Monday, September 28, 2009

Notes of Key-Value Stores

Key-Value Stores:

It’s a storage system that stores values, indexed by a key. [1]

Sometimes, a key-value store works as a under layer storage system for a data management system.

When to consider Key-Value Stores:

- You need something that can query simple data faster than a relational system can.

- There are scaling requirements that are difficult to meet with a relational system.

- You want to avoid tying an application to a requirement for a relational system that may well have its own maintenance needs over time.

- You just want something simpler than a relational database.

- ……

In all of these cases, a key-value store may be just the tool for you.

Typical data structure:


Hash based key-value stores have fast lookup and update speeds, and require that keys be unique.

Lookup in a hash based tree is limited to key-only lookup.

For example:

To an object storage system which stores billions of photos, its applications would only query/get a photo by its unique key/id, like Facebook’s Haystack. [2]

The general API: Put/Get/Delete.

The important issues of hash based key-value stores are to limit the size of in-memory index/metadata, and to implement the distribution.

Tree (e.g. B-tree or B+tree)

Tree based stores generally allow multiple identical keys, and because tree based structures (like the b-tree) are ordered, they allow you to query individual keys, as well as ranges of keys.

The drawback to tree based stores is that those structures are generally slower than simple hashes.

For example:

A log aggregation system that stores messages keyed by timestamp, but which uses a hash based store, will be troublesome to query in a useful way, despite the speed of lookup, because you can only lookup specific timestamps. A system that uses a b-tree based store may be marginally slower for individual queries, but by permitting lookup based on a range of timestamps, can easily query the records of interest. This more than offsets the difference in basic query speeds. BigTable is based on B-Tree based key-value store. [3]

The general API: Put/Get/Delete/Scan.

The important issue of tree based key-value stores is to implement distributed tree structures.


[1] Engine Yard, Key-Value Stores in Ruby (Key-Value Stores Part 1)

[2] Facebook, Needle in a haystack: efficient storage of billions of photos,

[3] Data Management Projects at Google-200803,

Sunday, September 27, 2009

Tips of Hadoop Runtime Environment


To setup a cluster for running Hadoop/HBase/Hive, etc., besides the configuration and tuning of these open-source programs themselves, the system hardware and software, and some useful utilities, should also be considered to improve the system performance and to ease the system maintenance.

1. Cluster Facilities and Hardware [1]

(1) Data center:

Usually, we run Hadoop/HBase/Hive in a single data center.

(2) Servers:

Clusters are often either capacity bound or CPU bound.

The 1U or 2U configuration is usually used.

The storage capability of each node is usually not too dense (<= 4TB is recommended).

Commodity server: 2x4 core CPU, 16 GB RAM, 4x1TB SATA, 2x1 GE NIC

Use ECC RAM and cheap hard drives: 7200 RPM SATA.

Start with standard 64-bit box for masters and workers.

(3) Network:

Gigabit Ethernet, 2 level tree, 5:1 oversubscription to core

May want redundancy at top of rack and core

Usually, for a small cluster, all nodes are under a single GE switch.

(4) RAID configuration: RAID0

If there are two or more disks in each machine, RAID0 can provide better disk throughput than other RAID levels (and JBOD?). The multiple data replicas of HDFS can tolerate failure and guarantee the data safety.

2. System Software

(1) Linux:

RedHat5+ or CentOS5+ (recommended). Now, we use CentOS5.3-x64.

(2) Local File System:

Ext3 is ok.

We usually configure a separate disk partition for Hadoop used local file system, create and mount separate local file system for Hadoop.

Mount with noatime and nodiratime for performance improvements. Default, Linux will update the atime of files and directories, which is unnecessary in most cases. [2]

-- Edit /etc/fstab:

e.g. /dev/VolGroup00/LogVol02 /data ext3 defaults,noatime,nodiratime 1 2

-- remound:

mount -o remount /data

(3) Swappiness configuration: [3]

With the introduction of version 2.6, the new variable "swappiness" was added in the Linux kernel memory management subsystem and a tunable was created for it. High value of swappiness will make the kernel page out application text in favour of another application or even file-system cache. The default value is 60 (see mm/vmscan.c).

If you end up swapping, you're going to start seeing some weird behavior and very slow GC runs, and likely killing off HBase regionservers as ZooKeeper times out and assume the RegionServer is dead. Suggest setting vm.swappiness = 0 or other low number (e.g. 10), and observe the state of swap.

-- Edit /etc/sysctl.conf : vm.swappiness=0

-- To check the current value on a running system: sysctl vm.swappiness

(4) Linux default file handle limit: [4]

Currently HBase is a file handle glutton. To up the users' file handles, edit /etc/security/limits.conf on all nodes.

* - nofile 32768

(5) Java: JRE/JDK1.6 latest and GC options [5]

For machine with 4-16 cores, our Hadoop/HBase and other java applications should use GC option as: “-XX:+UseConcMarkSweepGC”.

For machine with 2 cores, should use GC option as: “-XX:+UseConcMarkSweepGC -XX:+CMSIncrementalMode”.

(6) Apache ANT 1.7.1

(7) Useful Linux utilities:

top, sar, iostat, iftop, vmstat, nfsstat, strace, dmesg, and friends

Especially iostat is very useful for disk I/O analysis.

(8) Useful java utilities:

jps, jstack, jconsole

(9) Compression native library: Gzip and LZO [7]

(10) Ganglia:

To integrate metrics of Hadoop, HBase, Hive, applications, and Linux system.


[1] Hadoop and Cloudera, Managing Petabytes with Open Source, Jeff Hammerbacher, Aug. 21 or

[2] Set noatime of local file system:

[3] Linux Swappiness:

[4] HBase FAQ:

[5] Java SE 6 HotSpot[tm] Virtual Machine Garbage Collection Tuning,

[6] Apache Ant:


Saturday, September 12, 2009

HFile: A Block-Indexed File Format to Store Sorted Key-Value Pairs

1. Intruduction

HFile is a mimic of Google’s SSTable. Now, it is available in Hadoop HBase-0.20.0. And the previous releases of HBase temporarily use an alternate file format – MapFile[4], which is a common file format in Hadoop IO package. I think HFile should also become a common file format when it becomes mature, and should be moved into the common IO package of Hadoop in the future.

Following words of SSTable are from section 4 of Google’s Bigtable paper.

The Google SSTable file format is used internally to store Bigtable data. An SSTable provides a persistent, ordered immutable map from keys to values, where both keys and values are arbitrary byte strings. Operations are provided to look up the value associated with a specified key, and to iterate over all key/value pairs in a specified key range. Internally, each SSTable contains a sequence of blocks (typically each block is 64KB in size, but this is configurable). A block index (stored at the end of the SSTable) is used to locate blocks; the index is loaded into memory when the SSTable is opened. A lookup can be performed with a single disk seek: we first find the appropriate block by performing a binary search in the in-memory index, and then reading the appropriate block from disk. Optionally, an SSTable can be completely mapped into memory, which allows us to perform lookups and scans without touching disk.[1]

The HFile implements the same features as SSTable, but may provide more or less.

2. File Format

Data Block Size

Whenever we say Block Size, it means the uncompressed size.

The size of each data block is 64KB by default, and is configurable in HFile.Writer. It means the data block will not exceed this size more than one key/value pair. The HFile.Writer starts a new data block to add key/value pairs if the current writing block is equal to or bigger than this size. The 64KB size is same as Google’s [1].

To achieve better performance, we should select different block size. If the average key/value size is very short (e.g. 100 bytes), we should select small blocks (e.g. 16KB) to avoid too many key/value pairs in each block, which will increase the latency of in-block seek, because the seeking operation always finds the key from the first key/value pair in sequence within a block.

Maximum Key Length

The key of each key/value pair is currently up to 64KB in size. Usually, 10-100 bytes is a typical size for most of our applications. Even in the data model of HBase, the key (rowkey+column family:qualifier+timestamp) should not be too long.

Maximum File Size

The trailer, file-info and total data block indexes (optionally, may add meta block indexes) will be in memory when writing and reading of an HFile. So, a larger HFile (with more data blocks) requires more memory. For example, a 1GB uncompressed HFile would have about 15600 (1GB/64KB) data blocks, and correspondingly about 15600 indexes. Suppose the average key size is 64 bytes, then we need about 1.2MB RAM (15600X80) to hold these indexes in memory.

Compression Algorithm

- Compression reduces the number of bytes written to/read from HDFS.

- Compression effectively improves the efficiency of network bandwidth and disk space

- Compression reduces the size of data needed to be read when issuing a read

To be as low friction as necessary, a real-time compression library is preferred. Currently, HFile supports following three algorithms:

(1)NONE (Default, uncompressed, string name=”none”)

(2)GZ (Gzip, string name=”gz”)

Out of the box, HFile ships with only Gzip compression, which is fairly slow.

(3)LZO(Lempel-Ziv-Oberhumer, preferred, string name=”lzo”)

To achieve maximal performance and benefit, you must enable LZO, which is a lossless data compression algorithm that is focused on decompression speed.

Following figures show the format of an HFile.

In above figures, an HFile is separated into multiple segments, from beginning to end, they are:

- Data Block segment

To store key/value pairs, may be compressed.

- Meta Block segment (Optional)

To store user defined large metadata, may be compressed.

- File Info segment

It is a small metadata of the HFile, without compression. User can add user defined small metadata (name/value) here.

- Data Block Index segment

Indexes the data block offset in the HFile. The key of each index is the key of first key/value pair in the block.

- Meta Block Index segment (Optional)

Indexes the meta block offset in the HFile. The key of each index is the user defined unique name of the meta block.

- Trailer

The fix sized metadata. To hold the offset of each segment, etc. To read an HFile, we should always read the Trailer firstly.

The current implementation of HFile does not include Bloom Filter, which should be added in the future.

3. LZO Compression

LZO is now removed from Hadoop or HBase 0.20+ because of GPL restrictions. To enable it, we should install native library firstly as following. [6][7][8][9]

(1) Download LZO:, and build.

# ./configure --build=x86_64-redhat-linux-gnu --enable-shared --disable-asm

# make

# make install

Then the libraries have been installed in: /usr/local/lib

(2) Download the native connector library, and build.

Copy hadoo-0.20.0-core.jar to ./lib.

# ant compile-native

# ant jar

(3) Copy the native library (build/native/ Linux-amd64-64) and hadoop-gpl-compression-0.1.0-dev.jar to your application’s lib directory. If your application is a MapReduce job, copy them to hadoop’s lib directory. Your application should follow the $HADOOP_HOME/bin/hadoop script to ensure that the native hadoop library is on the library path via the system property -Djava.library.path=. [9]

4. Performance Evaluation


4 slaves + 1 master

Machine: 4 CPU cores (2.0G), 2x500GB 7200RPM SATA disks, 8GB RAM.

Linux: RedHat 5.1 (2.6.18-53.el5), ext3, no RAID, noatime

1Gbps network, all nodes under the same switch.

Hadoop-0.20.0 (1GB heap), lzo-2.0.3

Some MapReduce-based benchmarks are designed to evaluate the performance of operations to HFiles, in parallel.

Total key/value entries: 30,000,000.

Key/Value size: 1000 bytes (10 for key, and 990 for value). We have totally 30GB of data.

Sequential key ranges: 60, i.e. each range have 500,000 entries.

Use default block size.

The entry value is a string, each continuous 8 bytes are a filled with a same letter (A~Z). E.g. “BBBBBBBBXXXXXXXXGGGGGGGG……”.

We set to avoid client side bottleneck.

(1) Write

Each MapTask for each range of key, which writes a separate HFile with 500,000 key/value entries.

(2) Full Scan

Each MapTask scans a separate HFile from beginning to end.

(3) Random Seek a specified key

Each MapTask opens one separate HFile, and selects a random key within that file to seek it. Each MapTask runs 50,000 (1/10 of the entries) random seeks.

(4) Random Short Scan

Each MapTask opens one separate HFile, and selects a random key within that file as a beginning to scan 30 entries. Each MapTask runs 50,000 scans, i.e. scans 50,000*30=1,500,000 entries.

This table shows the average entries which are written/seek/scanned per second, and per node.

In this evaluation case, the compression ratio is about 7:1 for gz(Gzip), and about 4:1 for lzo. Even through the compression ratio is just moderate, the lzo column shows the best performance, especially for writes.

The performance of full scan is much better than SequenceFile, so HFile may provide better performance to MapReduce-based analytical applications.

The random seek in HFiles is slow, especially in none-compressed HFiles. But the above numbers already show 6X~10X better performance than a disk seek (10ms). Following Ganglia charts show us the overhead of load, CPU, and network. The random short scan makes the similar phenomena.


[1] Google, Bigtable: A Distributed Storage System for Structured Data,

[2] HBase-0.20.0 Documentation,

[3] HFile code review and refinement.

[4] MapFile API:

[5] Parallel LZO: Splittable Compression for Hadoop.

[6] Using LZO in Hadoop and HBase:

[7] LZO:

[8] Hadoop LZO native connector library:

[9] Hadoop Native Libraries Guide: