ADAPTIVE POWER CONSERVATION IN STORAGE CLUSTERS

• US 20150127967A1
• Filed: 01/16/2015
• Published: 05/07/2015
• Est. Priority Date: 07/20/2009
• Status: Active Grant

• Current Assignee: DataCore Software Corporation (Flagship Ventures 2004 Fund LLC)
• Original Assignee: Caringo Incorporated
• Inventors: DUTTON, James E., YOAKLEY, David, ARBILLA, Laura
• Granted Patent: US 9,348,408 B2
• Time in Patent Office: Days
• US Class Current: 713/323
• CPC Class Codes: G06F 1/08 Clock generators with changeable or programmable clock frequency
• CPC Class Codes: G06F 1/32 Means for saving power
• CPC Class Codes: G06F 1/3221 of disk drive devices
• CPC Class Codes: G06F 1/3268 Power saving in hard disk drive
• CPC Class Codes: G06F 1/3287 by switching off individual functional units in the computer system
• CPC Class Codes: G06F 1/3296 by lowering the supply or operating voltage
• CPC Class Codes: G06F 3/0625 Power saving in storage systems
• CPC Class Codes: G06F 3/0634 by changing the state or mode of one or more devices
• CPC Class Codes: G06F 3/067 Distributed or networked storage systems, e.g. storage area networks [SAN], network attached storage [NAS]
• CPC Class Codes: Y02D 10/00 Energy efficient computing, e.g. low power processors, power management or thermal management

Abstract

Each node and volume in a storage cluster makes a decision whether to reduce power consumption based on lack of requests from client applications and nodes over a time period. Node configuration parameters determine how long to wait until idling a node or volume, and how long to wait while idle before performing integrity checks. A bid value is calculated by each node and reflects how much it will cost for that node to write a file, read a file, or keep a copy. A node with the lowest bid wins, and nodes that are idle have a premium added to each bid to ensure that idle nodes are kept idle. In an archive mode, writes bids are reversed, nodes with less capacity submit lower bids, fuller nodes fill up faster and are then idled, while empty or near empty nodes may remain idle before winning a write bid.

Claims

1. A method of reading a computer file from a storage cluster having a plurality of computer nodes, said method comprising:

receiving at a first computer node of said storage cluster, a request from an external client application to retrieve said computer file from said storage cluster, each of said computer nodes of said storage cluster being in an idle state;

receiving, at said first computer node, bid values from a subset of said computer nodes for reading said computer file, each computer node in said subset including said computer file;

choosing one of said plurality of computer nodes based upon said bid values from said computer nodes, said computer file residing on said chosen computer node while said chosen computer node is in an idle state, calculation of said bid values not requiring a disk access of said computer nodes;

transitioning said chosen computer node and a disk volume of said chosen computer node that includes said computer file to an active state;

maintaining computer nodes of said storage cluster that are not chosen in said idle state; and

returning said computer file from said disk volume of said chosen computer node to said external client application.

2. A method as recited in claim 1 wherein said storage cluster is a write-once read-many (WORM) fixed-content stored cluster.

3. A method as recited in claim 1 wherein each of said idled computer nodes has a decreased clock rate of a CPU.

4. A method as recited in claim 1 wherein each disk volume of said idled computer nodes is in the process of being spun down.

5. A method of writing a computer file to a storage cluster having a plurality of computer nodes, said method comprising:

receiving at a first computer node of said storage cluster, a request from an external client application to write said computer file to said storage cluster, each of said computer nodes of said storage cluster being in an idle state, at least one of said computer nodes including at least one user computer file;

each of said computer nodes calculating a bid value reflecting how much it costs each node to write to said each computer node, calculation of said bid value not requiring a disk access of said each computer node;

receiving, at said first computer node, said bid values from said computer nodes;

choosing one of said plurality of computer nodes by choosing the computer node with the best bid value;

transitioning said chosen computer node and a disk volume of said chosen computer node to an active state;

maintaining computer nodes of said storage cluster that are not chosen in said idle state; and

writing said computer file from said external client application to said disk volume of said chosen computer node.

6. A method as recited in claim 5 wherein said storage cluster is a write-once read-many (WORM) fixed-content stored cluster.

7. A method as recited in claim 5 wherein each of said idled computer nodes has a decreased clock rate of a CPU.

8. A method as recited in claim 5 wherein each disk volume of said idled computer nodes is in the process of being spun down.

9. A method of writing a computer file to a storage cluster having a plurality of computer nodes, said method comprising:

receiving at a first computer node of said storage cluster, a request from an external archive application to write said computer file to said storage cluster;

each of said computer nodes calculating a bid value for the opportunity to write to said each of said computer nodes, wherein a subset of said computer nodes are designated as being in archive mode and wherein bid values from computer nodes in archive mode having greater capacity being higher than bid values from computer nodes in archive mode having less capacity;

receiving, at said first computer node, said bid values from said computer nodes;

choosing, by said first computer node, one of said plurality of computer nodes having the least bid value because said chosen node has the least capacity out of said computer nodes; and

writing said computer file from said external client application to a disk volume of said chosen computer node.

10. A method as recited in claim 9 wherein said bid value is calculated by taking into account a capacity of said each computer node, whether said each computer node is idle, how busy said each computer node has been, and a size of said computer file.

11. A method as recited in claim 9 wherein said storage cluster is a write-once read-many (WORM) fixed-content stored cluster.

12. A method as recited in claim 9 further comprising:

writing said computer file to a disk volume that has the least capacity of said chosen computer node.

13. A method as recited in claim 9 further comprising:

filling disk volumes of said second computer node to near capacity; and

idling said second computer node when it is determined that a predetermined amount of time has passed and no requests from said external archive application have been processed.

14. A method of storing a computer file in a storage cluster having a plurality of computer nodes, said method comprising:

receiving at a first computer node of said storage cluster a request from an external archive application to write said computer file to said storage cluster;

calculating by a second computer node in said storage cluster a bid value for the opportunity to write to said second computer node;

determining whether said second computer node is in an archive mode, and when it is determined that said second computer node is in an archive mode, adjusting said bid value such that the greater capacity of said second computer node the higher said bid value;

said second computer node winning a bid competition amongst said computer nodes on the basis of said calculated bid value being a least bid value; and

writing said computer file from said external client application to a disk volume of said second computer node.

15. A method as recited in claim 14 wherein said bid value is calculated by taking into account a capacity of said second computer node, whether said second computer node is idle, how busy said each computer node has been, and a size of said computer file.

16. A method as recited in claim 14 wherein said storage cluster is a write-once read-many (WORM) fixed-content stored cluster.

17. A method as recited in claim 14 further comprising:

writing said computer file to said disk volume that has the least capacity of said second computer node.

18. A method as recited in claim 14 further comprising:

filling disk volumes of said second computer node to near capacity; and

idling said second computer node when it is determined that a predetermined amount of time has passed and no requests from said external archive application have been processed.

19. A method as recited in claim 14 further comprising:

submitting said bid value to said first computer node, said first computer node receiving other bid values from other of said computer nodes for the opportunity to write to said other computer nodes.

20. A method as recited in claim 14 wherein said second computer node wins said bid competition because said second computer node has the least capacity out of said computer nodes.

21. A method of processing an external request in a storage cluster having a plurality of computer nodes, said method comprising:

receiving, at a first computer node of said storage cluster, an external request originating at an external client application;

calculating, by an idle computer node of said storage cluster, a bid value for the opportunity to process said external request;

determining whether said idle computer node is in an idle state, wherein when it is determined that said idle computer node is in an idle state, adding a premium value to said bid value resulting in a final bid value;

submitting said final bid value to said first computer node for processing said external request;

receiving notification by said idle computer node that said final bid value is not accepted to process said external request by virtue of no response being received to said submitted final bid value at said idle computer node, said final bid value not being accepted because said final bid value is higher than another bid value from another computer node of said storage cluster; and

not processing said external request by said idle computer node.

22. A method as recited in claim 21 wherein said storage cluster is a write-once read-many (WORM) fixed-content stored cluster.

23. A method as recited in claim 21 further comprising:

maintaining said idle computer node in an idle state, wherein said idle state is characterized in a decreased clock rate of a CPU of said idle computer node.

24. A method of writing a computer file to an idled storage cluster having a plurality of computer nodes, said storage cluster including a plurality of drives, including at least one solid-state drive (SSD) and at least one hard disk drive (HDD), said method comprising:

receiving at a first computer node of said storage cluster, a request from an external application to write said computer file to said storage cluster, all of said plurality of drives of said idled storage cluster being in an idle state;

each of said computer nodes calculating a bid value for the opportunity to write said computer file, wherein a bid value from a second computer node takes into account an SSD bid value from an SSD of said second computer node, said SSD bid value being calculated differently from an HDD bid value from an HDD of one of said computer nodes;

receiving, at said first computer node, said bid values from said computer nodes;

choosing said second computer node having the lowest bid value; and

writing said computer file from said external client application to said SSD of said chosen second computer node.

25. A method as recited in claim 24 wherein said storage cluster is a write-once read-many (WORM) fixed-content stored cluster.

26. A method as recited in claim 24 further comprising:

adding a premium value to bid values of said HDDs of said storage cluster because said HDDs are in an idle state; and

not adding said premium value to said SSD bid value of said SDD of said second computer node even though said SDD is in an idle state.