SYSTEM AND METHOD TO MAXIMIZE SERVER RESOURCE UTILIZATION AND PERFORMANCE OF METADATA OPERATIONS

• US 20140379722A1
• Filed: 06/24/2013
• Published: 12/25/2014
• Est. Priority Date: 06/24/2013
• Status: Active Grant

• Current Assignee: Hitachi, Ltd.
• Original Assignee: Hitachi, Ltd.
• Inventors: MYSUR, Kaushik, LIN, Wujuan, IKEDA, Hirokazu
• Granted Patent: US 9,197,703 B2
• Time in Patent Office: Days
• US Class Current: 707/740
• 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: G06F 3/067 Distributed or networked storage systems, e.g. storage area networks [SAN], network attached storage [NAS]
• CPC Class Codes: H04L 67/1097 for distributed storage of data in networks, e.g. transport arrangements for network file system [NFS], storage area networks [SAN] or network attached storage [NAS]

Abstract

An MDS (metadata server) in a distributed storage system includes data servers (DSs) storing file contents and one or more MDSs performing metadata operations in response to metadata requests of different types, the MDS including a controller having a processor and a memory, the MDS storing file system metadata. The controller is configured to: classify the metadata operations into different categories, which include a normal category and one or more special categories different from the normal category, the normal category having a primary stage which does not involve communication between the MDS and a component external to the MDS; for each special category, partition each metadata operation into a plurality of stages at least one of which involves communication between the MDS and a component external to the MDS; and dynamically assign resources to each of the partitioned stage based on monitored workloads of the different types of metadata requests.

Claims

1. An MDS (metadata server) in a distributed storage system which includes a plurality of data servers (DSs) storing file contents and one or more MDSs performing a plurality of metadata operations in response to metadata requests of different types, the MDS including a controller having a processor and a memory, the MDS storing file system metadata, the controller being configured to:

classify the metadata operations into different categories, which include a normal category and one or more special categories which are different from the normal category, the normal category having a primary stage which does not involve communication between the MDS and a component external to the MDS;

for each of the one or more special categories, partition each of the metadata operations into a plurality of stages at least one of which involves communication between the MDS and a component external to the MDS; and

dynamically assign resources to each of the partitioned stage based on monitored workloads of the different types of metadata requests.

2. The MDS of claim 1,

wherein the classifying comprises classifying the metadata operations into different categories based on type and amount of processing required for each category.

3. The MDS of claim 1,

wherein the one or more special categories of metadata operations comprise at least one of (i) a first special category of metadata operations that require communication between the MDS and one or more of the plurality of DSs;

or (ii) a second special category of metadata operations that require communication between the MDS and one or more other MDSs.

4. The MDS of claim 3,

wherein the stages for the first special category of metadata operations comprise the primary stage;

and

a pNFS manager stage for performing pNFS related metadata management including preparing requests to be sent to the DSs;

an MDS-DS asynchronous client stage for sending asynchronous requests to the DSs;

an asynchronous DS processing stage which is performed on the DSs for processing asynchronous requests from the MDS and sending a response back to the MDS after processing completes;

a DS response aggregator stage for receiving responses from the DSs and aggregating all the received responses to a single context representing a corresponding metadata operation; and

a secondary metadata processor stage for post processing on the MDS; and

wherein the stages for the second special category of metadata operations comprise the primary stage; and

an MDSC manager stage for performing MDS cluster management including preparing requests to be sent to the one or more other MDSs;

an inter-MDS asynchronous client stage for sending asynchronous requests to the one or more other MDSs;

an asynchronous inter-MDS processing stage which is performed on the one or more other MDSs;

an MDS response aggregator stage for receiving responses from the plurality of MDSs and aggregating all the received responses to a single context representing a corresponding metadata operation;

a secondary metadata processor for post processing on the MDS.

5. The MDS of claim 1,

wherein the partitioning comprises partitioning each metadata operation into a plurality of stages each of which (i) involves communication with a component external to the MDS, or (ii) involves a processing logic that is modularly different from its preceding processing logic and its succeeding processing logic.

6. The MDS of claim 5,

wherein a stage which involves communication with a component external to the MDS has a processing logic that treats the external component with which the stage of the metadata operation communicates as an asynchronous server component.

7. The MDS of claim 1, wherein the dynamically assigning resources comprises:

monitoring metadata workload of the normal category and the one or more special categories to obtain a number of metadata operations for each category and a total number of metadata operations for all categories;

calculating, for each special category, a ratio of the number of metadata operations for said each special category to the total number of metadata operations for all categories obtained from the monitoring;

calculating a processing time for each of the normal category and the one or more special categories; and

allocating a plurality of threads representing units of execution across all the stages by considering;

(i) an estimated processing time for each stage, (ii) the calculated ratio for each special category, (iii) the calculated processing time for each category, and (iv) a total number of threads allocated to the MDS.

8. The MDS of claim 7,

wherein the threads allocated to a particular special category are assigned to each stage involved in the particular special category in the ratio of the estimated processing time of each stage relative to the processing time of all the stages of the particular special category.

9. The MDS of claim 1, wherein the controller is configured to:

identify, from the one or more special categories of metadata operations, one or more candidate metadata operations to be executed in batch mode;

wherein the one or more candidate metadata operations each (i) has a potential to be batched together in a single network call to perform similar metadata operations speculatively, or (ii) has a potential to be locally completed asynchronously within the MDS and, at a later point in time, to be batched together with similar metadata operations to complete inter-server processing between the MDS and one or more external components.

10. The MDS of claim 9, wherein the controller is configured:

(i) when the one or more candidate metadata operations each has a potential to be batched together in a single network call to perform similar metadata operations speculatively, to identify data structures required to be stored in the memory of the MDS in order to perform a batch operation to speculatively fetch information from the external component;

or

(ii) when the one or more candidate metadata operations each has a potential to be locally completed asynchronously within the MDS and, at a later point in time, to be batched together with similar metadata operations to complete inter-server processing between the MDS and one or more external components, to identify a data consistency protocol for batch mode execution involving asynchronous processing.

11. The MDS of claim 9, wherein the controller is configured to:

count a total number of each candidate metadata operation to be executed in batch mode based on monitoring the metadata workload; and

when the total number of a particular candidate metadata operation exceeds a preset threshold for the particular candidate metadata operation, select the particular candidate metadata operation to be executed in batch mode.

12. The MDS of claim 11, wherein the dynamically assigning resources comprises:

monitoring metadata workload of the normal category and the one or more special categories to obtain a number of metadata operations for each category and a total number of metadata operations for all categories;

calculating, for each special category, a ratio of (the number of metadata operations for said each special category minus a number of metadata operations for said each special category which have been selected to be executed in batch mode) to the total number of metadata operations for all categories obtained from the monitoring;

calculating a processing time for each of the normal category and the one or more special categories; and

allocating a plurality of threads representing units of execution across all the stages by considering;

(i) an estimated processing time for each stage, (ii) the calculated ratio for each special category, (iii) the calculated processing time for each category, and (iv) a total number of threads allocated to the MDS.

13. A method of managing resources of an MDS (metadata server) in a distributed storage system which includes a plurality of data servers (DSs) storing file contents and one or more MDSs performing a plurality of metadata operations in response to metadata requests of different types, the MDS including a controller having a processor and a memory, the MDS storing file system metadata, the method comprising:

classifying the metadata operations into different categories, which include a normal category and one or more special categories which are different from the normal category, the normal category having a primary stage which does not involve communication between the MDS and a component external to the MDS;

for each of the one or more special categories, partitioning each of the metadata operations into a plurality of stages at least one of which involves communication between the MDS and a component external to the MDS; and

dynamically assigning resources to each of the partitioned stage based on monitored workloads of the different types of metadata requests.

14. The method of claim 13, wherein dynamically assigning resources comprises:

monitoring metadata workload of the normal category and the one or more special categories to obtain a number of metadata operations for each category and a total number of metadata operations for all categories;

calculating, for each special category, a ratio of the number of metadata operations for said each special category to the total number of metadata operations for all categories obtained from the monitoring;

calculating a processing time for each of the normal category and the one or more special categories; and

allocating a plurality of threads representing units of execution across all the stages by considering;

(i) an estimated processing time for each stage, (ii) the calculated ratio for each special category, (iii) the calculated processing time for each category, and (iv) a total number of threads allocated to the MDS.

15. The method of claim 13, further comprising:

identifying, from the one or more special categories of metadata operations, one or more candidate metadata operations to be executed in batch mode, wherein the one or more candidate metadata operations each (i) has a potential to be batched together in a single network call to perform similar metadata operations speculatively, or (ii) has a potential to be locally completed asynchronously within the MDS and, at a later point in time, to be batched together with similar metadata operations to complete inter-server processing between the MDS and one or more external components;

counting a total number of each candidate metadata operation to be executed in batch mode based on monitoring the metadata workload; and

when the total number of a particular candidate metadata operation exceeds a preset threshold for the particular candidate metadata, selecting the particular candidate metadata operation to be executed in batch mode;

wherein dynamically assigning resources comprises;

monitoring metadata workload of the normal category and the one or more special categories to obtain a number of metadata operations for each category and a total number of metadata operations for all categories;

calculating, for each special category, a ratio of (the number of metadata operations for said each special category minus a number of metadata operations for said each special category which have been selected to be executed in batch mode) to the total number of metadata operations for all categories obtained from the monitoring;

calculating a processing time for each of the normal category and the one or more special categories; and

allocating a plurality of threads representing units of execution across all the stages by considering;

(i) an estimated processing time for each stage, (ii) the calculated ratio for each special category, (iii) the calculated processing time for each category, and (iv) a total number of threads allocated to the MDS.