There is a lot of confusion about the place and purpose of the many recent non-relational database solutions ("NoSQL databases") compared to the relational database solutions that have been around for so many years. In this presentation I will first clarify what exactly these database solutions are, compare them, and discuss the best use cases for each. I'll discuss topics involving OLTP, scaling, data warehousing, polyglot persistence, and the CAP theorem. We will even touch on a new type of database solution called NewSQL. If you are building a new solution it is important to understand all your options so you take the right path to success.
2. About Me
Microsoft, Big Data Evangelist
In IT for 30 years, worked on many BI and DW projects
Worked as desktop/web/database developer, DBA, BI and DW architect and developer, MDM
architect, PDW/APS developer
Been perm employee, contractor, consultant, business owner
Presenter at PASS Business Analytics Conference, PASS Summit, Enterprise Data World conference
Certifications: MCSE: Data Platform, Business Intelligence; MS: Architecting Microsoft Azure
Solutions, Design and Implement Big Data Analytics Solutions, Design and Implement Cloud Data
Platform Solutions
Blog at JamesSerra.com
Former SQL Server MVP
Author of book “Reporting with Microsoft SQL Server 2012”
3. Agenda
Definition and differences
ACID vs BASE
Four categories of NoSQL
Use cases
CAP theorem
On-prem vs cloud
Product categories
Polyglot persistence
Architecture samples
4. Goal
My goal is to give you a high level overview of all the technologies so you know where to start and put you on
the right path to be a hero!
5. Relational and non-relational defined
Relational databases (RDBMS, SQL Databases)
• Example: Microsoft SQL Server, Oracle Database, IBM DB2
• Mostly used in large enterprise scenarios
• Analytical RDBMS (OLAP, MPP) solutions are Analytics Platform System, Teradata, Netezza
Non-relational databases (NoSQL databases)
• Example: Azure Cosmos DB, MongoDB, Cassandra
• Four categories: Key-value stores, Wide-column stores, Document stores and Graph stores
Hadoop: Made up of Hadoop Distributed File System (HDFS), YARN and MapReduce
6. Origins
Using SQL Server, I need to index a few thousand documents and search them.
No problem. I can use Full-Text Search.
I’m a healthcare company and I need to store and analyze millions of medical claims per day.
Problem. Enter Hadoop.
Using SQL Server, my internal company app needs to handle a few thousand transactions per second.
No problem. I can handle that with a nice size server.
Now I have Pokémon Go where users can enter millions of transactions per second.
Problem. Enter NoSQL.
But most enterprise data just needs an RDBMS (89% market share – Gartner).
7. Main differences (Relational)
Pros
• Works with structured data
• Supports strict ACID transactional consistency
• Supports joins
• Built-in data integrity
• Large eco-system
• Relationships via constraints
• Limitless indexing
• Strong SQL
• OLTP and OLAP
• Most off-the-shelf applications run on RDBMS
8. Main differences (Relational)
Cons
• Does not scale out horizontally (concurrency and data size) – only vertically, unless use sharding
• Data is normalized, meaning lots of joins, affecting speed
• Difficulty in working with semi-structured data
• Schema-on-write
• Cost
9. Main differences (Non-relational/NoSQL)
Pros
• Works with semi-structured data (JSON, XML)
• Scales out (horizontal scaling – parallel query performance, replication)
• High concurrency, high volume random reads and writes
• Massive data stores
• Schema-free, schema-on-read
• Supports documents with different fields
• High availability
• Cost
• Simplicity of design: no “impedance mismatch”
• Finer control over availability
• Speed, due to not having to join tables
10. Main differences (Non-relational/NoSQL)
Cons
• Weaker or eventual consistency (BASE) instead of ACID
• Limited support for joins, does not support star schema
• Data is denormalized, requiring mass updates (i.e. product name change)
• Does not have built-in data integrity (must do in code)
• No relationship enforcement
• Limited indexing
• Weak SQL
• Limited transaction support
• Slow mass updates
• Uses 10-50x more space (replication, denormalized, documents)
• Difficulty tracking schema changes over time
• Most NoSQL databases are still too immature for reliable enterprise operational applications
11. Main differences (Hadoop)
Pros
• Not a type of database, but rather a open-source software ecosystem that allows for massively
parallel computing
• No inherent structure (no conversion to relational or JSON needed)
• Good for batch processing, large files, volume writes, parallel scans, sequential access
• Great for large, distributed data processing tasks where time isn’t a constraint (i.e. end-of-day
reports, scanning months of historical data)
• Tradeoff: In order to make deep connections between many data points, the technology
sacrifices speed
• Some NoSQL databases such as HBase are built on top of HDFS
12. Main differences (Hadoop)
Cons
• File system, not a database
• Not good for millions of users, random access, fast individual record lookups or updates (OLTP)
• Not so great for real-time analytics
• Lacks: indexing, metadata layer, query optimizer, memory management
• Same cons at non-relational: no ACID support, data integrity, limited indexing, weak SQL, etc
• Security limitations
• More complex debugging
Hadoop adoption has slowed
• Too much hype
• Companies adopt is without understanding use cases (i.e. real big data)
• Difficulty in finding skillset
• Pace of change too fast
• Too many products involved in a solution
• Other technologies (RDBMS, NoSQL) improving and expanding use cases
• Higher learning curve
13. ACID (RDBMS) vs BASE (NoSQL)
ATOMICITY: All data and commands in a
transaction succeed, or all fail and roll back
CONSISTENCY: All committed data must be
consistent with all data rules including
constraints, triggers, cascades, atomicity,
isolation, and durability
ISOLATION: Other operations cannot access
data that has been modified during a
transaction that has not yet completed
DURABILITY: Once a transaction is
committed, data will survive system failures,
and can be reliably recovered after an
unwanted deletion
Needed for bank transactions
Basically Available: Guaranteed Availability
Soft-state: The state of the system may change, even
without a query (because of node updates)
Eventually Consistent: The system will become
consistent over time
Ok for web page visits
ACID BASE
Strong Consistency Weak Consistency – stale data OK
Isolation Last Write Wins
Transaction Programmer Managed
Available/Consistent Available/Partition Tolerant
Robust Database/Simpler Code Simpler Database, Harder Code
14. Data stored in tables.
Tables contain some number of columns, each of a type.
A schema describes the columns each table can have.
Every table’s data is stored in one or more rows.
Each row contains a value for every column in that table.
Rows aren’t kept in any particular order.
15. Thanks to: Harri Kauhanan, http://www.slideshare.net/harrikauhanen/nosql-3376398
Relational stores
16. Key-value stores offer very high speed via the
least complicated data model—anything can
be stored as a value, as long as each value is
associated with a key or name.
Key Value
18. Wide-column stores are fast and can be nearly as simple as key-value stores. They include a primary
key, an optional secondary key, and anything stored as a value. Also called column stores
Values
Primary key Keys and values can
be sparse or
numerous
Secondary
key
24. Use cases for NoSQL categories
• Key-value stores: [Redis] For cache, queues, fit in memory, rapidly changing data, store blob data.
Examples: shopping cart, session data, leaderboards, stock prices. Fastest performance
• Wide-column stores: [Cassandra] Real-time querying of random (non-sequential) data, huge
number of writes, sensors. Examples: Web analytics, time series analytics, real-time data analysis,
banking industry. Internet scale
• Document stores: [MongoDB] Flexible schemas, dynamic queries, defined indexes, good
performance on big DB. Examples: order data, customer data, log data, product catalog, user
generated content (chat sessions, tweets, blog posts, ratings, comments). Fastest development
• Graph databases: [Neo4j] Graph-style data, social network, master data management, network and
IT operations. Examples: social relations, real-time recommendations, fraud detection, identity and
access management, graph-based search, web browsing, portfolio analytics, gene sequencing, class
curriculum
Note: Many NoSQL solutions are now multi-model
25. Velocity
Volume Per
Day
Real-world
Transactions
Per Day
Real-world
Transactions
Per Second
Relational DB Document DB Key Value or
Wide Column
8 GB 8.64B 100,000 As Is
86 GB 86.4B 1M Tuned* As Is
432 GB 432B 5M Appliance Tuned* As Is
864 GB 864B 10M Clustered
Appliance
Clustered
Servers
Tuned*
8,640 GB 8.64T 100M Many
Clustered
Servers
Clustered
Servers
43,200 GB 43.2T 500M Many
Clustered
Servers
* Tuned means tuning the model, queries, and/or hardware (more CPU, RAM, and Flash)
26. Focus of different data models
…you may not have the data volume for NoSQL (yet), but there are other reasons to use
NoSQL (semi-structured data, schemaless, high availability, etc)
27. Relational NewSQL stores are designed for web-scale
applications, but still require up-front schemas, joins, and
table management that can be labor intensive.
Blend RDBMS with NoSQL: provide the same scalable
performance of NoSQL systems for OLTP read-write
workloads while still maintaining the ACID guarantees of
a traditional relational database system.
28. Use case for different database technologies
• Traditional OLTP business systems (i.e. ERP, CRM, In-house app): relational database (RDBMS)
• Data warehouses (OLAP): relational database (SMP or MPP)
• Web and mobile global OLTP applications: non-relational database (NoSQL)
• Data lake: Hadoop
• Relational and scalable OLTP: NewSQL
29. CAP Theorem
Impossible for any shared data system to guarantee simultaneously all of the
following three properties:
Consistency: Once data is written, all future requests will contain the data. “Is
the data I’m looking at now the same if I look at it somewhere else?”
Availability: The database is always available and responsive. “What happens
if my database goes down?”
Partitioning: If part of the database is unavailable, other parts are unaffected.
“What if my data is on a different node?”
Relational: CA (i.e. SQL Server with no replication)
Non-relational: AP (Cassandra, CoachDB, Riak); CP (Hbase, Cosmos DB, MongoDB, Redis)
NoSQL can’t be both consistent and available. If two nodes (A and B) and B goes down, if
the A node takes requests, it is available but not consistent with B node. If A node stops
taking requests, it remains consistent with B node but it is not available. RDBMS is
consistent and available because it only has one node/partition (so no partition tolerance)
30. Microsoft data platform solutions
Product Category Description More Info
SQL Server 2016 RDBMS Earned top spot in Gartner’s Operational Database magic
quadrant. JSON support
https://www.microsoft.com/en-us/server-
cloud/products/sql-server-2016/
SQL Database RDBMS/DBaaS Cloud-based service that is provisioned and scaled quickly.
Has built-in high availability and disaster recovery. JSON
support
https://azure.microsoft.com/en-
us/services/sql-database/
SQL Data Warehouse MPP RDBMS/DBaaS Cloud-based service that handles relational big data.
Provision and scale quickly. Can pause service to reduce
cost
https://azure.microsoft.com/en-
us/services/sql-data-warehouse/
Analytics Platform System (APS) MPP RDBMS Big data analytics appliance for high performance and
seamless integration of all your data
https://www.microsoft.com/en-us/server-
cloud/products/analytics-platform-
system/
Azure Data Lake Store Hadoop storage Removes the complexities of ingesting and storing all of
your data while making it faster to get up and running with
batch, streaming, and interactive analytics
https://azure.microsoft.com/en-
us/services/data-lake-store/
Azure Data Lake Analytics On-demand analytics job
service/Big Data-as-a-
service
Cloud-based service that dynamically provisions resources
so you can run queries on exabytes of data. Includes U-
SQL, a new big data query language
https://azure.microsoft.com/en-
us/services/data-lake-analytics/
HDInsight PaaS Hadoop compute A managed Apache Hadoop, Spark, R, HBase, and Storm
cloud service made easy
https://azure.microsoft.com/en-
us/services/hdinsight/
Azure Cosmos DB PaaS NoSQL: Document
Store
Get your apps up and running in hours with a fully
managed NoSQL database service that indexes, stores, and
queries data using familiar SQL syntax
https://azure.microsoft.com/en-
us/services/cosmos-db/
Azure Table Storage PaaS NoSQL: Key-value
Store
Store large amount of semi-structured data in the cloud https://azure.microsoft.com/en-
us/services/storage/tables/
32. Azure Cosmos DB consistency options
• Strong, which is the slowest of the four, but is guaranteed to always return correct data
• Bounded staleness, which ensures that an application will see changes in the order in which they were
made. This option does allow an application to see out-of-date data, but only within a specified
window, e.g., 500 milliseconds
• Session, which ensures that an application always sees its own writes correctly, but allows access to
potentially out-of-date or out-of-order data written by other applications
• Consistent Prefix (new), updates returned are some prefix of all the updates, with no gaps
• Eventual, which provides the fastest access, but also has the highest chance of returning out-of-date
data
33. On-prem vs Cloud
• On-prem: SQL Server, APS, MongoDB, Oracle, Cassandra, Neo4J
• IaaS Cloud: SQL Server in Azure VM, Oracle in Azure VM
• DBaaS/PaaS Cloud: SQL Database, SQL Data Warehouse, Azure Cosmos DB, Redshift, RDS, MongoLab
38. db-engines.com/en/ranking
Method of calculation:
• Number of mentions of the system
on websites
• General interest in the system
• Frequency of technical discussions
about the system
• Number of job offers, in which the
system is mentioned
• Number of profiles in professional
networks, in which the system is
mentioned
• Relevance in social networks
db-engines.com/en/ranking_definition
40. Polyglot Persistence
• Sometimes a relational store is the right choice, sometimes a NoSQL store is the right choice
• Sometimes you need more than one store: Using the right tool for the right job
41.
42.
43. Summary
Choose NoSQL when…
• You are bringing in new data with a lot of volume and/or variety
• Your data is non-relational/semi-structured
• Your team will be trained in these new technologies (NoSQL)
• You have enough information to correctly select the type and product of NoSQL for your situation
• You can relax transactional consistency when scalability or performance is more important
• You can service a large number of user requests vs rigorously enforcing business rules
Relational databases are created for strong consistency, but at the cost of speed and scale. NoSQL slightly sacrifices
consistency across nodes for both speed and scalability.
NoSQL and Hadoop are viable technologies for a subset of specialized needs and use cases.
Lines are getting blurred – do your homework!
44. Bottom line!
• RDBMS for enterprise OLTP and ACID compliance, or db’s under 5TB
• NoSQL for scaled OLTP and JSON documents
• Hadoop for big data analytics (OLAP)
45. Resources
Relational database vs Non-relational databases: http://bit.ly/1HXn2Rt
Types of NoSQL databases: http://bit.ly/1HXn8Zl
What is Polyglot Persistence? http://bit.ly/1HXnhMm
Hadoop and Data Warehouses: http://bit.ly/1xuXfu9
Hadoop and Microsoft: http://bit.ly/20Cg2hA
46. Q & A ?
James Serra, Big Data Evangelist
Email me at: JamesSerra3@gmail.com
Follow me at: @JamesSerra
Link to me at: www.linkedin.com/in/JamesSerra
Visit my blog at: JamesSerra.com (where this slide deck is posted via the “Presentations” link on the top menu)
Editor's Notes
There is a lot of confusion about the place and purpose of the many recent non-relational database solutions (“NoSQL databases”) compared to the relational database solutions that have been around for so many years. In this presentation I will first clarify what exactly these database solutions are, how they compare to Hadoop, and discuss the best use cases for each. I’ll discuss topics involving OLTP, scaling, data warehousing, polyglot persistence, and the CAP theorem. We will even touch on a new type of database solution called NewSQL. If you are building a new solution it is important to understand all your options so you take the right path to success.
Fluff, but point is I bring real work experience to the session
My goal is to give you a high level overview of all the technologies so you know where to start Make you a hero
Hadoop started 2006. NoSQL started 2009
DocumentDB has done 5m/tps per region for 4 regions, so 20m/tps. DocumentDB uses local storage
Kevin Cox: What is the highest performance (transactions per second) you have seen out of SQL Server? Over 500k/sec. Very dependent on using flash-type storage for tran log; i.e. FusionIO or similar. Also short transactions (stock trades).
Matt Goswell: Please see attached. SDX offers 171,800 TPS however this is using SQL 2014. We are waiting on updated numbers for SQL 2016.
Arvind Shyamsundar: The question is fairly open-ended and the answer is dependent on the workload pattern. On the in-memory OLTP front, we achieved 1.2 million batch requests / second on a Fujitsu Primergy server (4 sockets, 72 cores, 144 logical procs) last October. The Superdome X can go up to 16-sockets and hundreds of cores, but with the form factor beyond 4 sockets comes increased NUMA memory latency. So more sockets does not necessarily translate to more throughput. The recent 10TB TPC-H numbers we released were all on 8-socket Lenovo boxes, and the workload involved is predominantly read-workload
https://blogs.msdn.microsoft.com/sqlcat/2016/10/26/how-bwin-is-using-sql-server-2016-in-memory-oltp-to-achieve-unprecedented-performance-and-scale/
sql server: 1.2m batch requests/sec (30-40 sql statements each batch)
Batch requests / second is the nearest equivalent to compare transactions / second. Statements is not an accurate comparison. Transactions / second is too overloaded / ambiguous because it could mean any of:
Business transactions / second (one business transactions could mean multiple SQL batches)
Batch requests / second (assuming one business transaction == one SQL batches)
Some other number involving interplay between SQL commands and external web services etc.
So from a pure OLTP perspective we prefer to quote batch requests / second in this ‘benchmark’. Proper benchmarks like TPC benchmarks have their own clearly defined unit of meansurement (http://www.tpc.org/tpcc/detail.asp)
Arvind Shyamsundar
OLTP DBMS now called Operational DBMS: http://www.gartner.com/technology/reprints.do?id=1-2RIVJYE&ct=151104&st=sb
Hadoop is kind of FileSystem on which Several Ecosystem can work. Its not a DB.Nosql is a kind of DB, Which having specific property.
The diff between filesystem and database is subtle. Anyway databases store all data in files or in RAM. Also we have "object storages"(like S3), or " key-value data stores"(like Riak), or "data structure stores"(like Redis) and we can treat them as the databases.Hadoop is file system and technology stack including NoSQL solutions(HBase for example). NoSQL is a set of methods or ways of data handling.
Hadoop HDFS + YARN is a file system on steroids... i.e. it is neither a relational DBMS's nor non-relational (NoSql) DBMS's... it is optimized for string processing (large strings in large amounts of data)... Hadoop allows users to interact with the data via SQL (multiple options of SQL dialects) and NoSql (multiple options of procedural languages)... unfortunately, in a sub-optimal performance and functionally restrictive for all non string related processing... that's the reason for all vendors and gurus to be so emphatic about Hadoop costs...
For any real-time processing or analytics, NoSQL would be a better use case, rather than Hadoop. However, there are several factors to keep in mind. NoSQL is better suited for simple data structure (key-value, doc etc), but Hadoop has no inherent structure. Hadoop is better for volume writes and parallel scans, but NoSQL is better for high volume random reads (indexed access) and writes. Finally, it would be important to look at what type of analytics you want to do: statistical (with R), Visualization etc to pick the right store. Sometimes it would mean to have both hadoop and NoSQL
On SQL, you nay not need to define schema, but you still need to convert to key/value or JSON before you can store Hadoop is good for batch processing and you don't want to expose to millions of users
Historically Hadoop ecosystem(hdfs,map reduce,yarn etc) targeted OLAP use cases and No Sql (Cassandra, Couchbase etc) were more towards OLTP work loads. However lines are getting blurred. You gave a good example of Map Reduce on Couchbase. Or Hbase on Hadoop ecosystem targeting real time use cases.
HDFS (Hadoop File System) has been built for large files and is very efficient in batch processing ,supports sequential access of data only , hence no support for random access and fast individual record lookups and data update is not efficient either, while NoSQL database addresses all the these challenges.
To reiterate in short, Hadoop is a computation platform, while NoSQL is an unstructured database.
Hadoop on its most basic constituent is a distributed file system HDFS built to store large volume of string data in parallel with redundancy. But the filesystem by itself is of little use without the rest of the ecosystem like YARN, HBASE, HIVE, etc (and now SPARC for more realtime usage) providing more user friendly usage. HBASE also falls under the noSQL category. NoSQL come in different flavors based on the inherent architecture and use-cases they support.
OLTP DBMS now called Operational DBMS: http://www.gartner.com/technology/reprints.do?id=1-2RIVJYE&ct=151104&st=sb
Hadoop is kind of FileSystem on which Several Ecosystem can work. Its not a DB.Nosql is a kind of DB, Which having specific property.
The diff between filesystem and database is subtle. Anyway databases store all data in files or in RAM. Also we have "object storages"(like S3), or " key-value data stores"(like Riak), or "data structure stores"(like Redis) and we can treat them as the databases.Hadoop is file system and technology stack including NoSQL solutions(HBase for example). NoSQL is a set of methods or ways of data handling.
Hadoop HDFS + YARN is a file system on steroids... i.e. it is neither a relational DBMS's nor non-relational (NoSql) DBMS's... it is optimized for string processing (large strings in large amounts of data)... Hadoop allows users to interact with the data via SQL (multiple options of SQL dialects) and NoSql (multiple options of procedural languages)... unfortunately, in a sub-optimal performance and functionally restrictive for all non string related processing... that's the reason for all vendors and gurus to be so emphatic about Hadoop costs...
For any real-time processing or analytics, NoSQL would be a better use case, rather than Hadoop. However, there are several factors to keep in mind. NoSQL is better suited for simple data structure (key-value, doc etc), but Hadoop has no inherent structure. Hadoop is better for volume writes and parallel scans, but NoSQL is better for high volume random reads (indexed access) and writes. Finally, it would be important to look at what type of analytics you want to do: statistical (with R), Visualization etc to pick the right store. Sometimes it would mean to have both hadoop and NoSQL
On SQL, you nay not need to define schema, but you still need to convert to key/value or JSON before you can store Hadoop is good for batch processing and you don't want to expose to millions of users
Historically Hadoop ecosystem(hdfs,map reduce,yarn etc) targeted OLAP use cases and No Sql (Cassandra, Couchbase etc) were more towards OLTP work loads. However lines are getting blurred. You gave a good example of Map Reduce on Couchbase. Or Hbase on Hadoop ecosystem targeting real time use cases.
HDFS (Hadoop File System) has been built for large files and is very efficient in batch processing ,supports sequential access of data only , hence no support for random access and fast individual record lookups and data update is not efficient either, while NoSQL database addresses all the these challenges.
To reiterate in short, Hadoop is a computation platform, while NoSQL is an unstructured database.
Hadoop on its most basic constituent is a distributed file system HDFS built to store large volume of string data in parallel with redundancy. But the filesystem by itself is of little use without the rest of the ecosystem like YARN, HBASE, HIVE, etc (and now SPARC for more realtime usage) providing more user friendly usage. HBASE also falls under the noSQL category. NoSQL come in different flavors based on the inherent architecture and use-cases they support.
NoSQL: Analogy of building a race car from a regular car…stripping off the parts
scalable because all data within one doc and no need to move data to join tables
Join not a problem for OLTP, but a problem for OLAP
NoSQL: Analogy of building a race car from a regular car…stripping off the parts
scalable because all data within one doc and no need to move data to join tables
Join not a problem for OLTP, but a problem for OLAP
http://www.jamesserra.com/archive/2014/05/hadoop-and-data-warehouses/
Hadoop Common – Contains libraries and utilities needed by other Hadoop modules
Hadoop Distributed File System (HDFS) – A distributed file-system that stores data on commodity machines, providing very high aggregate bandwidth across the cluster
Hadoop MapReduce – A programming model for large scale data processing. It is designed for batch processing. Although the Hadoop framework is implemented in Java, MapReduce applications can be written in other programming languages (R, Python, C# etc). But Java is the most popular
Hadoop YARN – YARN is a resource manager introduced in Hadoop 2 that was created by separating the processing engine and resource management capabilities of MapReduce as it was implemented in Hadoop 1 (see Hadoop 1.0 vs Hadoop 2.0). YARN is often called the operating system of Hadoop because it is responsible for managing and monitoring workloads, maintaining a multi-tenant environment, implementing security controls, and managing high availability features of Hadoop
http://www.jamesserra.com/archive/2014/05/hadoop-and-data-warehouses/
Hadoop Common – Contains libraries and utilities needed by other Hadoop modules
Hadoop Distributed File System (HDFS) – A distributed file-system that stores data on commodity machines, providing very high aggregate bandwidth across the cluster
Hadoop MapReduce – A programming model for large scale data processing. It is designed for batch processing. Although the Hadoop framework is implemented in Java, MapReduce applications can be written in other programming languages (R, Python, C# etc). But Java is the most popular
Hadoop YARN – YARN is a resource manager introduced in Hadoop 2 that was created by separating the processing engine and resource management capabilities of MapReduce as it was implemented in Hadoop 1 (see Hadoop 1.0 vs Hadoop 2.0). YARN is often called the operating system of Hadoop because it is responsible for managing and monitoring workloads, maintaining a multi-tenant environment, implementing security controls, and managing high availability features of Hadoop
https://www.linkedin.com/pulse/hadoop-falling-george-hill
http://www.jamesserra.com/archive/2014/05/hadoop-and-data-warehouses/
Hadoop Common – Contains libraries and utilities needed by other Hadoop modules
Hadoop Distributed File System (HDFS) – A distributed file-system that stores data on commodity machines, providing very high aggregate bandwidth across the cluster
Hadoop MapReduce – A programming model for large scale data processing. It is designed for batch processing. Although the Hadoop framework is implemented in Java, MapReduce applications can be written in other programming languages (R, Python, C# etc). But Java is the most popular
Hadoop YARN – YARN is a resource manager introduced in Hadoop 2 that was created by separating the processing engine and resource management capabilities of MapReduce as it was implemented in Hadoop 1 (see Hadoop 1.0 vs Hadoop 2.0). YARN is often called the operating system of Hadoop because it is responsible for managing and monitoring workloads, maintaining a multi-tenant environment, implementing security controls, and managing high availability features of Hadoop
Am I doing bank transactions or counting web page visits?
In NoSQL, to maintain high availability or for performance reasons, data has multiple copies. These copies will not all be updated instantaneously when there is a data change, but will all eventually be updated (“eventually consistent”)
Use cases: scale, cache, store blob data, shopping cart, session data, leaderboards, queues
See http://www.slideshare.net/harrikauhanen/nosql-3376398
Also called Columnar stores or column stores
Use cases: scale, real-time querying of random (non-sequential) data. Web analytics, time series analytics, huge number of writes, big data storage. Like document stores except data is stored on nodes
Use cases: social network, master data management, network and IT operations, real-time recommendations, fraud detection, identity and access management, graph-based search, web browsing, portfolio analytics, gene sequencing, class curriculum
MongoDB vs Cassandra: http://theprofessionalspoint.blogspot.com/2014/01/mongodb-vs-cassandra-difference-and.html:
Cassandra is much better suited for highly distributed applications due to its tunable replication engine. It was built from the ground up to be a shared-nothing data engine. MongoDB, by contrast, is better suited for applications that need a dynamic schema-less approach.
https://www.youtube.com/watch?v=PENcqjVKqr4c
https://www.youtube.com/watch?v=gJFG04Sy6NY
http://maxivak.com/differences-between-nosql-databases-cassandra-vs-mongodb-vs-couchdb-vs-redis-vs-riak-vs-hbase-vs-membase-vs-neo4j/
http://www.infoworld.com/article/2848722/nosql/mongodb-cassandra-hbase-three-nosql-databases-to-watch.html
MongoDB vs Cassandra: http://theprofessionalspoint.blogspot.com/2014/01/mongodb-vs-cassandra-difference-and.html:
Cassandra is much better suited for highly distributed applications due to its tunable replication engine. It was built from the ground up to be a shared-nothing data engine. MongoDB, by contrast, is better suited for applications that need a dynamic schema-less approach.
https://www.youtube.com/watch?v=PENcqjVKqr4c
https://www.youtube.com/watch?v=gJFG04Sy6NY
http://maxivak.com/differences-between-nosql-databases-cassandra-vs-mongodb-vs-couchdb-vs-redis-vs-riak-vs-hbase-vs-membase-vs-neo4j/
http://www.infoworld.com/article/2848722/nosql/mongodb-cassandra-hbase-three-nosql-databases-to-watch.html
http://db.cs.cmu.edu/papers/2016/pavlo-newsql-sigmodrec2016.pdf
Use cases: scale,
A class of modern RDBMS’s that seek to provide the same scalable performance of NoSQL systems for OLTP read-write workloads while still maintaining the ACID guarantees of a traditional relational database system. The disadvantages is they are not for OLAP-style queries, and they are inappropriate for databases over a few terabytes. Aims to blend NoSQL and Relational/SQL. VoltDB, NuoDB, MemSQL, SAP HANA, Splice Machine, Clustrix, Altibase
If you would rather go the route of using Hadoop software, many of the above technologies have Hadoop or open source equivalents: AtScale and Apache Kylin create SSAS-like OLAP cubes on Hadoop, Jethro Data creates indexes on Hadoop data, Apache Atlas for metadata and lineage tools, Apache Drill to query Hadoop files via SQL, Apache Mahout or Spark MLib for machine learning, Apache Flink for distributed stream and batch data processing, Apache HBase for storing non-relational streaming data and supporting fast query response times, SQLite/MySQL/PostgreSQL for storing relational data, Apache Kafka for event queuing, Apache Falcon for data and pipeline management (ETL), and Apache Knox for authentication and authorization.
https://codahale.com/you-cant-sacrifice-partition-tolerance/
Emails don’t need to be consistent, stock prices do
http://www.3pillarglobal.com/insights/short-history-databases-rdbms-nosql-beyond
In NoSQL, to maintain high availability or for performance reasons, data has multiple copies. These copies will not all be updated instantaneously when there is a data change, but will all eventually be updated (“eventually consistent”)
https://www.infoq.com/news/2014/04/bitcoin-banking-mongodb
https://azure.microsoft.com/en-us/blog/json-functionalities-in-azure-sql-database-public-preview/ “If you need a specialized JSON database in order to take advantage of automatic indexing of JSON fields, tunable consistency levels for globally distributed data, and JavaScript integration, you may want to choose Azure DocumentDB as a storage engine.”
https://blogs.msdn.microsoft.com/jocapc/2015/05/16/json-support-in-sql-server-2016/
https://msdn.microsoft.com/en-us/library/dn921897.aspx “If you have pure JSON workloads where you want to use some query language that is customized and dedicated for processing of JSON documents, you might consider Microsoft Azure DocumentDB.”
http://demo.sqlmag.com/scaling-success-sql-server-2016/integrating-big-data-and-sql-server-2016
https://www.simple-talk.com/sql/learn-sql-server/json-support-in-sql-server-2016/
So now that you’re convinced of the benefits of PaaS, let’s take a look at the menu of available PaaS data services on Azure. It’s important to remember that with any application, you can use multiple data stores.
Cache and Search are specialized data stores that you wouldn’t use as a primary data store, but they are worth mentioning here.
Note: speaker should do a brief verbal overview of the information contained in this chart.
Presenter guidance:
Introduce the family portrait.
Slide talk track:
This is how we think about the core differences across the data services for capturing and managing data
On the left, you have more database imposed structure on the left and this loosens as you move to the right, ending in blobs which is just large containers of binary data.
Presenter guidance:
At this point, let’s take a slight detour to mention SQL Server in a VM and how it fits into the mix. It’s important in the context of dev/test and lift and shift (or migrating existing apps).
Establish app dev scenarios as common ground.
Slide talk track:
Let’s first orient ourselves in what we see as common application scenarios.
Are you seeing these?
Are you interested in these scenarios?
Do these represent scenarios you would be willing to move to the cloud?
The services listed are generally those that we would see you using for these scenarios, but this is just what we see. There are infinite ways to do things and at the end of the day, it’s your decision. Azure is there to make sure you have all of the options and choices available that you need.
https://msdn.microsoft.com/en-us/library/mt143171.aspx
When it comes to key BI investments we are making it much easier to manage relational and non-relational data with Polybase technology that allows you to query Hadoop data and SQL Server relational data through single T-SQL query. One of the challenges we see with Hadoop is there are not enough people out there with Hadoop and Map Reduce skillset and this technology simplifies the skillset needed to manage Hadoop data. This can also work across your on-premises environment or SQL Server running in Azure.
Clock- 47 Minutes
In this scenario based HOL, you will learn how to build a ‘polyglot persistence’ data pattern that is common in modern cloud hosted applications. Requirements of modern applications, such as, greater scale and availability, have driven the industry to begin using a much broader range of technologies for storing data within an application. Microsoft Azure provides a range of storage technologies that support these architectures and this HOL provides an example of the use of these in the well understood scenario of e-Commerce. With data services in Microsoft Azure, you can quickly design, deploy, and manage highly-available apps that scale without downtime and that enable you to rapidly respond to security threats. Features built into services like Azure SQL Database, Azure Search, and Azure DocumentDB help your apps scale smartly, run faster, and stay available and secure.
In this HOL you will see a browser based e-commerce application running under the LCA approved sample company name ‘AdventureWorks’. It has been created to demonstrate functionality provided by the following data storage technologies (SQL Database, DocumentDB, Search, Table Storage). In a real application, decisions will need to be made as to where data is stored. In this HOL we wish to highlight; how using multiple Azure data service technologies allows you to take a modern approach to data in your applications.
Note- The website (which gets built out of this HOL) is not intended to be a fully functioning site. It is not designed to be a reference e-commerce implementation nor a starting point for a customer’s implementation of an e-commerce site on Azure, rather it will provide the following functionality in order to demonstrate the selected storage technologies.
In the course of this lab, you will gain greater familiarity with Azure SQL Database, DocumentDB, Azure Search and Table Storage through performing the following tasks:
Familiarize yourself with one of the tenant-company’s websites and its Azure SQL Database backend.
Create a new database using the Azure portal.
Configure and implement vertical scaling by increasing the capacity of a database.
Use Azure SQL Database auditing features to track down an erroneous deletion from a database.
Use Azure SQL Database point-in-time restore to correct the deletion (Optional)
Configure and implement Azure SQL Database geographic disaster recovery to prevent large-scale data loss.
Locate data using Azure Search.
Modernize and create an iterative experience using DocumentDB.
http://INMMDDYYYY.azurewebsites.net/
Show a couple of examples of using multiple data services.
Show a couple of examples of using multiple data services.
1) Copy source data into the Azure Data Lake Store (twitter data example)2) Massage/filter the data using Hadoop (or skip using Hadoop and use stored procedures in SQL DW/DB to massage data after step #5)3) Pass data into Azure ML to build models using Hive query (or pass in directly from Blob Storage). You can use a Python package to pull data directly from the Azure Data Lake Store4) Azure ML feeds prediction results into the data warehouse (you can also pull in data from SQL Database or SQL Data Warehouse)5) Non-relational data in Azure Data Lake Store copied to data warehouse in relational format (optionally use PolyBase with external tables to avoid copying data)6) Power BI pulls data from data warehouse to build dashboards and reports7) Azure Data Catalog captures metadata from Azure Data Lake Store, SQL DW/DB, and SSAS cubes8) Power BI can pull data from the Azure Data Lake Store via HDInsight/Spark (beta) or directly. Excel can pull data from the Azure Data Lake Store via Hive ODBC or PowerQuery/HDInsight9) To support high concurrency if using SQL DW, or for easier end-user data layer, create an SSAS cube