Reducing Power Consumption of Distributed Storage Systems

• US 20110213994A1
• Filed: 02/26/2010
• Published: 09/01/2011
• Est. Priority Date: 02/26/2010
• Status: Active Grant

• Current Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Original Assignee: Microsoft Corporation
• Inventors: Naraynanan, Dushyanth, Thereska, Eno, Donnelly, Austin
• Granted Patent: US 8,370,672 B2
• Time in Patent Office: Days
• US Class Current: 713/320
• CPC Class Codes: G06F 1/3221 of disk drive devices
• CPC Class Codes: G06F 1/3287 by switching off individual functional units in the computer system
• CPC Class Codes: G06F 11/2023 Failover techniques
• CPC Class Codes: G06F 11/2041 with more than one idle spare processing component
• CPC Class Codes: G06F 11/2097 maintaining the standby controller/processing unit updated initialisation or re-synchronisation thereof G06F11/1658 and subgroups
• CPC Class Codes: G06F 16/1737 for reducing power consumption or coping with limited storage space, e.g. in mobile devices saving storage space on storage devices G06F3/0608; power saving in storage systems G06F3/0625
• CPC Class Codes: G06F 16/1752 based on file chunks
• CPC Class Codes: G06F 16/1844 Management specifically adapted to replicated file systems
• CPC Class Codes: Y02D 10/00 Energy efficient computing, e.g. low power processors, power management or thermal management
• CPC Class Codes: Y02D 30/50 in wire-line communication networks, e.g. low power modes or reduced link rate

Abstract

Methods for reducing the power consumption of distributed storage systems are described. An embodiment describes a storage system which is adapted to reduce its power consumption at times of low load by reducing the number of active versions of the stored data. The data to be stored in the storage system is divided into chunks and in an example, each chunk is replicated on a number of different servers. At times of low load, the system enters a mode of operation in which the number of active replicas is reduced and servers that do not store any active replicas are put into a low power state. When in this mode, writes are written to a versioned store and the data is subsequently copied to servers storing replicas once all the servers have returned to normal power state.

Claims

1. A storage system for storing a plurality of data objects, each data object comprising at least one chunk of data and the system comprising:

a plurality of chunk servers arranged to store multiple blocks of data derived from each chunk of data;

a metadata service arranged to store mappings between data objects and chunks of data; and

a versioned store,

wherein the system is arranged to operate in one of a plurality of modes of operation, the plurality of modes of operation comprising;

a first mode in which a first number of blocks of data derived from a chunk of data are available, and

at least one reduced power mode in which a reduced number of blocks of data derived from the chunk of data are available and in which a subset of the chunk servers are in a low power state,

and wherein the versioned store is arranged to store multiple versions of any data written when the system is in a reduced power mode of operation.

2. A storage system according to claim 1, wherein each block of data derived from a chunk comprises a replica of the chunk.

3. A storage system according to claim 2, wherein one replica of the chunk is assigned as a primary replica and wherein the metadata service is further arranged to reassign the primary replica to a different replica before the system switches to a reduced power mode of operation if the primary replica is stored on a server scheduled to enter a low power state.

4. A storage system according to claim 1, wherein each chunk server is assigned to one of a plurality of gear groups and wherein each block of data derived from a chunk of data is stored on a chunk server assigned to a different gear group.

5. A storage system according to claim 4, wherein in a reduced power mode, the subset of the chunk servers in a low power state comprises all the chunk servers assigned to at least one gear group.

6. A storage system according to claim 4, divided into a plurality of fault domains and wherein each block of data derived from a chunk of data is stored on a chunk server assigned to a different gear group and located in a different fault domain.

7. A storage system according to claim 1, further comprising a scheduling service arranged to compute system load for a period of time based on at least one of historical load data and measurements of system load and to select one of the plurality of modes of operation based on the computed system load.

8. A storage system according to claim 7, wherein the scheduling service is further arranged to select the subset of the chunk servers to be in a low power state in the selected one of the plurality of modes of operation.

9. A storage system according to claim 8, wherein the scheduling service is further arranged to transmit schedules to each chunk server identifying one or more periods of time when the chunk server is to enter a low power state.

10. A storage system according to claim 1, wherein the versioned store is further arranged to transfer data from the store to one of more chunk servers when the system changes from a reduced power mode of operation to the first mode of operation.

11. A storage system according to claim 1, wherein the versioned store comprises:

a plurality of loggers arranged to store data; and

a log client associated with each chunk server and arranged to send data to multiple loggers on receipt of a write request when the system is in a reduced power mode of operation.

12. A storage system according to claim 11, wherein a logger and a chunk server are co-located on a server.

13. A storage system according to claim 1, wherein the metadata service is further arranged to detect failure of a chunk server and if a chunk server fails when the system is in a reduced power mode of operation, the metadata service is arranged to cause any chunk servers storing blocks of data derived from those chunks with blocks stored on the failed chunk server to exit low power state.

14. A method of storing data objects in a storage system, the system comprising a plurality of groups of servers and the method comprising:

dividing each data object into a plurality of chunks of data;

for each chunk of data, storing a block of data derived from the chunk of data on each of a number of servers in different groups; and

operating the system in one of a plurality of operating modes selected based on system load, the plurality of operating modes comprising;

a first mode in which a first number of blocks of data derived from a chunk of data are available; and

at least one reduced power mode in which at least one group of servers are put into a low power state such that a subset of blocks of data derived from each chunk of data are available and in which data written to the storage system is stored in a versioned store.

15. A method according to claim 14, wherein the plurality of operating modes comprises the first mode and a plurality of reduced power modes in which different numbers of blocks of data derived from each chunk of data are available.

16. A method according to claim 14, wherein operating the system in one of a plurality of operating modes selected based on system load comprises:

computing system load based on at least one of historical load data and measurements of system load; and

selecting an operating mode based on the computed system load.

17. A method according to claim 14, further comprising, on returning to the first mode from a reduced power mode:

copying any data stored in the versioned store and relating to a chunk of data to each server storing a block of data derived from the chunk of data.

18. A method according to claim 14, wherein each block of data derived from a chunk of data comprises a replica of the chunk of data and the method further comprising:

assigning one replica of each chunk of data as a primary replica; and

prior to entering a reduced power mode;

identifying any primary replicas stored in the at least one group of servers to be put into a low power state; and

for each identified primary replica, assigning a replica of the chunk of data stored on a server in another group as primary replica.

19. A method according to claim 18, further comprising:

assigning each chunk of data to a chunk group, wherein chunks within a chunk group are stored on the same servers and wherein the same server stores the primary replica for each chunk within a chunk group.

20. A storage system for storing a plurality of data objects, each data object comprising at least one chunk of data and the system comprising:

a plurality of servers arranged to store chunks of data and wherein each chunk is replicated across a number of servers;

a metadata service storing mappings between data objects and chunks of data;

a scheduling service arranged to predict system load and at times of reduced system load to schedule a subset of the servers to enter a low power mode such that a reduced number of replicas of a chunk of data are available; and

a versioned store arranged to store data written when at least one server is in a low power mode and to transfer the data stored to the servers when all servers have exited from the low power mode.