SQL vs. NoSQL

2014 © Trivadis
BASEL BERN BRUGG LAUSANNE ZÜRICH DÜSSELDORF FRANKFURT A.M. FREIBURG I.BR. HAMBURG MUNICH STUTTGART VIENNA
2014 © Trivadis
SQL versus NoSQL
Guido Schmutz
Philipp Salvisberg
9th December 2014
SQL versus NoSQL
1

2014 © Trivadis
Guido Schmutz
 Working for Trivadis for more than 17 years
 Oracle ACE Director for Fusion Middleware and SOA
 Co-Author of different books
 Consultant, Trainer Software Architect for Java, Oracle, SOA and
Big Data / Fast Data
 Member of Trivadis Architecture Board
 Technology Manager @ Trivadis
 More than 25 years of software development
experience
 Contact: guido.schmutz@trivadis.com
 Blog: http://guidoschmutz.wordpress.com
 Twitter: gschmutz
9th December 2014
SQL versus NoSQL
2

2014 © Trivadis
Philipp Salvisberg
 With Trivadis since April 2000
 Senior Principal Consultant, Partner
 Member of the Board of Directors
 philipp.salvisberg@trivadis.com
 www.salvis.com/blog
 @phsalvisberg
 Main focus on database centric
development with Oracle database
 Application Development
 Business Intelligence
 Application Performance Management
 Over 20 years experience in using Oracle products
9th December 2014
SQL versus NoSQL
3

2014 © Trivadis
Agenda
1. Motivation
2. Overview of SQL and NoSQL Data Stores
3. Use Cases – Let's Get Ready to Rumble
4. Core Messages
9th December 2014
SQL versus NoSQL
4

2014 © Trivadis
Understanding the Merits
9th December 2014
SQL versus NoSQL
5
Why NoSQL?
Can do everything with RDBMSs!
Having performance, scalability,
transactions and SQL.
A key-value store is adequate for key
lookups and easier to understand than
a RDBMS.
RDBMSs do not scale as good as
NoSQL systems like Google’s
BigTable.
RDBMSs provide a common interface
with SQL, transactions, and relational
schema.
Some applications require a flexible
schema.
Adding new attributes at runtime in
RDBMSs is typically not possible.…
…
Source: Scalable SQL and NoSQL Data Stores, Rick Catell, 2010

2014 © Trivadis
Agenda
1. Motivation
4. Conclusion
9th December 2014
SQL versus NoSQL
6

2014 © Trivadis
SQL Data Stores
 Relational Model
 Standardized, SQL:2011 is the 7th major revision since SQL-86
 9 parts, more than 4000 pages
 But no single database implements all standards/features
 Rich set of features
 Incl. SQL/PSM, SQL/MED, SQL/XML, SQL/RPR, Temporal Features
 Incl. User-defined Types and Collection Types (since SQL:1999)
 ACID Transactions
 Atomicity: all or nothing
 Consistency: from valid state to valid state considering constraints, triggers, …
 Isolation: result is not affected through concurrent execution
 Durability: committed data stays available after crash, power loss or errors
 Good support by different languages, frameworks and tools
 Good understanding of basic concepts by IT professionals
9th December 2014
SQL versus NoSQL
7

2014 © Trivadis
NoSQL Definition
 Next Generation Databases mostly addressing some of the points:
 being non-relational,
 distributed,
 open-source and
 horizontally scalable.
 Often more characteristics apply such as:
 schema-free,
 easy replication support,
 simple API,
 eventually consistent / BASE (not ACID),
 a huge amount of data
 and more.
 The misleading term "nosql" (the community now translates it mostly with
"not only sql") should be seen as an alias to something like the definition
above
9th December 2014
SQL versus NoSQL
8
Source: http://nosql-database.org
BASE
 Basically Available: Availability is
more important than consistency
 Soft State: Higher availability results
in an eventual consistent state
 Eventually Consistent: If no new
updates are made to a given data
item, eventually all accesses to that
item will return the last updated
value

2014 © Trivadis
Brewer's CAP Theorem
Any networked shared-data system can have at most two of the three
desirable properties:
 Consistency
All of the nodes see the same data at
the same time, regardless of
where the data is stored
 Availability
Node failures do not prevent
survivors from continuing to
operate
 Network Partition tolerance
The system continues to operate
despite arbitrary message loss
9th December 2014
SQL versus NoSQL
9
Availability
Consistency
Network
Partition
Tolerance
n/a
CA CP
AP

2014 © Trivadis
Data Store Positioning
9th December 2014
SQL versus NoSQL
10
Scalability
Standardized Model, Tooling, Complexity
Key-value
Wide Column (Column Families / Extensible Records)
Document
Graph
Relational
SQL Comfort Zone
Multi Dimensional

2014 © Trivadis
Agenda
1. Motivation
4. Core Messages
9th December 2014
SQL versus NoSQL
11

2014 © Trivadis
9th December 2014
SQL versus NoSQL
12
Round 1
Smart Meter

2014 © Trivadis
Smart Meter – Customer Dashboard
9th December 2014
SQL versus NoSQL
13

2014 © Trivadis
Smart Meter – Use Cases
 Store sensor and its
sub-sensor values
 2 Mio sensors, up to 10 sub-sensors
 Energy consumption per second per sensor (kWh)
 Delivery interval between 1 second and 5 minutes
 Query usage per sensor and its
sub-sensors to visualize a
time series on a customer dashboard
 Available in different granularities, values are aggregated in
- Minute
- Quarter of hour (15-minutes)
- Hour
- Day
 Responsive UI
9th December 2014
SQL versus NoSQL
14
Smart Meter
Store data of
millions of
sensors
Household
Dashboard

2014 © Trivadis
Cassandra NoSQL Datastore
• Wide-Column Store
• Developed at Facebook
• Professional grade support from DataStax
• Main Features
 Real-Time
 Highly Distributed
 Support for Multiple Data Center
 Highly Scalable
 No Single Point of Failure
 Fault Tolerant
 Tunable Consistency
 CQL – Cassandra Query
Language
9th December 2014
SQL versus NoSQL
15

2014 © Trivadis
The Wide Column Store Way (Cassandra)
9th December 2014
SQL versus NoSQL
16
Household Bucket
AFG10 MINUTE-2014/03/5 sensor 1 1 1 … 2 2 …
at 11:59 11:58 11:57 … 11:59 11:58 …
kwh 7.05 7.10 8.11 … 6.95 7.04 …
AFG10 QHOUR-2014/03 sensor 1 1 1 … 2 2 …
at 5T11:4
5
5T11:3
0
5T11:1
5
… 5T11:4
5
5T11:3
0
…
kwh 105.78 104.73 102.29 … 102.78 101.61 …
AFG10 HOUR-2014/03 sensor 1 1 1 … 2 2 …
at 5T11 5T10 5T09 … 5T11 5T10 …
kwh 423.00 410.33 395.99 … 598.32 522.12 …
AFG10 DAY-2014 sensor 1 1 1 … 2 2 …
at 5T 3T 2T … 5T 4T …
kwh 10100.2 9892.2 8987.4 … 879.8 912,3 …
GXK11 MINUTE-2014/03/5 sensor 1 1 1 … 2 2 …
at 11:59 11:03 11:04 … 11:01 11:02 …
kwh 100.10 90.88 95.00 … 92.50 88.50 …
Growth
24h * 60m * 11 sensor = 15’840 cols
30d * 24h * 4q * 11 sensor = 31’680 cols
30d * 24h * 11 sensor = 7’920 cols
365d * 11 sensor = 4’011 cols

2014 © Trivadis
The Cassandra Way
9th December 2014
SQL versus NoSQL
17
 288 nodes on EC2
 Over 1 Mio writes/sec => 60Mio writes/min
 Rolling counters, always up to date

2014 © Trivadis
Relational Architecture
 Active Data Guard Configuration
 Global Data Services redirects
requests based on
 Server loads
 Request type (read/write)
 Reader farm is geographically
spread
 Failover/switchover to any node in
the reader farm is possible
 Read services are not affected
 Write services are unavailable for a
short period of time
 Scalability of the write services is
the bottleneck of the system
9th December 2014
SQL versus NoSQL
18
Global Data Services
Client
Read/Write Read-only Read-only Read-only
Sensor Sensor

2014 © Trivadis
Relational Data Model
 SENSOR_READINGS_...
 Index-organized tables
 Daily partitions
 JDBC Batch Merges
 A transaction per sensor delivery
 A single network roundtrip to merge 55
readings of a sensor delivery
 Average between
- 0.4 Mio tpm (delivery per 5 minutes)
- 120 Mio tpm (delivery per second)
 Top TPC-C Benchmark: 8.5 Mio tpm
 Batch job to aggregate readings every
15 minutes, avoiding intermediate
results (updates)
 Quarter of hour (5760 times a day)
 Hour (24 times a day)
 Day (once a day)
9th December 2014
SQL versus NoSQL
19

2014 © Trivadis
Query Sensor Data – The SQL Way
MERGE INTO sensor_readings_minute t
USING (SELECT sensor_id, TRUNC(start_time, 'MI') AS start_time,
usage_kwh FROM TABLE(:p_sensor_data_list)) s
ON (t.sensor_id = s.sensor_id AND t.start_time = s.start_time)
WHEN NOT MATCHED THEN
INSERT (t.sensor_id, t.start_time, t.usage_kwh)
VALUES (s.sensor_id, s.start_time, s.usage_kwh)
WHEN MATCHED THEN
UPDATE SET t.usage_kwh = t.usage_kwh + s.usage_kwh;
9th December 2014
SQL versus NoSQL
20
SELECT sensor_id, start_time, usage_kwh
FROM sensor_readings_minute
WHERE sensor_id = :p_sensor_id
AND start_time BETWEEN :p_from AND :p_to
ORDER BY sensor_id, start_time;
Use aggregate tables to change granularity (quarter of hours, hours, days)

2014 © Trivadis
9th December 2014
SQL versus NoSQL
21
Smart Meter
0 – 1

2014 © Trivadis
9th December 2014
SQL versus NoSQL
22
Round 2
Order Entry

2014 © Trivadis
Order Entry – Example
9th December 2014
SQL versus NoSQL
23

2014 © Trivadis
Order Entry – Use Cases
 Update the quantity in stock of
all ordered products
 When order status changes
from "incomplete" to "complete"
 When order status changes
from "complete" to "cancelled"
 Ensure that the quantity in stock is always correct (no lost updates or similar)
 Create a report for the 5 top-selling products for a year
9th December 2014
SQL versus NoSQL
24
Order Entry
Update
quantity in
stock
Top 5
selling
products

2014 © Trivadis
Relational Model
9th December 2014
SQL versus NoSQL
25

2014 © Trivadis
Change Quantity in Stock – The SQL Way
Single Transaction
UPDATE ORDERS
SET order_status = :p_value_for_complete
WHERE order_id = :p_order_id;
MERGE INTO PRODUCTS t
USING (SELECT product_id,
SUM(quantity) AS quantity
FROM order_items
WHERE order_id = :p_order_id
GROUP BY product_id) s
ON (t.product_id = s.product_id)
WHEN MATCHED THEN
UPDATE SET t.quantity_on_stock =
t.quantity_on_stock - s.quantity;
COMMIT;
9th December 2014
SQL versus NoSQL
26

2014 © Trivadis
5 Top-Selling Products – The SQL Way
SELECT p.name AS product_name,
SUM(i.quantity * i.unit_price) AS sales_volume
FROM order_items i
INNER JOIN orders o
ON o.order_id = i.order_id
INNER JOIN products p
ON p.product_id = i.product_id
WHERE o.order_date > DATE '2013-01-01'
AND o.order_date <= DATE '2014-01-01'
AND o.order_status = :p_value_for_complete
GROUP BY p. name
ORDER BY 2 DESC
FETCH FIRST 5 ROWS WITH TIES;
9th December 2014
SQL versus NoSQL
27

2014 © Trivadis
Customer
MongoDB Document Data Model (Aggregate Pattern)
9th December 2014
SQL versus NoSQL
28
Order
Order Items
Order Item 1
Order Item 2
Order Item n
Shipping Address
Billing Address
Product
Category
Suppliers
Supplier 1
Address
Supplier n
Address
Supplier 1
Address
Address

2014 © Trivadis
Update Quantity in Stock – The MongoDB Way
ForEach orderItems.item {
db.products.update( { productId : 101 },
{ $inc : { quantity: -10 } },
{ multi: false }
);
}
9th December 2014
SQL versus NoSQL
29
db.orders.find ( { orderId: 1} );
db.orders.update( { orderId: 1},
{ $set : { orderStatus: "COMPLETE" } },
{ multi: false } );
Transaction 1
Read Operation
Transaction 2 .. n

2014 © Trivadis
5 Top-Selling Products – The MongoDB Way
db.orders.aggregate([
{ $match : {
orderStatus: "COMPLETE",
orderDate: { $gt: ISODate("2014-01-01"),
$lt: ISODate("2014-04-01") }
} },
{ $unwind : "$orderItems" },
{ $project : { _id: 0,
productId: "$orderItems.productId",
total : { $multiply : ["$orderItems.quantity",
"$orderItems.unitPrice”] }
} },
{ $group : { _id: "$productId",
total : { $sum : "$total"} } },
{ $sort : { total: -1 }},
{ $limit : 5 }
])
9th December 2014
SQL versus NoSQL
30

2014 © Trivadis
Spotify – Use Cases
 Playlist, Showing Ads, Following Artists …
are all uses cases which have to be
highly available, and accessible worldwide
 Needs to be distributed to be fast
 Service should be available even if a partition happen (due to network
failure/machine failure)
 First time subscription and subscription renewal must be absolutely
consistent
 Customer should only pay once!
9th December 2014
SQL versus NoSQL
34
Spotify
Renewals
(payments)
Globally
available
playlists

2014 © Trivadis
Polyglot Persistence – SQL And NoSQL
9th December 2014
SQL versus NoSQL
35
Spotify Platform
Playlists, Ads,
Followings
Subscriptions
and Payments
Wide-column
store RDBMS

2014 © Trivadis
Core Messages
 We will see a major consolation
in the NoSQL area
 SQL is and stays important
 Polyglot persistence will be part
of every solution design in the
near future
 Enterprise capabilities are
required
 Tooling (monitoring, backup &
recovery, data security, …)
 Organization, skills
 Opportunity for cloud based
solutions
9th December 2014
SQL versus NoSQL
38

2014 © Trivadis
Questions and answers ...
2014 © Trivadis
BASEL BERN BRUGES LAUSANNE ZÜRICH DÜSSELDORF FRANKFURT A.M. FREIBURG I.BR. HAMBURG MUNICH STUTTGART VIENNA
Guido Schmutz
Technology Manager
Philipp Salvisberg
Senior Principal Consultant
9th December 2014
SQL versus NoSQL
39

2014 © Trivadis
Further information ...
9th December 2014
SQL versus NoSQL
40
 http://martinfowler.com/books/nosql.html
 http://www.manning.com/mccreary/
 http://highlyscalable.wordpress.com
 http://nosql-database.org
 http://db-engines.com/

SQL vs. NoSQL

More Related Content

What's hot (20)

Viewers also liked (20)

Similar to SQL vs. NoSQL (20)

More from Guido Schmutz (20)

Recently uploaded (20)

SQL vs. NoSQL

Editor's Notes