Parallel file system and method anticipating cache usage patterns

• US 5,940,838 A
• Filed: 07/11/1997
• Issued: 08/17/1999
• Est. Priority Date: 07/11/1997
• Status: Expired due to Fees

• Current Assignee: International Business Machines Corporation
• Original Assignee: International Business Machines Corporation
• Inventors: Schmuck, Frank B., Mandler, Benjamin, Wyllie, James Christopher, Yehudai, Zvi Yosef
• Primary Examiner(s): Lintz, Paul R.
• Assistant Examiner(s): SHAH, SANJIV
• Application Number:
  US08/893,733
• Time in Patent Office: 767 Days
• Field of Search: 707/200, 707/8, 707/201
• US Class Current: 1/1
• CPC Class Codes: G06F 12/0862 with prefetch
• CPC Class Codes: G06F 12/0866 for peripheral storage systems, e.g. disk cache
• CPC Class Codes: G06F 2212/6026 Prefetching based on access pattern detection, e.g. stride based prefetch

Abstract

A computer system having a shared disk file system running on on multiple computers each having their own instance of an operating system and being coupled for parallel data sharing access to files residing on network attached shared disks. Methods are provided for use as a parallel file system in a shared disk environment by use of a scalable directory service for the system with a stable cursor, a segmented allocation map. Dynamic prefetch and cached balance pools for multiple accesses improve the system. Extended file attributes are used for implementation of Access Control Lists in a parallel file system. Improvements to caching and cache performance developments balance pools for multiple accesses. A metadata node manages file metadata for parallel read and write actions. For our system, tokens are used for metadata node selection and identification, and we have enhanced token modes for controlling file size, as well as smart caching of byte range tokens using file access patterns and a byte range lock algorithm using a byte range token interface. Locking techniques reduce the overhead of a token manager which is also used in the file system recovery if a computer participating in the management of shared disks becomes unavailable or failed.

Claims

1. A method of improving a files system'"'"'s cache utilization in a computer system having cache space requests of varying sizes presented to a file system in an unpredictable fashion, comprising the method of

making a periodic identification of workload, and

splitting said cache into different working cache units, each controlling a portion of the cache state,

periodically consulting with a cache manager for optimizing cache utilization via analyzing current cache usage patterns for anticipating the nature of incoming requests and preparing for the requests such that each incoming request is directed towards the cache unit which with a high probability will be used to satisfy the request,

determining the space that can be devoted for each workload in each working unit and adjusting other system actions in response to said determined space that can be devoted for each workload in each working unit, and

wherein when two types of workloads are detected and responded to by said cache manager, the complete cache is split into different working units, each controlling a portion of the complete cache space, and responsible for buffers of a given pre-defined size with each working unit comprised of two sub-units that monitor the two kinds of workloads the system operates with, the amount of different working units and the buffer sizes that they are responsible for change dynamically.

2. A method of improving a files system'"'"'s cache according to claim 1, wherein said other system actions includes the pre-fetching rate.

3. A method of improving a files system'"'"'s cache according to claim 1, wherein in the periodic identification of workload, two types of workloads, sequential and random workloads, are detected and responded to after the current usage pattern in the system has been established, and assumed to be relatively stable, and said cache manager for said cache responds accordingly.

4. A method of improving a files system'"'"'s cache according to claim 3, wherein there is in the system a working unit that takes care of incoming requests for buffer sizes that differ from all other working units fixed size for enhancing cache response time by pointing incoming requests directly to the cache portion which hosts buffers of the desired size.

5. A method of improving a files system'"'"'s cache according to claim 1, wherein the cache manager recognizes at each moment in time the buffer sizes for which, with a high probability, there will be a lot of demand, and sets up the working units accordingly.

6. A method of improving a files system'"'"'s cache utilization in a computer system having cache space requests of varying sizes presented to a file system in an unpredictable fashion, comprising the method of

making a periodic identification of workload usage satitistics for examination, and

splitting said cache into different working cache units, each controlling a portion of the cache state,

periodically consulting with a cache manager for optimizing cache utilization via analyzing current cache usage patterns for anticipating the nature of incoming requests and preparing for the requests such that each incoming request is directed towards the cache unit which with a high probability will be used to satisfy the request,

determining the space that can be devoted for each workload in each working unit and adjusting other system actions in response to said determined space that can be devoted for each workload in each working unit, and

wherein as periodically the gathered usage statistics are examined, said cache manager has its options for control over the cache space re-divided among the different working units and wherein future usage patterns are predicted by analyzing current ones, so new space re-division is not acted upon immediately but takes effect upon demand.

7. A method of improving a files system'"'"'s cache utilization in a computer system having cache space requests of varying sizes presented to a file system in an unpredictable fashion, comprising the method of

making a periodic identification of workload, and

splitting said cache into different working cache units, each controlling a portion of the cache state,

periodically consulting with a cache manager for optimizing cache utilization via analyzing current cache usage patterns for anticipating the nature of incoming requests and preparing for the requests such that each incoming request is directed towards the cache unit which with a high probability will be used to satisfy the request,

determining the space that can be devoted for each workload in each working unit and adjusting other system actions in response to said determined space that can be devoted for each workload in each working unit, and

wherein each working unit has two kinds of space limits, namely, an internal and an external one, and the internal space limit divides between the two sub-working units, and the external space limit of each working unit is further divided into two kinds of limits, namely the physical limit and the virtual limit, where the physical limit represents the actual amount of space under control of the usage pattern scheme distribution that belongs to the individual working unit, and the virtual limit, used to deduce whether a specific working unit physical limit is allowed to grow or whether it is forced to give up a portion of the space under its control upon a request from a working unit that is allowed to grow, thus in essence it is allowed to shrink, is projected by the usage pattern prediction process as the physical limit this working unit should attempt to achieve.

8. A method of improving a files system'"'"'s cache according to claim 7, further comprising setting new virtual units by analyzing sub-working units'"'"' statistics in order to deduce the usage pattern and activity level that determine the space optimally needed by it, and wherein when a sub-working unit attempts to obtain the amount of space it determined to be optimal working set size for its needs, the relative activity level of the sub-working unit presents a cap on the optimally needed space.

9. A method of improving a files system'"'"'s cache according to claim 8, where for new space acquisition, when a request for new buffer space arrives it is served by the working unit which controls the size requested, and if there is a free or a very easy and quick to obtain buffer in the working unit, it is used to satisfy the incoming request, and the working unit then proceeds to compare its physical limit with its virtual limit such that if the physical limit is not smaller than the virtual one, the working unit proceeds to find the easiest to obtain space already under its control, but otherwise the current working unit finds the working unit that is allowed to shrink the most and directs a space acquisition request to it, whereby the receiving working unit finds the easiest to obtain space under its control, and gives up the control over it, and then the original working unit proceeds to assume control over the new space and uses it to satisfy the incoming request.

10. A method of improving a files system'"'"'s cache according to claim 8, wherein during a periodic usage detection, the cache manger determines the time when the periodic usage detection should be next run, such that the calculation is based on the expected time for all the working units to access all the space under their control so the expected time period is subjected to pre-defined upper and lower bounds such that the interval between pre-defined upper and lower bounds permits the usage pattern process to deduce the current workload distribution without being affected by a single straining event, and a working set of random workload clients can be deduced as well as the space needed for read-ahead of sequential workload clients.