Parallel I/O read processing for use in clustered file systems having cache storage

• US 10,657,108 B2
• Filed: 11/29/2017
• Issued: 05/19/2020
• Est. Priority Date: 04/29/2014
• Status: Active Grant

• Current Assignee: International Business Machines Corporation
• Original Assignee: International Business Machines Corporation
• Inventors: Gunda, Kalyan C., Hildebrand, Dean, Naik, Manoj P., Shiraguppi, Riyazahamad M.
• Primary Examiner(s): Nguyen, Thu N
• Application Number:
  US15/826,492
• Publication Number:
  US 20180129680A1
• Time in Patent Office: 902 Days
• Field of Search: None
• CPC Class Codes: G06F 16/172 Caching, prefetching or hoarding of files
• CPC Class Codes: G06F 16/183 Provision of network file services by network file servers, e.g. by using NFS, CIFS network file access protocols H04L67/1097
• CPC Class Codes: H04L 1/1614 using bitmaps
• CPC Class Codes: H04L 67/1008 based on parameters of servers, e.g. available memory or workload monitoring of computer activity G06F11/30
• CPC Class Codes: H04L 67/101 based on network conditions
• CPC Class Codes: H04L 67/568 Storing data temporarily at an intermediate stage, e.g. caching
• CPC Class Codes: H04L 67/5681 Pre-fetching or pre-delivering data based on network characteristics

Abstract

In one embodiment, a method includes determining a home node that corresponds to gateway (GW) nodes in a clustered file system, each GW node being eligible to process one or more read tasks, determining a peer GW eligibility value for more than one of the GW nodes in the clustered file system eligible to process one or more read tasks, and determining a single GW node from amongst the GW nodes having a highest peer GW eligibility value for each home node. Additionally, the method includes assigning and defining a size for one or more read task items for the GW nodes having the highest peer GW eligibility value for multiple home nodes based on a current dynamic profile of the GW nodes, and distributing workload to the GW nodes according to the size for each of the one or more read task items assigned to the GW nodes.

Claims

1. A computer program product, comprising a computer readable storage medium having program code embodied therewith, the embodied program code executable by a processor to cause the processor to:

determine, by the processor, a home node that corresponds to gateway (GW) nodes in a clustered file system, each GW node being eligible to process one or more read tasks;

determine, by the processor, a peer GW eligibility value for more than one of the GW nodes in the clustered file system eligible to process the one or more read tasks;

determine, by the processor, a single GW node from amongst the GW nodes having a highest peer GW eligibility value for each home node;

assign and define, by the processor, a size for one or more read task items for the GW nodes having the highest peer GW eligibility value for multiple home nodes based on a current dynamic profile of the GW nodes;

distribute, by the processor, workload to the GW nodes according to the size for each of the one or more read task items assigned to the GW nodes;

receive, by the processor, a remote block read request for an uncached file from an application node;

determine, by the processor, whether the remote block read request is sent in response to a foreground read request or initiation of a manual file prefetch;

determine, by the processor, whether a size of the requested file exceeds a parallel read threshold in response to a determination that the remote block read request is sent in response to the manual file prefetch and invoke a parallel input/output (I/O) code path with handling of sparse blocks and small read scenarios in response to a determination that the size of the requested file exceeds the parallel read threshold; and

determine, by the processor, whether at least two blocks of the requested file have been cached in response to a determination that the remote block read request is sent in response to the foreground read request and invoke the parallel I/O code path with handling of sparse blocks and small read scenarios in response to a determination that at least two blocks of the requested file have been cached and a percentage of cached blocks of the requested file versus all blocks of the requested file exceeds a predetermined prefetch threshold.

2. The computer program product as recited in claim 1, wherein the embodied program code is further executable by the processor to cause the processor to determine, by the processor, an amount of workload corresponding to the one or more read task items based on information selected from the group consisting of:

an amount of data present in input/output (I/O) requests that are queued for the one or more read task items, an amount of data in outstanding I/O requests for the one or more read task items that have been issued, an amount of data in outstanding I/O requests for the one or more read task items that have been received but not yet completed, an amount of data in outstanding I/O requests for the one or more read task items that have been issued and received but not yet completed, an I/O workload at each individual GW node in the clustered file system eligible to process the one or more read tasks due to traffic from one or more other read task items, and an I/O workload at each individual GW node in the clustered file system eligible to process the one or more read tasks that are functioning as an application node serving local applications.

3. The computer program product as recited in claim 2, wherein the embodied program code is further executable by the processor to cause the processor to determine, by the processor, the amount of workload corresponding to the one or more read task items for other GW nodes which have connectivity with any of the individual GW nodes eligible to process the one or more read tasks in the clustered file system.

4. The computer program product as recited in claim 1, wherein the embodied program code is further executable by the processor to cause the processor to:

receive, by the processor, indication of the one or more read task items to be split; and

reassign, by the processor, any read task items that were assigned to a first GW node to another GW node eligible to process the one or more read tasks in the clustered file system in response to a determination that an error has occurred at the first GW node during read task processing.

5. The computer program product as recited in claim 1, wherein the embodied program code is further executable by the processor to cause the processor to:

receive, by the processor, a task response from a slave GW node corresponding to the one or more read task items, the task response indicating at least a read start value and an end offset value of data corresponding to the one or more read task items;

retrieve, by the processor, a cached block bitmap of the data corresponding to the one or more read task items, the cached block bitmap being an indication of which portions of the data corresponding to the one or more read task items are stored to a cache storage;

determine, by the processor, whether, for each read request waiting for remote read corresponding to the one or more read task items, all blocks of the data corresponding to the one or more read task items are stored in the cache storage;

send, by the processor, a response to the slave GW node indicating that caching has been performed for the data corresponding to the one or more read task items in response to a determination that all blocks of the data corresponding to the one or more read task items are stored in the cache storage; and

send, by the processor, a request to cache portions of the data corresponding to the one or more read task items which are not currently in the cache storage in response to a determination that all blocks of the data corresponding to the one or more read task items are not stored in the cache storage.

6. A computer-implemented method, comprising:

determining a home node that corresponds to gateway (GW) nodes in a clustered file system, each GW node being eligible to process one or more read tasks;

determining a peer GW eligibility value for more than one of the GW nodes in the clustered file system eligible to process one or more read tasks;

determining a single GW node from amongst the GW nodes having a highest peer GW eligibility value for each home node;

assigning and defining a size for one or more read task items for the GW nodes having the highest peer GW eligibility value for multiple home nodes based on a current dynamic profile of the GW nodes;

distributing workload to the GW nodes according to the size for each of the one or more read task items assigned to the GW nodes;

receiving a task response from a slave GW node corresponding to the one or more read task items, the task response indicating at least a read start value and an end offset value of data corresponding to the one or more read task items;

retrieving a cached block bitmap of the data corresponding to the one or more read task items, the cached block bitmap being an indication of which portions of the data corresponding to the one or more read task items are stored to a cache storage;

determining whether, for each read request waiting for remote read corresponding to the one or more read task items, all blocks of the data corresponding to the one or more read task items are stored in the cache storage;

sending a response to the slave GW node indicating that caching has been performed for the data corresponding to the one or more read task items in response to a determination that all blocks of the data corresponding to the one or more read task items are stored in the cache storage; and

sending a request to cache portions of the data corresponding to the one or more read task items which are not currently in the cache storage in response to a determination that all blocks of the data corresponding to the one or more read task items are not stored in the cache storage.

7. The method as recited in claim 6, further comprising determining an amount of workload corresponding to the one or more read task items based on information selected from the group consisting of:

an amount of data present in input/output (I/O) requests that are queued for the one or more read task items, an amount of data in outstanding I/O requests for the one or more read task items that have been issued, an amount of data in outstanding I/O requests for the one or more read task items that have been received but not yet completed, an amount of data in outstanding I/O requests for the one or more read task items that have been issued and received but not yet completed, an I/O workload at each individual GW node in the clustered file system eligible to process the one or more read tasks due to traffic from one or more other read task items, and an I/O workload at each individual GW node in the clustered file system eligible to process the one or more read tasks that are functioning as an application node serving local applications.

8. The method as recited in claim 7, further comprising determining the amount of workload corresponding to the one or more read task items for other GW nodes which have connectivity with any of the individual GW nodes eligible to process the one or more read tasks in the clustered file system.

9. The method as recited in claim 6, further comprising:

receiving indication of the one or more read task items to be split; and

reassigning any read task items that were assigned to a first GW node to another GW node eligible to process the one or more read tasks in the clustered file system in response to a determination that an error has occurred at the first GW node during read task processing.

10. The method as recited in claim 6, further comprising:

receiving a remote block read request for an uncached file from an application node;

determining whether the remote block read request is sent in response to a foreground read request or initiation of a manual file prefetch;

determining whether a size of the requested file exceeds a parallel read threshold in response to a determination that the remote block read request is sent in response to the manual file prefetch and invoke a parallel I/O code path with handling of sparse blocks and small read scenarios in response to a determination that the size of the requested file exceeds the parallel read threshold; and

determining whether at least two blocks of the requested file have been cached in response to a determination that the remote block read request is sent in response to the foreground read request and invoke the parallel I/O code path with handling of sparse blocks and small read scenarios in response to a determination that at least two blocks of the requested file have been cached and a percentage of cached blocks of the requested file versus all blocks of the requested file exceeds a predetermined prefetch threshold.

11. A system, comprising a processor and logic integrated with the processor, executable by the processor, or integrated with and executable by the processor, the logic being configured to cause the processor to:

determine a home node that corresponds to gateway (GW) nodes in a clustered file system, each GW node being eligible to process one or more read tasks;

determine a peer GW eligibility value for more than one of the GW nodes in the clustered file system eligible to process the one or more read tasks;

determine a single GW node from amongst the GW nodes having a highest peer GW eligibility value for each home node;

assign and define a size for one or more read task items for the GW nodes having the highest peer GW eligibility value for multiple home nodes based on a current dynamic profile of the GW nodes;

distribute workload to the GW nodes according to the size for each of the one or more read task items assigned to the GW nodes; and

receive a remote block read request for an uncached file from an application node;

determine whether the remote block read request is sent in response to a foreground read request or initiation of a manual file prefetch;

determine whether a size of the requested file exceeds a parallel read threshold in response to a determination that the remote block read request is sent in response to the manual file prefetch and invoke a parallel input/output (I/O) code path with handling of sparse blocks and small read scenarios in response to a determination that the size of the requested file exceeds the parallel read threshold; and

determine whether at least two blocks of the requested file have been cached in response to a determination that the remote block read request is sent in response to the foreground read request and invoke the parallel I/O code path with handling of sparse blocks and small read scenarios in response to a determination that at least two blocks of the requested file have been cached and a percentage of cached blocks of the requested file versus all blocks of the requested file exceeds a predetermined prefetch threshold.

12. The system as recited in claim 11, wherein the logic is further configured to cause the processor to determine an amount of workload corresponding to the one or more read task items based on information selected from a group consisting of:

an amount of data present in input/output (I/O) requests that are queued for the one or more read task items, an amount of data in outstanding I/O requests for the one or more read task items that have been issued, an amount of data in outstanding I/O requests for the one or more read task items that have been received but not yet completed, an amount of data in outstanding I/O requests for the one or more read task items that have been issued and received but not yet completed, an I/O workload at each individual GW node in the clustered file system eligible to process the one or more read tasks due to traffic from one or more other read task items, and an I/O workload at each individual GW node in the clustered file system eligible to process the one or more read tasks that are functioning as an application node serving local applications.

13. The system as recited in claim 12, wherein the logic is further configured to cause the processor to determine the amount of workload corresponding to the one or more read task items for other GW nodes which have connectivity with any of the individual GW nodes eligible to process the one or more read tasks in the clustered file system.

14. The system as recited in claim 11, wherein the logic is further configured to cause the processor to:

receive indication of the one or more read task items to be split; and

reassign any read task items that were assigned to a first GW node to another GW node eligible to process the one or more read tasks in the clustered file system in response to a determination that an error has occurred at the first GW node during read task processing.

15. The system as recited in claim 11, wherein the logic is further configured to cause the processor to:

receive a task response from a slave GW node corresponding to the one or more read task items, the task response indicating at least a read start value and an end offset value of data corresponding to the one or more read task items;

retrieve a cached block bitmap of the data corresponding to the one or more read task items, the cached block bitmap being an indication of which portions of the data corresponding to the one or more read task items are stored to a cache storage;

determine whether, for each read request waiting for remote read corresponding to the one or more read task items, all blocks of the data corresponding to the one or more read task items are stored in the cache storage;

send a response to the slave GW node indicating that caching has been performed for the data corresponding to the one or more read task items in response to a determination that all blocks of the data corresponding to the one or more read task items are stored in the cache storage; and

send a request to cache portions of the data corresponding to the one or more read task items which are not currently in the cache storage in response to a determination that all blocks of the data corresponding to the one or more read task items are not stored in the cache storage.