8. column oriented databases

Column-Oriented Databases
In Depth
Ciao
ciao
Vai a fare
ciao ciao
Dr. Fabio Fumarola

Outline
• Column-Oriented Introduction
• Bigtable:
– Features
– Data Model
– Rows and Column Family
– Timestamp
– API
– Implementation
• Other Open Source solutions
2

Column-Oriented Introduction
• We analyzed that key-value database are simple
hash table, where:
– All the access are via primary keys and return an object as
result.
• Column-Oriented databases are motivated by the
necessity to model something more than object-
values.
3

Bigtable
• It is used pervasively at Google in several projects:
– Google Analytics, Google Finance, Google Plus,
– Personalized Search, Google Docs, and Google Maps.
• These products use Bigtable for a variety of
demanding workloads ranging from:
– Batch oriented processing to
– Low latency serving of data to end users
4

Bigtable: Main Characteristics
• It does not support a full relational data model
• It provides clients with a simple data model that
– Supports dynamic control over data layout and format
– Allows clients to reason about the locality of the data in
the underlying storage
• Data is indexed using row and column names that
can be arbitrary strings
• Data is stored as uninterpreted strings supporting
several kind of data.
5

Data Model
• A Bigtable is a sparse, distributed and persistent
multi-dimensional sorted map.
• The map is indexed at the first lever by a row-key
• And at second and third level by a column-key and a
timestamp respectively.
• Each value in the map is an array of bytes
(row:string, column:string, time:int64) → string
6

Data Model
• They settled on this data model after reasoning on
potential uses of Bigtable.
• The example that drove their decision was storing
the crawling results and to enable analysis.
• The main goal is to “keep a copy of a large collection
of web pages and related information that could be
used by many different projects” (Cache).
8

Webtable example
• The URLs is used as row keys
• various aspects of web pages as column names,
• and store the contents of the web pages in the contents:
column under the timestamps when they were fetched.
9

Data Model: Rows
• The row keys are arbitrary strings (up to 64KB)
• Every read or write of data under a single row key is
atomic (no considering the number of different
columns being read or written).
• Data are maintained in lexicographic order by row
key.
• Ordered set of row-key are grouped in ranges and
dynamically partitioned
10

Data Model: Rows
• Each row range is called tablet,
• Tablets are the unit of distribution and load
balancing.
• As a result, reads of short row ranges are
efficient and typically require communication
with only a small number of machines.
• For example, in Webtable, pages in the same
domain are grouped together into contiguous
rows by reversing the hostname components
of the URLs.
11

Data Model: Column Families
• Column keys are grouped into sets called column
families.
• All data stored in a column family is usually of the
same type (values of the same family are
compressed together).
• The design philosophy is to have a low number of
column-families (< 101).
• In contrast, a table may have an unbounded number
of columns.
12

• A column key is named using the following syntax:
family:qualifier.
• Column family names must be printable, but
qualifiers may be arbitrary strings.
• An example of column family is language
• It uses only one column key in the language family,
and it stores each web page’s language ID.
13

• Another useful column family for this table is anchor;
• Each column key in this family represents a single anchor
• The qualifier is the name of the referring site; the cell
contents is the link text.
14

Timestamps
• Each cell in a Bigtable can contain multiple versions
of the same data;
• These versions are indexed by timestamp in
microseconds.
• Bigtable timestamps are 64-bit integers.
• Different versions of a cell are stored in decreasing
timestamp order, so that the most recent versions
can be read first.
15

Timestamps
• It normally stores 3 versions of the same data
• Stale data is garbage-collected automatically
• In our Webtable example, we set the timestamps of
the crawled pages stored in the contents.
16

API
• Operator supported:
– put(key, columnFamily, columnQualifier, value)
– get(key)
– Scan(startKey, endKey)
– delete(key)
• It also provides functions for changing cluster, table,
and column family metadata, such as access control
rights.
17

API: Get & Scan
• A Get is simply a Scan limited by the API to one row.
• A Scan fetches zero or more rows of a table.
• By default, a Scan reads the entire table from start to
end.
– Scan()
– Scan(byte[] startRow)
– Scan(byte[] startRow, byte[] stopRow)
18

API
• A scan can be limited in several ways:
– Start and Stop row
– Caching size: to limit the number of rows:
– Batch size: to limit the number of columns
– Max result size: to limit the number of bytes
– Setting the column names and qualifiers
• Bigtable can be used with MapReduce, a frame-
work for running large-scale parallel computations
developed at Google.
19

Building Blocks
• Google File System (GFS)
• Google SSTable
• Distributed lock service called Chubby (Zookeper)
20

Google File Systems
• Bigtable uses the distributed Google File System
(GFS) to store log and data files.
21
http://static.googleusercontent.com/media/research.google.com/it//
archive/gfs-sosp2003.pdf

Google SSTable
• File format is used internally to store Bigtable data.
• An SSTable provides a persistent, ordered immutable
map from keys to values.
• Provided operations are:
– lookup the value associated with a key
– To iterate over all key/value pairs in a specified key range
• Internally, each SSTable contains a sequence of
blocks (typically of 64KB size)
22

Google SSTable
• 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.
23

Chubby
• It is a highly-available and persistent distributed lock service.
• It consists by at least five replicas, one of which is elected as
master.
• Chubby provides a namespace that consists of directories and
small files.
24
http://static.googleusercontent.com/media/research.go
ogle.com/it//archive/chubby-osdi06.pdf

Chubby and Bigtable
• Bigtable uses Chubby for a variety of tasks:
– to ensure that there is at most one active master at any
time;
– to store the bootstrap location of Bigtable data;
– to discover tablet servers and finalize tablet server
deaths;
– to store Bigtable schema information (the column family
information for each table);
– and to store access control lists.
25

Implementation
• It is composed by three major components:
– a library that is linked into every client,
– one master server,
– and many tablet servers
• Tablet servers can be dynamically added (or
removed) from a cluster to accomodate changes in
workloads.
26

Implementation: Master
• It is responsible for:
– assigning tablets to tablet servers,
– detecting the addition and expiration of tablet servers,
– balancing tablet-server load,
– and garbage collection of files in GFS.
• In addition, it handles schema changes such as table
and column family creations.
27

Implementation: Tablet Server
• Each tablet server manages a set of tablets (typically
we have somewhere between ten to a thousand
tablets per tablet server).
• The tablet server handles read and write requests to
the tablets that it has loaded, and also splits tablets
that have grown too large.
28

Implementation: Clients
• clients communicate directly with tablet servers for
reads and writes.
• Because Bigtable clients do not rely on the master
for tablet location information, most clients never
communicate with the master.
• As a result, the master is lightly loaded in practice
(differently from HBase).
29

Implementation: Cluster
• A Bigtable cluster stores a number of tables.
• Each table consists of a set of tablets, and each
tablet contains all data associated with a row range.
• Initially, each table consists of just one tablet. As a
table grows, it is automatically split into multiple
tablets, each approximately 100-200 MB in size by
default.
30

Tablet Location
• It is used a three-level hierarchy similar to a B+ tree
to store tablet location information
32

Tablet Location
• The first level is a file stored in Chubby that contains the
location of the root tablet.
• The root tablet contains the location of all tablets in a special
METADATA table.
33

Tablet Location
• The second level contains the location of a set of user tablets.
• The METADATA table stores the location of a tablet under a
row key.
34

Tablet Location
• The third level contains the data.
• The client library caches tablet locations.
• If the client does not know the location of a tablet then it
recursively moves up the tablet location hierarchy.
35

Tablet Location
• They also store secondary information in the
METADATA table, including a log of all events
pertaining to each tablet.
• This information is helpful for debugging and
performance analysis.
36

Tablet Assignment and Serving
37

Tablet Assignment
• Each tablet is assigned to one tablet server at a time.
• The master keeps track of the set of live tablet
servers, that is assigned and unassigned tablets.
• When a tablet is unassigned, and a tablet server with
sufficient room for the tablet is available, the master
assigns the tablet by sending a tablet load request to
the tablet server.
38

Tablet Assignment
• Bigtable uses Chubby to keep track of tablet servers.
• When a tablet server starts, it creates, and acquires
an exclusive lock on, a uniquely-named file in a
specific Chubby directory
• When a tablet server ends the master is responsible
for reassigning those tablets as soon as possible.
39

Tablet Serving
• The persistent state of a tablet is stored in GFS
• Updates are committed to a commit log that stores redo
records
40
It stores recently
committed updates
It stores the older
updates

Tablet Recover
• It is done by a tablet server reading tablet metadata
from the METADATA table.
• This table contains an event log with a set of redo
point, which are pointer to commit logs containing
data for the tablet.
• The server reads the indices of the SSTables into
memory and reconstructs the memtable by applying
all of the updates that have committed since the
redo points.
41

Tablet Serving
• When a read operation arrives at a tablet server, the
operation is executed on a merged view of the
sequence of SSTables and the memtable.
• This operation is efficient because SSTable and
memtable are lexicographically sorted data
structures.
• Incoming read and write operations can continue
while tablets are split and merged.
42

Compactions
• As write operations execute, the size of the
memtable increases (immutable).
• When the memtable size reaches a threshold, the
memtable is frozen, a new memtable is created, and
the frozen memtable is converted to an SSTable and
written to GFS.
• There are three type of compactions: Minor,
Merging, and Major.
43

Minor Compactions
It has two goals:
•it shrinks the memory usage of the tablet server,
•it reduces the amount of data that has to be read from
the commit log during recovery if this server dies.
Incoming read and write operations can continue while
compactions occur.
44

Merging Compactions
• Every minor compaction generates a new SSTable.
• Read operations might need to merge updates from
an arbitrary number of SSTables.
• This operations are bounded by allowing periodically
merging compaction.
• A merging compaction reads the contents of a few
SSTables and the memtable, and writes out a new
SSTable.
45

Major Compaction
• It is a compaction that rewrites all SSTables into
exactly one SSTable.
• SSTables produced by non-major compactions can
contain special deletion entries.
• A major compaction, on the other hand, produces an
SSTable that contains no deletion information or
deleted data.
• These major compactions allow Bigtable to reclaim
resources used by deleted data.
46

Locality groups
• Clients can group multiple column families together
into a locality group.
• A separate SSTable is generated for each locality
group in each tablet.
• Segregating column families that are not typically
accessed together into separate locality groups
enables more efficient reads.
48

Compression
• A user-specified compression format is applied to
each SSTable block.
• A used compression approach is based on:
– first pass uses Bentley and McIlroy’s scheme
– The second pass uses a fast compression algorithm that
looks for repetitions in a small 16 KB window of the data.
• This scheme applied to Webtable achieved a 10-to-1
reduction in space.
49

Caching for read
To improve read performance, tablet servers use two
levels of caching.
•The Scan Cache is a higher-level cache that caches the
key-value pairs returned by the SSTable interface to the
tablet server code.
•The Block Cache is a lower-level cache that caches
SSTables blocks that were read from GFS.
50

Bloom filters
• The number of accesses are reduced by allowing
clients to specify that Bloom filters should be create
for SSTables.
• A Bloom filter allows us to ask whether an SSTable
contains any data for a specified row/column pair.
• http://billmill.org/bloomfilter-tutorial/
51

Commit-log
• All the operation of write and delete are persisted on
a global commit-log (inter-tablet).
• Using one log provides significant performance
benefits during normal operation, but it complicates
recovery.
• To recover the state for a tablet, the new tablet
server needs to reapply the mutations for that tablet
from the commit log written by the original tablet
server.
52

Commit-log
• To avoid duplicating log reads, it is at firs sorted by
table, row name, log sequence number .⟨ ⟩
• In the sorted output, all mutations for a particular
tablet are contiguous and can therefore be read
efficiently with one disk seek followed by a
sequential read.
• To parallelize the sorting, the log file is partitioned
into files of 64MB and sorted using MapReduce.
53

Exploiting immutability
• The Bigtable system is simplified by the fact that all of the
SSTables generated are immutable.
• This allows us to not do any sync of accesses.
• The only mutable data structure that is accessed by both
reads and writes is the memtable.
• The immutability of SSTables enables us to split tablets
quickly.
• Instead of generating a new set of SSTables for each child
tablet, we let the child tablets share the SSTables of the
parent tablet.
54

Lessons
• Distributed systems are vulnerable to many types of
failures: memory and network corruption, large clock
skew, hung machines, extended and asymmetric
network partitions, bugs in other systems.
• Add new features only when you know how to use it.
56

Lessons
• A practical lesson learned from supporting Bigtable is
the importance of proper system-level monitoring.
• Clarity of Design. “Given both the size of our system
(about 100,000 lines of non-test code), as well as the
fact that code evolves over time in unexpected ways,
we have found that code and design clarity are of
immense help in code maintenance and debugging.”
57

Open Source Column Stores
• Hbase (http://hbase.apache.org/)
• Hypertable (http://hypertable.org/)
• Cassandra (http://cassandra.apache.org/)
• Accumulo (http://accumulo.apache.org/)
• Parquet (http://parquet.io/)
59

HBase
Apache HBase™ is the Hadoop database to use when you need
you need random, realtime read/write access to your Data.
•Automatic and configurable sharding of tables
•Automatic failover support between RegionServers.
•Convenient base classes for backing Hadoop MapReduce jobs
with Apache HBase tables.
•Easy to use Java API for client access.
•To be distributed, it has to run on top of hdfs
•Integrated with MapReduce
60

Data Location, Assignment and
Serving
62

Coprocessors
63
https://blogs.apache.org/hbase/entry/coprocessor_introduction

Hypertable
• Hypertable is an open source database system
inspired by publications on the design of Google's
BigTable.
• Hypertable runs on top of a distributed file system
such as the Apache Hadoop DFS, GlusterFS, or the
Kosmos File System (KFS). It is written almost entirely
in C++.
64
http://hypertable.com/documentation/

Cassandra
• Big-Table extension:
– All nodes are similar.
– Can be used as a distributed hash-table, with an "SQL-like"
language, CQL (but no JOIN!)
• Data can have expiration (set on INSERT)
• Map/reduce possible with Apache Hadoop
• Rich Data Model (columns, composites, counters,
secondary indexes, map, set, list, counters)
65

Reading from Cassandra
clientclient
memtable
sstable
sstable
sstable
Row cache
key cache

Writing to Cassandra
clientclient Commit
log (Disk)
Memtable
(memory)
sstable
Flush
Replication factor: 3Replication factor: 3
sstable sstablesstable

When not to use
• There are not suitable for systems:
– that requires ACID transactions for writes and reads,
– That needs to aggregate data (SUM and AVG) you need to
do it at client side,
– Are in an initial stage. If we need to change the queries we
could have to change the columns.
71

8. column oriented databases

More Related Content

What's hot (20)

Similar to 8. column oriented databases (20)

More from Fabio Fumarola (20)

Recently uploaded (20)

8. column oriented databases

Editor's Notes