Replication is expensive -- the default 3x replication scheme in HDFS has 200% overhead in storage space and other resources (e.g., network bandwidth).
However, for warm and cold datasets with relatively low I/O activities, additional block replicas are rarely accessed during normal operations, but still consume the same amount of resources as the first replica.
Therefore, a natural improvement is to use Erasure Coding (EC) in place of replication, which provides the same level of fault-tolerance with much less storage space. In typical Erasure Coding (EC) setups, the storage overhead is no more than 50%.
In storage systems, the most notable usage of EC is Redundant Array of Inexpensive Disks (RAID). RAID implements EC through striping, which divides logically sequential data (such as a file) into smaller units (such as bit, byte, or block) and stores consecutive units on different disks. In the rest of this guide this unit of striping distribution is termed a striping cell (or cell). For each stripe of original data cells, a certain number of parity cells are calculated and stored -- the process of which is called encoding. The error on any striping cell can be recovered through decoding calculation based on surviving data and parity cells.
Integrating EC with HDFS can improve storage efficiency while still providing similar data durability as traditional replication-based HDFS deployments.
As an example, a 3x replicated file with 6 blocks will consume 6*3 = 18 blocks of disk space. But with EC (6 data, 3 parity) deployment, it will only consume 9 blocks of disk space.
In the context of EC, striping has several critical advantages. First, it enables online EC (writing data immediately in EC format), avoiding a conversion phase and immediately saving storage space. Online EC also enhances sequential I/O performance by leveraging multiple disk spindles in parallel; this is especially desirable in clusters with high end networking. Second, it naturally distributes a small file to multiple DataNodes and eliminates the need to bundle multiple files into a single coding group. This greatly simplifies file operations such as deletion, quota reporting, and migration between federated namespaces.
In typical HDFS clusters, small files can account for over 3/4 of total storage consumption. To better support small files, in this first phase of work HDFS supports EC with striping. In the future, HDFS will also support a contiguous EC layout. See the design doc and discussion on [HDFS-7285](https://issues.apache.org/jira/browse/HDFS-7285) for more information.
***NameNode Extensions** - Striped HDFS files are logically composed of block groups, each of which contains a certain number of internal blocks.
To reduce NameNode memory consumption from these additional blocks, a new hierarchical block naming protocol was introduced. The ID of a block group can be inferred from the ID of any of its internal blocks. This allows management at the level of the block group rather than the block.
On the output / write path, DFSStripedOutputStream manages a set of data streamers, one for each DataNode storing an internal block in the current block group. The streamers mostly
work asynchronously. A coordinator takes charge of operations on the entire block group, including ending the current block group, allocating a new block group, and so forth.
On the input / read path, DFSStripedInputStream translates a requested logical byte range of data as ranges into internal blocks stored on DataNodes. It then issues read requests in
parallel. Upon failures, it issues additional read requests for decoding.
***DataNode Extensions** - The DataNode runs an additional ErasureCodingWorker (ECWorker) task for background recovery of failed erasure coded blocks. Failed EC blocks are detected by the NameNode, which then chooses a DataNode to do the recovery work. The recovery task is passed as a heartbeat response. This process is similar to how replicated blocks are re-replicated on failure. Reconstruction performs three key tasks:
2._Decode the data and generate the output data:_ New data and parity blocks are decoded from the input data. All missing data and parity blocks are decoded together.
To accommodate heterogeneous workloads, we allow files and directories in an HDFS cluster to have different replication and EC policies.
Information on how to encode/decode a file is encapsulated in an ErasureCodingPolicy class. Each policy is defined by the following 2 pieces of information:
There are four policies currently being supported: RS-DEFAULT-3-2-64k, RS-DEFAULT-6-3-64k, RS-DEFAULT-10-4-64k and RS-LEGACY-6-3-64k. All with default cell size of 64KB. The system default policy is RS-DEFAULT-6-3-64k which use the default schema RS_6_3_SCHEMA with a cell size of 64KB.
Intel ISA-L stands for Intel Intelligent Storage Acceleration Library. ISA-L is a collection of optimized low-level functions used primarily in storage applications. It includes a fast block Reed-Solomon type erasure codes optimized for Intel AVX and AVX2 instruction sets.
HDFS EC can leverage this open-source library to accelerate encoding and decoding calculation. ISA-L supports most of major operating systems, including Linux and Windows. By default, ISA-L is not enabled in HDFS.
The default implementations for all of these codecs are pure Java. For default RS codec, there is also a native implementation which leverages Intel ISA-L library to improve the performance of codec. For XOR codec, a native implementation which leverages Intel ISA-L library to improve the performance of codec is also supported. Please refer to section "Enable Intel ISA-L" for more detail information.
1.`dfs.datanode.ec.reconstruction.stripedread.timeout.millis` - Timeout for striped reads. Default value is 5000 ms.
1.`dfs.datanode.ec.reconstruction.stripedread.threads` - Number of concurrent reader threads. Default value is 20 threads.
1.`dfs.datanode.ec.reconstruction.stripedread.buffer.size` - Buffer size for reader service. Default value is 64KB.
1.`dfs.datanode.ec.reconstruction.stripedblock.threads.size` - Number of threads used by the Datanode for background reconstruction work. Default value is 8 threads.
HDFS native implementation of default RS codec leverages Intel ISA-L library to improve the encoding and decoding calculation. To enable and use Intel ISA-L, there are three steps.
1. Build ISA-L library. Please refer to the offical site "https://github.com/01org/isa-l/" for detail information.
2. Build Hadoop with ISA-L support. Please refer to "Intel ISA-L build options" section in "Build instructions for Hadoop"(BUILDING.txt) document. Use -Dbundle.isal to copy the contents of the isal.lib directory into the final tar file. Deploy hadoop with the tar file. Make sure ISA-L library is available on both HDFS client and DataNodes.
3. Configure the `io.erasurecode.codec.rs-default.rawcoder` key with value `org.apache.hadoop.io.erasurecode.rawcoder.NativeRSRawErasureCoderFactory` on HDFS client and DataNodes.
To check ISA-L library enable state, try "Hadoop checknative" command. It will tell you if ISA-L library is enabled or not.
It also requires three steps to enable the native implementation of XOR codec. The first two steps are the same as the above step 1 and step 2. In step 3, configure the `io.erasurecode.codec.xor.rawcoder` key with `org.apache.hadoop.io.erasurecode.rawcoder.NativeXORRawErasureCoderFactory` on both HDFS client and DataNodes.
`policyName`: The ErasureCoding policy to be used for files under this directory. This is an optional parameter, specified using ‘-p’ flag. If no policy is specified, the system default ErasureCodingPolicy will be used.
Unset an ErasureCoding policy from a directory at the specified path when previously user sets the ErasureCoding policy on this directory via "setPolicy" command. If the directory inherits the ErasureCoding policy from its parent group, "unsetPolicy" command on this directory will not have any effect. Unset ErasureCoding policy on a directory which doesn't have ErasureCoding policy will not return an error.
