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

Patent metadata

Field	Value
Publication number	US-9614926-B2
Application number	US-201414265152-A
Country	US
Kind code	B2
Filing date	Apr 29, 2014
Priority date	Apr 29, 2014
Publication date	Apr 4, 2017
Grant date	Apr 4, 2017

Abstract

In one embodiment, a method includes using an owner gateway node to write data for a fileset, determining whether to utilize other gateway nodes to handle a portion of write traffic for the fileset, selecting a set of eligible gateway nodes based on: a current internal workload, a network workload, and recent performance history data in regard to workload distribution across the other gateway nodes, assigning and defining a size for one or more write task items for each gateway node in the set based on a current dynamic profile of each gateway node, providing availability to in-memory and/or I/O resources at each gateway node in the set to handle assigned write task items, and distributing workload to the set of eligible gateway nodes according to the size for each of the assigned write task items for each individual gateway node in the set of eligible gateway nodes.

First claim

What is claimed is:

1. A computer program product for providing parallel writing in a clustered file system having cache storage, the computer program product comprising a non-transitory computer readable storage medium having program code embodied therewith, the program code executable by a processor to cause the processor to:
write, by the processor, data for a first fileset using an owner gateway (GW) node;
determine, by the processor, whether to utilize one or more other GW nodes to handle at least a portion of write traffic for the first fileset;
select, by the processor, a set of eligible GW nodes based on at least one of: a current internal workload, a network workload, and recent performance history data in regard to workload distribution across the one or more other GW nodes;
assign and define, by the processor, a size for one or more write task items for each GW node in the set of eligible GW nodes based on a current dynamic profile of each individual GW node in the set of eligible GW nodes;
provide, by the processor, availability to in-memory and input/output (I/O) resources at the each GW node in the set of eligible GW nodes to handle one or more assigned write task items;
distribute, by the processor, workload to the set of eligible GW nodes according to the size for each of the one or more assigned write task items for the each individual GW node in the set of eligible GW nodes;
receive, by the processor, a write operation from an application node, the received write operation including a file identifier, a start offset value, and an end offset value related to the one or more write task items;
search, by the processor, a write queue to determine whether a write message for the file identifier is already in the write queue;
merge, by the processor, any already-queued write messages with the received write operation in response to a determination that the file identifier is already in the write queue and write chunks are overlapping for the already-queued write messages and the received write operation; and
prepare, by the processor, a scatter-gather list including write chunk information for the received write operation in response to a determination that the file identifier is not already in the write queue.

2. The computer program product as recited in claim 1 , wherein the program code configured to determine whether to utilize the one or more other GW nodes to handle the portion of the write traffic for the first fileset is performed as one of: a background process periodically, and a foreground process before initiating a large I/O request according to a predetermined I/O size threshold.

3. The computer program product as recited in claim 1 , wherein the program code is further configured to determine, by the processor, an amount of workload corresponding to the first fileset at the owner GW node based on one or more of: an amount of data present in I/O requests that are queued for the first fileset, an amount of data in outstanding I/O requests for the first fileset that have been issued and but not yet completed, an I/O workload at one or more other GW nodes due to traffic from one or more other filesets, and an I/O workload at the one or more other GW nodes that are functioning as an application node serving local applications.

4. The computer program product as recited in claim 3 , wherein the program code is further configured to determine, by the processor, the amount of workload corresponding to the first fileset for other GW nodes which have connectivity with the owner GW node.

5. The computer program product as recited in claim 1 , wherein the program code is further configured to:
receive, by the processor, indication of the one or more write task items to be split;
determine, by the processor, the one or more GW nodes to split each of the one or more write task items across individually;
remove, by the processor, any GW node from the set of eligible GW nodes in response to a determination that an error occurs at a GW node during write task processing; and
reassign, by the processor, any write task items that were assigned to a GW node that was removed from the set of eligible GW nodes to another GW node from the set of eligible GW nodes.

6. The computer program product as recited in claim 1 , wherein the write queue is stored as a hash table.

7. The computer program product as recited in claim 1 , wherein the program code is further configured to:
coalesce and sort, by the processor, a write chunk list for a write operation related to the one or more write task items;
determine, by the processor, a size of each write chunk in the write chunk list;
aggregate, by the processor, all write chunks from the write chunk list which have a size that is less than a predetermined minimum chunk size into a small chunk list;
create, by the processor, a write work item including all write chunks in the small chunk list in response to a determination that a sum of all write chunks in the small chunk list exceeds the predetermined minimum chunk size and a determination that no more write chunks are left to be processed in the write chunk list;
create, by the processor, a write work item for each write chunk from the write chunk list which has a size that is greater than the predetermined minimum chunk size and less than a predetermined maximum chunk size, the write work item having a size related to a size of the write chunk from the write chunk list having the size that is greater than the predetermined minimum chunk size and less than the predetermined maximum chunk size; and
create, by the processor, more than one write work item for each write chunk from the write chunk list which has a size that is greater than the predetermined maximum chunk size, a number of write work items created being related to how many times greater the size of the write chunk is than the predetermined maximum chunk size.

8. The computer program product as recited in claim 1 , wherein the program code is further configured to:
receive, by the processor, a task response from a slave GW node, the task response indicating at least an update time for a remote file, and attributes at a cache side GW node; and
store, by the processor, the update time and the attributes for the remote file in response to a determination that: the task response is a first task response or the update time for the remote file is more recent than a stored update time.

9. A method for providing parallel writing in a clustered file system having cache storage, the method comprising:
using an owner gateway (GW) node to write data for a first fileset;
determining whether to utilize one or more other GW nodes to handle at least a portion of write traffic for the first fileset;
selecting a set of eligible GW nodes based on at least one of: a current internal workload, a network workload, and recent performance history data in regard to workload distribution across the one or more other GW nodes;
assigning and defining a size for one or more write task items for each GW node in the set of eligible GW nodes based on a current dynamic profile of each individual GW node in the set of eligible GW nodes;
providing availability to in-memory and input/output (I/O) resources at the each GW node in the set of eligible GW nodes to handle one or more assigned write task items;
distributing workload to the set of eligible GW nodes according to the size for each of the one or more assigned write task items for the each individual GW node in the set of eligible GW nodes,
receiving a write operation from an application node, the received write operation including a file identifier, a start offset value, and an end offset value related to the one or more write task it

Assignees

• IBM

Inventors

• Gunda Kalyan C
• Hildebrand Dean
• Naik Manoj P
• Shiraguppi Riyazahamad M

Classifications

• H04L67/101

  based on network conditions · CPC title

• H04L67/2842 Primary

  Electricity · mapped topic

• H04L67/1008 Primary

  based on parameters of servers, e.g. available memory or workload (monitoring of computer activity G06F11/30) · CPC title

• G06F17/30203

  Physics · mapped topic

• H04L67/568

  Storing data temporarily at an intermediate stage, e.g. caching · CPC title