Error detection/correction and fault detection/recovery – Data processing system error or fault handling – Reliability and availability
Reexamination Certificate
1999-12-20
2003-12-02
Beausoliel, Robert (Department: 2184)
Error detection/correction and fault detection/recovery
Data processing system error or fault handling
Reliability and availability
C707S793000
Reexamination Certificate
active
06658589
ABSTRACT:
FIELD OF THE INVENTION
This invention is directed towards data storage systems, and more particularly towards physical backup and restore of databases serving multi-processor computers.
BACKGROUND
Computer systems allow the processing of massive quantities of data for a variety of purposes. As the ability to process data has increased, so has the need for data storage systems which provide massive data storage capabilities combined with fast access for host systems. Another feature required by many businesses and industries is continuous availability. Many businesses operate on a world-wide basis, and have a need for round-the-clock access to databases stored in one or more data storage systems. The data stored in these data storage systems is changing at an incredible rate, for example with transaction processing, reservation systems and data mining, the data is changing and updating many times per second.
Another requirement for data storage systems is periodic backup of data both for archival purposes and for data recovery in case of a system failure. For many businesses, a loss of data can be catastrophic. Therefore, system backups must be performed on a frequent basis.
However, the need for system backups often interferes with the need for continuous availability. With many data storage systems, performing a system backup requires taking the data storage system offline, thereby denying continuous access to the data.
One solution to this problem is used for RAID (Redundant Array of Independent Disks) systems. In RAID-1 systems, two physical storage devices, such as disks, each store identical data, in a process known as “mirroring”. This provides a very high level of fault tolerance in the form of redundancy, and it also allows data backups to be performed while still allowing continuous data access. Typically, the mirroring process is stopped (referred to as splitting the mirrors), and one of the disks is taken off-line and backed up, while the other disk remains online and available. When the first disk is completely backed up, the two disks are resynchronized (so that the data is identical on both), and the data storage system returns to full operation.
An overview of major components of a backup data system
10
is shown in FIG.
1
. One or more host computer systems
12
access, process and store data from a data storage system
14
. The host systems
12
, including Massively Parallel Processor (MPP) or Symmetric Multi-Processor (SMP) systems are interfaced to the data storage system
14
over an interface
16
, which may be any of various types of interface such as a network or SCSI interface. The host systems
12
are also interfaced
20
to a backup system
22
, which provides data backup and restore to appropriate storage devices
24
, for example via tape storage. This interface
20
between the host systems
12
and the backup system
22
is also any of various types of interface, such as a TCP/IP connection.
The data storage system
14
is any of various types of mass data storage systems, including for example a RAID system with multiple disks. A RAID-1 system is illustrated with two mirrored disk volumes (mirrors)
18
a
,
18
b
. The mirrors
18
a
,
18
b
are connected
21
such that the data is replicated on both mirrors
18
a
,
18
b
. Although the mirrors
18
a
,
18
b
are illustrated in a same data storage system
14
enclosure, the mirrors can be physically remote from each other, but still support RAID-1 mirroring using a remote data facility option, including a high-speed connection
21
such as an ESCON® fibre link connection.
For backup and restore of data stored on the data storage system
14
, a standard method for backup requires the host systems
12
to extract the data from the databases on the data storage system
14
and pipe the data over to the backup management system
22
. This method is incredibly slow, and it requires tying up the host system's
12
time in the form of database access operations and data pipelining. A better solution is known as “direct connect”. A high speed direct connection
26
is provided between the data storage system
14
and the backup management system
22
, thereby allowing fast transfer of data directly to the backup management system
22
, without the need for host system
12
intervention. This high speed direct connection
26
can be over any of various types of interfaces, such as a SCSI connection.
An example data storage system
14
is the Symmetrix mass storage system provided by EMC Corporation of Hopkinton, Mass. An example backup management system
22
is the EMC Data Manager (EDM). EDM can support backup and restore via three different methods, each tailored to particular backup environments and needs. The same EDM can support three different backup methods simultaneously, and enables host systems
12
and users to stay operational, with continued access to the data storage system
14
while backup occurs.
There are several types of database servers available, including parallel server databases. A parallel server database is a database server with enhancements that allow a common database to be shared among the nodes of an MPP or loosely coupled SMP system. A node can be an independent processor on an MPP or SMP machine, or a separate machine belonging to a clustered hardware environment. Parallel server databases provide processor scalability, where additional processing power may be added through the addition of more processor nodes, as well as high availability (also known as fault tolerance) in that if one processor node goes down, the other processor nodes can transparently take over the work of the down processor node.
However, there are problems related to parallel server databases. Since several nodes are accessing and writing the same database, there are problems relating to the coherency of the data. Further, backup and restore of the parallel server database becomes very complicated. Current systems are designed to handle backup and restore of a database which exists on a single database client machine, and which has only one node instance (thread)
30
FIG. 2
associated with it. A thread is one process of multiple processes running on a computer system. A thread of change log information keeps track of the database changes made by a single instance.
Typical parallel server database applications, such as an Oracle® database from Oracle Corporation of Redwood Shores, Calif., have an architecture which includes several data files to maintain database coherence and availability. Such data files include a control file
32
, which is a small administrative file required by every database, necessary to start and run a database system. A control file
32
is paired with a database, not with an instance
30
. Multiple identical control files (not shown) are preferred to a single file
32
, for reasons of data security. Other data files include a “redo” log, which is a sequential log of actions that are to be reapplied to the database if they did not get written to disk. The log usually consists of at least two files; one is optionally being spooled to disk (archived)
34
while the other is being written (online)
36
. Online redo logs
36
are redo logs that have not been archived. The online redo logs
36
may be available to the instance
30
for recording activity, or have previously been written but are awaiting archiving. Finally, there is typically a parameter file
38
which maintains instance-specific information, for example buffer sizes, archived redo log locations, and other routine information.
In this architecture (as shown in FIG.
2
), the backup system is only concerned with backup and restore of (1) data files (not shown) as seen from the database client, (2) backup copy of the control file
32
as seen from the database client, and (3) archived redo logs
34
which reside in the archived log directory as seen from the database client. The online redo logs
36
, and parameter or config files
38
are not backed up. Parallel server database providers recommend that onli
Beausoliel Robert
Brown Rudnick Berlack & Israels LLP
EMC Corporation
McCarthy Christopher S.
McHugh Steven M.
LandOfFree
System and method for backup a parallel server data storage... 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 and method for backup a parallel server data storage..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and System and method for backup a parallel server data storage... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3142394