Metadata storage system and management method for cluster file system

• US 9,449,005 B2
• Filed: 05/15/2014
• Issued: 09/20/2016
• Est. Priority Date: 11/17/2011
• Status: Active Grant

• Current Assignee: Huawei Technologies Co., Ltd. (Huawei Investment & Holding Co., Ltd.)
• Original Assignee: Huawei Technologies Co., Ltd. (Huawei Investment & Holding Co., Ltd.)
• Inventors: Chen, Zhuo, Xiong, Jin
• Primary Examiner(s): Dwivedi, Mahesh
• Assistant Examiner(s): Johnson, Johnese
• Application Number:
  US14/278,744
• Publication Number:
  US 20140250155A1
• Time in Patent Office: 859 Days
• Field of Search: 707/823
• US Class Current: 1/1
• CPC Class Codes: G06F 16/13 File access structures, e.g. distributed indices arrangements of input from, or output to, record carriers G06F3/06
• CPC Class Codes: G06F 16/1847 specifically adapted to static storage, e.g. adapted to flash memory or SSD
• CPC Class Codes: G06F 16/1858 Parallel file systems, i.e. file systems supporting multiple processors

Abstract

The present invention provides a metadata storage system and a management method for a cluster file system. The metadata of the cluster file system is stored by using a DRAM+SSD+HDD three-tier architecture, and the metadata is written sequentially into a metadata buffer in the DRAM based on write-in time. When a metadata buffer fully written, it is written back into the SSD sequentially to form a new metadata storage unit, and corresponding metadata index information in a header structure thereof is stored into the DRAM in a form of a metadata index unit.

Claims

1. A metadata storage system in a cluster file system, comprising:

a Dynamic Random Access Memory (DRAM);

a Solid-state Drive (SDD);

wherein the DRAM comprises multiple metadata buffers;

metadata in each metadata buffer is written sequentially based on write-in time, each metadata buffer corresponds to a header structure for storing metadata index information of the buffer;

a new first metadata index unit is formed in the DRAM each time data in a first metadata buffer is written back into the SSD, to store a metadata index information comprised in a header structure of the first metadata buffer; and

wherein the SSD comprises a metadata file, and the metadata file comprises multiple metadata storage units organized in a sequence of write-back time;

a new first metadata storage unit is formed in the SSD each time the data in the first metadata buffer is written back into the SSD, to store the metadata written into the SSD from the first metadata buffer; and

the first metadata storage unit corresponds to the first metadata index unit in the DRAM, wherein the first metadata buffer is anyone of the multiple metadata buffers in the DRAM.

2. The metadata storage system according to claim 1, further comprising a Hard Disk Drive (HDD) for storing metadata from the SSD.

3. The metadata storage system according to claim 1, wherein the header structure corresponding to the first metadata buffer comprises a Bloom Filter and a hash table, the Bloom Filter is used to record what metadata items are stored in the first metadata buffer, and the hash table is used to maintain a mapping between identifiers of the metadata items and offsets of the metadata items in the first metadata buffer.

4. The metadata storage system according to claim 1, wherein the SSD further comprises a metadata index file for storing the metadata index information.

5. The metadata storage system according to claim 1, wherein the DRAM further comprises a read buffer module, configured to buffer metadata read from the SSD.

6. A metadata management method for a metadata storage system, comprising followed steps, wherein the metadata storage system comprises a dynamic random access memory (DRAM) and a solid state drive (SSD), and the DRAM comprises multiple metadata buffers with each metadata buffer corresponding to a header structure for storing metadata index information of the buffer, the method comprising:

writing metadata sequentially into a first metadata buffer in the DRAM based on write-in time; and

when a first metadata buffer is fully written, writing the metadata in the first metadata buffer back into the SSD sequentially to form a new first metadata storage unit, and storing corresponding metadata index information in the header structure corresponding to the first metadata buffer into the DRAM in a form of a metadata index unit, wherein the first metadata storage unit corresponds to the first metadata index unit in the DRAM, the metadata storage unit is organized into a metadata file in a sequence of write-back time, and the first metadata buffer is any one of the multiple metadata buffers in the DRAM.

7. The method according to claim 6, wherein the header structure corresponding to the first metadata buffer comprises a Bloom Filter and a hash table, the Bloom Filter is used to record what metadata items are stored in the first metadata buffer, and the hash table is used to maintain a mapping between identifiers of the metadata items and offsets of the metadata items in the first metadata buffer.

8. The method according to claim 6, wherein writing the metadata in the first metadata buffer back into the SSD further comprises writing the corresponding metadata index information back into a metadata index file on the SSD sequentially when the metadata in the first metadata buffer is written back into the SSD sequentially.

9. The method according to claim 7, wherein writing the metadata in the first metadata buffer back into the SSD further comprises:

determining which data items on the SSD are already invalid according to the identifiers of the metadata items in the header structure when the metadata in the first metadata buffer is written back into the SSD; and

making an invalid mark for the metadata items in the first metadata index units corresponding to the first metadata storage units in which the already invalid metadata items are located.

10. The method according to claim 7, wherein the Bloom Filter is organized in a manner of combining a 64-ary tree and a link, a depth of each 64-ary tree is no more than 2 layers, and multiple 64-ary trees form a linked list, wherein a Bloom Filter corresponding to a non-leaf node of each 64-ary tree is an exclusive OR result of a Bloom Filter of a child node thereof.

11. The method according to claim 9, further comprising determining whether to migrate the metadata storage unit from the SSD to the HDD, wherein the metadata storage system further comprises a HDD, wherein determining whether to migrate the metadata storage unit from the SSD to the HDD comprises:

comparing access heat of the first metadata storage unit with a preset threshold to determine whether the first metadata storage unit needs to be migrated; and

when the access heat of the first metadata storage unit is smaller than the preset threshold, migrating the first metadata storage unit from the SSD to the HDD, otherwise writing valid metadata items in the first metadata storage unit into the metadata buffer in the DRAM.

12. The method according to claim 10, wherein the access heat of the metadata storage unit is calculated by using the following formula:

13. The method according to claim 11, further comprising:

directly migrating the first metadata storage units with a ratio of invalid metadata items over 50% from the SSD to the HDD without judging access heat thereof.

14. The method according to claim 11, wherein when a ratio of SSD available space to the SSD total space is smaller than a certain space threshold, determining whether to migrate the first metadata storage unit from the SSD to the HDD.

15. The method according to claim 14, wherein the space threshold ranges between 15% and 20%.

16. The method according to claim 11, wherein migrating the metadata storage unit from the SSD to the HDD is performed at a fixed time interval.

17. A cluster file system, comprising a metadata storage system comprising:

a Dynamic Random Access Memory (DRAM);

a Solid-state Drive (SDD);

wherein the DRAM comprises multiple metadata buffers;

metadata in each metadata buffer is written sequentially based on write-in time, each metadata buffer corresponds to a header structure for storing metadata index information of the buffer;

a new first metadata index unit is formed in the DRAM each time data in a first metadata buffer is written back into the SSD, to store a metadata index information comprised in a header structure of the first metadata buffer; and

wherein the SSD comprises a metadata file, and the metadata file comprises multiple metadata storage units organized in a sequence of write-back time;

a new first metadata storage unit is formed in the SSD each time the data in the first metadata buffer is written back into the SSD, to store the metadata written into the SSD from the first metadata buffer; and

the first metadata storage unit corresponds to the first metadata index unit in the DRAM, wherein the first metadata buffer is anyone of the multiple metadata buffers in the DRAM.