658478 (2003-12-01), Singhal
Grieff Thomas
Pellegrino Greg
Beausoliel Robert
Chu Gabriel
LandOfFree
System, machine, and method for maintenance of mirrored... does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with System, machine, and method for maintenance of mirrored..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and System, machine, and method for maintenance of mirrored... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3224381