Sql server 2012_parallel_data_warehouse_breakthrough_platform_white_paper

Microsoft SQL
Server 2012
Parallel Data
Warehouse
A Breakthrough
Platform for NextGeneration Data
Warehousing and Big
Data Solutions

Contents
4

Summary Statement

4

Introduction

6

PDW’s Capabilities
6

Scalable, Fast, and Reliable

7

Easy to Use and Manage

7
Platform for Data Mining and
Analysis
8

What Customers Are Saying

10
Purpose-Built for Data Warehousing
Workloads
10

Designed for Parallel Processing

11
Hardware and Software
Designed Together
13
Scalable Solution for Today’s Data
Challenges
13

Easy to Add Capacity

15

Easy to Load Data Fast

17

Easy to Retain More Data

18

Easy to Consolidate Data

Microsoft SQL Server 2012 Parallel Data Warehouse

19

Why Queries Run Fast
19

Queries Run on Distributed Data

20

Queries Are Highly Parallelized

22
Data Movement Service
Transfers Data Fast
22
In-Memory Clustered
Columnstore Indexes Improve
23

Why You Can Trust PDW

24

PDW’s Hardware Redundancy

24

High Availability

25

PDW is Easy to Use and Manage

27

Platform for Data Mining and Analysis
28
PolyBase for Hadoop
Integration
30

30

© 2013 Microsoft Corporation. All rights
reserved. This document is provided “asis.” Information and views expressed in
this document, including URL and other
Internet Web site references, may change
without notice. You bear the risk of using
it. This document does not provide you
with any legal rights to any intellectual
property in any Microsoft product. You
may copy and use this document for your
internal, reference purposes. You may
modify this document for your internal,
reference purposes. Published October
22, 2013.

Business Intelligence Integration

Summary


Summary
Statement

This paper explains PDW’s thoughtful design so that you can learn
more about how PDW achieves breakthrough results, and how
upgrading to SQL Server 2012 PDW will benefit your business.
PDW’s benefits are so remarkable that you might think it all sounds
too good to be true! However, PDW’s benefits are real and the
result of a well-engineered system that is purpose-built for data
warehousing workloads, including the growing business needs for
data warehousing to integrate the analysis of Hadoop and
relational data.
Microsoft’s® SQL Server® 2012 Parallel Data Warehouse (PDW) is a
next-generation platform built to run your data analysis fast and to scale
storage from a few terabytes to over 6 petabytes in a single appliance.
PDW ships to your data center as an appliance with hardware and
software pre-installed and pre-configured for maximum performance.
With PDW’s Massively Parallel Processing (MPP) design, queries run in
minutes instead of hours, and in seconds instead of minutes in
comparison to today’s Symmetric Multi-Processing (SMP) databases.
PDW is not only fast and scalable, it is designed with high redundancy
and high availability, making it a reliable platform you can trust with
your most business critical data. PDW is designed for simplicity which
makes it easy to learn and to manage. PDW’s PolyBase technology for
analyzing Hadoop data, and its deep integration with Business
Intelligence tools make it a comprehensive platform for building end-toend solutions.

Introduction

If your business relies on data, you know that it is a constant challenge
to store, manage, and analyze it effectively as data continues to
proliferate. It’s also expensive to keep enough data on “hot” storage
where it is readily available for analysis. Even when you have the data
you need on hot storage, it can take hours or even days to run analysis
and reports on today’s symmetric multi-processing (SMP) systems. To
add more to the challenges, businesses today are struggling to figure
out how to add the value of non-relational Hadoop data into their
analysis.
As a result, business analysts are held back from making faster and more
accurate data-driven business decisions that are needed to compete in
today’s marketplace. This is the modern data challenge.
If you can relate to today’s data dilemma, consider upgrading to SQL
Server 2012 Parallel Data Warehouse (PDW), Microsoft’s next generation
platform for data warehousing and Big Data integration. With PDW’s
massively parallel processing (MPP) design, queries commonly complete
50 times faster than traditional data warehouses built on symmetric
multi-processing (SMP) database management systems.


4

50 times faster means that queries
complete in minutes instead of hours,
or seconds instead of minutes. With
this breakthrough performance, your
business analysts can generate more
extensive results faster, and can easily
perform ad-hoc queries or drill down
into the details. As a result, your
business can make better decisions,
faster.
In addition to achieving breakthrough query performance, PDW makes it
easy to:









Grow your data warehouse to anywhere from a few terabytes to over
6 petabytes of data in a single appliance by adding “scale units” to
your existing system,
Trust your data will be there when you need it because of the builtin high redundancy and high availability,
Solve modern data challenges of loading and consolidating data,
Integrate Hadoop data with relational data for fast analysis by using
PDW’s highly parallelized PolyBase technology,
Use Business Intelligence tools to build comprehensive end-to-end
solutions.

How is this possible? Would you like to understand how PDW achieves
such remarkable results? This paper details how PDW performs in realworld situations and then explains how PDW’s design is able to achieve
these outcomes through a well-engineered data platform that is
purpose-built for data warehousing workloads and Big Data integration.


5

PDW’s
Capabilities

PDW is more than a product, it’s a data platform for next-generation
data warehousing and Big Data solutions.

Scalable,
Fast, and
Reliable

Scalable
PDW achieves breakthrough scalability to expand storage from a few
terabytes to over 6 petabytes in a single appliance. Unlike SMP systems
that require purchasing a new server and storage in order to add more
processing and storage capacity, PDW expands by adding “scale units”
to the existing appliance. After adding capacity, PDW does all the work
of redistributing data and balancing I/O to the additional scale units.

The following benefits of PDW are all the result of a well-engineered
system that is designed specifically for data warehousing workloads.

Fast
PDW’s Massively Parallel Processing (MPP) design can load data in realtime and finish complex queries up to 50 times faster than today’s
Symmetric Multi-Processing (SMP) systems. Queries complete in
seconds instead of minutes, and minutes instead of hours compared to
today’s SMP databases such as SQL Server 2008 R2, SQL Server 2012,
Oracle, or DB2. Some queries that are impossible to run on today’s SMP
systems can complete on PDW. Companies no longer need to settle for
canned reports and summarized results in exchange for quicker
response times. Ad-hoc analysis and detailed reporting are possible and
on a sustained basis.
Reliable
PDW is a platform you can trust to keep your data safe and available. It
is well-engineered with high-redundancy and high availability built-in
across hardware and software components. User data is always stored
and managed by SQL Server 2012, which has an impeccable reputation
for keeping data safe.


6

Easy to Use
and Manage

PDW is designed for simplicity. The complexity is already engineered
into the appliance so that you don’t have to handle the details. For
example, PDW handles all the details of distributing the data across the
appliance nodes, performs all the extra steps required to process queries
in parallel, and manages the low-level hardware and software
configuration settings.
PDW arrives at your data center pre-configured and pre-tested; simply
plug PDW in and configure it for your networking environment. You can
start using it the same day it arrives!
As a result, the learning curve is minimal. You can easily retrain your staff
without hiring new PDW experts. Your SQL Server DBAs can easily
transfer their SQL Server knowledge to PDW.
PDW Simplicity. No need to manage…



storage space



parallel queries



distributed data



software configuration



Platform for
Data Mining
and Analysis

disk or database subsystems

hardware configuration

Integrated with Hadoop
PDW’s PolyBase technology enables you to query Hadoop data and
merge it with your relational data by using Transact-SQL and without
learning a host of new skills. By using PolyBase’s Transact-SQL interface
you can do more in-depth data mining, reporting, and analysis without
acquiring the skills to run MapReduce queries in Hadoop. For example,
queries can combine Hadoop and PDW data in a single step, Hadoop
data can be stored as relational data in PDW, and query results can be
stored back to Hadoop.
PDW’s PolyBase is an easy way to use the power of PDW’s MPP
architecture to analyze Hadoop data fast. PolyBase gives you the
flexibility to structure the Hadoop data you need, when you need it, as
you bring it into PDW for analysis. With PolyBase, queries that run too
slow in Hadoop can now run quickly in PDW. This opens up a new world
of data integration and analysis possibilities.


7

Integrated with Business Intelligence Tools
PDW’s deep integration with Business Intelligence (BI) tools makes PDW
a comprehensive platform for building end-to-end data mining and
analysis solutions. PDW integrates with the Microsoft BI Stack including
Reporting Services, Analysis Services, PowerPivot for Excel, and
PowerView. But, PDW also integrates with a growing list of leading nonMicrosoft BI platforms, such as Business Objects, Cognos, SAP Data
Integrator, Tableau, MicroStrategy, QlikView, Oracle Business
Intelligence, and TIBCO Spotfire.

What
Customers
Are Saying

PDW customers are experiencing fantastic results. The following results
from our customers are similar to what most people are saying about
PDW.
Queries Are Fast
 Our queries are completing 76 times faster on PDW. This was after
PDW compressed 1.5 TB to 134 GB.








Our queries are completing in less than 1.5 seconds on a month of
data. In comparison, using SQL Server 2008 R2, the same queries
completed in 2-3 minutes on 7 days of data.
Our queries run 25 times faster on PDW. The queries complete with
an average of 23 seconds per query on PDW. On our SMP system,
the queries completed with an average of 5 minutes and 36 seconds.
Our queries complete 18 times faster on PDW. The queries complete
with an average of 53 seconds per query on PDW. In comparison, on
our SMP system, the queries complete with an average of 16
minutes per query.
With a heavy load on our PDW, our individual query times only
slowed down by 16 percent.

Scales Past SMP
 PDW made it possible for us to complete queries that could not
complete our current system. Even though we ran our queries on a
DL980 with 40 cores, 2 terabytes of RAM, and a sizeable storage
array, the queries could not finish without running out of memory.
After loading our data onto PDW, we ran queries on 7 years of data
with 200 users on the system. These queries finished with an average
time of 689 seconds per query.


We had 40 databases sharded among 10 servers because of SMP
performance limitations. With PDW, we now have all of the
databases consolidated into one database. The data is much easier
to maintain and load. Even better, we are able to run aggregation
queries on a single database.


8

Loading Is Fast, from hours to minutes
 We must load one hour of data in less than an hour. One hour of
data contains 6 billion records and we are able to load this onto
PDW in less than an hour.






Our daily load now takes 5.5 minutes on PDW whereas it used to
take 2.5 hours without PDW. The load is 27 times faster on PDW.
We can load 520 million rows into PDW in 13 minutes, whereas our
old system took 16 hours to load the same data.
PDW performs well with our mixed workload; our users continue to
run their queries while PDW loads occur in the background.

High Data Compression Rates
By using in-memory clustered columnstore indexes to store data on
disk, PDW achieves high data compression rates that save storage costs
and improve query performance. These are typical examples of customer
compression rates.


1.5 TB (uncompressed) compressed to 134 GB (7x compression).



5.5 TB (uncompressed) compressed to 400 GB (14x compression).



120 GB (uncompressed) compressed to 2 GB (60x compression).

Customer Reference
In a case study
(http://www.microsoft.com/casestudies/Case_Study_Detail.aspx?CaseStu
dyID=710000002669), the sales and marketing service firm,
CROSSMARK, needed faster and more detailed insight into terabytes of
information about product supply and demand. With PDW they
achieved this and as a result, CROSSMARK was able to improve service,
create reports at least 50 percent faster, improve employee efficiency,
and save money.


9

PurposeBuilt for
Data Warehousing
Workload

“We started using PDW the same day
it arrived at our data center. The
appliance arrived with the hardware
already configured and the software
already installed.”
PDW is an appliance solution designed specifically for data warehousing
workloads. From the ground up, the hardware and software were
carefully designed together to achieve high scalability, performance,
reliability, and high availability. Whereas, SQL Server 2012 works great
for online transaction processing (OLTP) workloads that update
thousands or millions of records on a daily basis, PDW is designed to be
great at loading, storing, and analyzing the OLTP or Hadoop data that
your business generates on a daily basis.

Designed for
Parallel
Processing

Instead of simply extending the SQL Server 2012 Symmetric MultiProcessing (SMP) design, PDW uses a Massively Parallel Processing
(MPP) design to achieve breakthrough query results and other
performance capabilities discussed in this paper.

Figure 1: SQL Server 2012 PDW uses an MPP design to achieve high performance and scalability.


10

Query processing is highly parallelized. User data is distributed across
processing and storage units called Compute nodes. Each Compute
node has its own direct attached storage, processors, and memory that
run as an independent processing unit. The Control node is the brains of
PDW and figures out how to run each user query in parallel across all of
the Compute nodes. As a result, queries run fast!
SQL Server 2012 is foundational to PDW and runs on each Compute
node to store, manage, and run queries on distributed user data. PDW
uses a special build of SQL Server 2012 that uses updateable in-memory
clustered columnstore indexes for high compression rates and fast
performance on the individual Compute nodes. Plus, by using SQL
Server 2012, PDW has the backing of SQL Server 2012’s impeccable
reputation to keep your data safe.

Hardware
and
Software
Designed
Together

PDW is an appliance solution with hardware and software designed
together to achieve maximum performance and scalability. PDW arrives
at your data center with the software and hardware already installed,
configured, and tested. No tuning is required because the appliance is
already built and tuned to balance CPU, memory, I/O, storage, network,
and other resources. Simply configure PDW for your own network
environment, and it is ready to use.
The first rack of the PDW appliance is called the base rack. Every
appliance has at least one base rack with 2 or 3 Compute nodes,
depending on vendor hardware.
As your business requirements change, you can expand PDW by adding
scale units to the base rack. When the base rack is full, PDW expands by
adding additional racks, called expansion racks, and adding scale units
to them.
The base rack has two InfiniBand and two Ethernet switches for
redundant network connectivity. A dedicated server runs the Control
node and the Management node. A spare server ships in the rack for
failover clustering. Optionally, you can add a second spare server.
The number of Compute nodes in the base rack varies according to the
hardware vendor. For example, HP has 2 Compute nodes per scale unit
and Dell has 3.
Pictured below is an HP base rack plus 3 scale units for a total of 8
Compute nodes. Dell’s base rack scales to 9 Compute nodes.


11

Figure 2: Software and hardware are designed together for parallel processing and scalability.

High-Speed InfiniBand Network
Networking is also very important for the appliance to achieve
breakthrough query performance. The hardware ships with dual
InfiniBand networks, and supports fast data transfer up to 56 Gb/s by
using FDR InfiniBand and Mellanox ConnectX-3 FDR InfiniBand network
adapters. Each rack has redundant Ethernet switches for connecting to
your corporate network.
For fast loading and database backups, we recommend connecting your
own loading and backup servers to the appliance’s InfiniBand network.


12

Scalable
Solution for
Today’s Data
Challenges

“We are able to consolidate our data
from multiple databases onto PDW
where we can run analysis queries
that simply weren’t possible before on
our old databases.”
PDW does more than achieve breakthrough query performance and
scalability. The thoughtful design of PDW results in a solution capable of
solving data challenges that face businesses today as they try to load,
store, manage, and analyze data in order to make faster and more
accurate business decisions.

Easy to Add
Capacity

“Our data is growing faster than
expected and we need more capacity
within the next 12 months.”
Eventually, as your data grows, you need to plan for more capacity. With
SMP systems, such as SQL Server 2008 R2, SQL Server 2012, Oracle, or
DB2, the only way to add more capacity is to purchase a bigger system
that has faster processors, more memory, and more storage.
Incremental growth is not a viable option, especially for adding more
processors.
Even when you purchase a bigger system, it inevitably becomes difficult
to efficiently use one system to meet your growing processing and
storage requirements. As a result, businesses are often burdened with
the complexity and cost of developing and maintaining their own
software to distribute their workload across multiple computing
resources.


13

Figure 3: Scalability of today’s SMP systems compared to PDW’s MPP design

With PDW’s MPP design, you do not need to buy a new system in order
to add capacity. Instead, PDW grows by adding to the existing system.
You don’t need to over-buy and waste storage that you don’t need and
if you under-buy you can quickly add more capacity if your data growth
is faster than projected. You also don’t need to migrate your data to a
new system in order to add capacity. You can scale out without having
to redesign your application or re-engineer the distribution mechanism.
PDW is designed to expand processing, memory, and storage by adding
scale units consisting of 2 to 3 Compute nodes. By scaling out, you can
easily expand capacity to handle a few terabytes to over 6 petabytes in a
single appliance. When one rack is full, you can purchase another rack
and start filling it with Compute nodes. There is no need to migrate your
data to a new system, and no need to restructure your database files to
accommodate more nodes. PDW takes care of redistributing your data
across more Compute nodes.
The following diagram shows a DELL expansion rack with scalability
options.


14

Figure 4: After the base rack is full, PDW grows by adding scale units to expansion racks

Easy to Load
Data Fast

“With our old system, it took 4 hours
to load one day of data. With PDW,
the same load takes only a few
minutes and we are able to load the
data without incurring any downtime.”
You might have plenty of storage space available in your data
warehouse, but if you can’t get the data inserted into SQL Server fast
enough, your loading process will be the bottleneck.
PDW loads data up to 7 times faster than SQL Server 2012. Loads are
faster with PDW than SMP SQL Server because the data is loaded, in
parallel, into multiple instances of SQL Server. For example, if you have
10 Compute nodes and you load 1 Terabyte of data, you will have 10
independent SQL Server 2012 databases that are each compressing and

15

bulk inserting 100 GB of data at the same time. For most customers, this
is 7 times faster than loading 1 TB into one instance of SQL Server 2012.

Figure 5: Data loads in parallel into multiple instances of SQL Server

Using PDW, loading is simpler than loading into SMP SQL Server. For
SMP SQL Server, the Data Loading Performance Guide
(http://msdn.microsoft.com/en-us/library/dd425070(v=SQL.100).aspx)
explains optimizations for loading data. These are suggestions and
configurations that you can perform manually. PDW has many of these
optimizations already built-in to the PDW architecture and design,
making it easier to optimize than SMP SQL Server.
PDW ships with a command-line loading tool called dwloader. Using this
tool is the fastest way to load data into PDW and popular among
customers who do in-database set-based transformation of data. You
can also use SQL Server Integration Services (SSIS) to bulk load data into
PDW. The data is loaded from an off-appliance loading staging server
into PDW Compute nodes. Informatica also works with PDW.
Loading Tools:


dwloader Command-line Loading Tool



SQL Server Integration Services (SSIS)

To load data into the Compute nodes by using dwloader, first use your
own ETL process to create the source data you wish to load. The source
data must be formatted to match the schema of the destination table.
Store the source data into one or more text files or gzip files, and copy
files into the same directory on your loading server. Then, run dwloader
on your loading server (or ETL server) and dwloader will communicate
with the Control node to perform the load.


16

Easy to
Retain More
Data

“With PDW, we can store enough
data to analyze trends over the last
several years. On our old system, we
could only store enough data to
analyze trends for just 1 year.”
As the demand for more data analysis and insight grows, businesses are
faced with the challenge of retaining more online data in their data
warehouse. Often, the data warehouse is not big enough to store all of
the data that analysts need to make more accurate business decisions.
You also might have compliance reasons for keeping a particular
number of years of data online.
A common way to manage data growth is to use the sliding window
approach to data retention in which older data is moved onto tapes or
other cheap storage in order to make room for more recent data. For
example, you could keep 12 months of data online in your data
warehouse. Every month, you might archive the oldest partition to tape,
and then reuse the online storage for the current month.
Data stored offline onto tapes is hard to access and therefore impractical
to analyze. To analyze data stored on tapes, the data must first be
restored to an online system. This takes time and requires available disk
space to hold the restored data. Sometimes it’s a challenge just to find
the right tapes.
PDW’s scale-out architecture makes it easy to add capacity so that you
can retain more data online for analysis and reporting. Plus, just like SMP
systems, you still have the flexibility to partition your data and to archive
portions of it off the data warehouse according to your business needs.


17

Easy to
Consolidate
Data

“Our data was in so many different
places, that we couldn’t keep track of
it, or even begin to analyze it. Now,
with PDW, we have consolidated
many databases onto PDW.”
Businesses get data from web clicks, customer transactions, social
networks, and a plethora of other data sources. Consolidating disparate
data into one data warehouse is often not feasible because of the data
size. As a result, management of the data is decentralized and complex.
Wouldn’t it be nice to consolidate your data into one central data
warehouse? This makes the data warehouse much more usable because
the data is in one place and can be managed centrally. Simplifying data
warehouse management lowers the total cost of ownership. You only
need to manage one system, and plan for capacity growth for that one
system. Your business also benefits from being able to run queries on
any of the data which is all stored on one system, or to run queries that
select from multiple databases.
Consolidation has several business benefits:










As hardware reaches end-of-life, instead of a complicated process of
refreshing many hardware systems, you can migrate the data to
PDW and then manage the data on one system.
DBA’s can manage all of the data warehouse data in one place which
makes their job simpler, and reduces management costs.
PDW uses Windows Server Update Services to perform software
updates on all of the servers. One software update through Windows
Server Update Services updates all of the servers in the appliance.
All of your data sets benefits from the redundancy and high
availability of the appliance.
You can monitor the health and state of the appliance from one
centralized place. PDW has an admin console that DBA’s and
appliance administrators can use to see the state of queries and the
health of the appliance.


18



Why Queries
Run Fast

Perform more comprehensive analysis where analysts can ask
questions involving multiple data sources that are all centrally
located. This makes it easy to cross-analyze data from different data
sources.

“When we first started to use PDW,
we couldn’t believe how fast our
queries were finishing. In fact, we kept
running them “again” just to convince
ourselves the queries actually ran.”
PDW’s MPP design is the cornerstone for breakthrough query
performance. Queries run fast because they run in parallel on distributed
data. PDW’s cost-based query optimizer figures out how to run each
parallel query fast, and Data Movement Service (DMS) efficiently
transfers data among the Compute nodes.

Queries Run
on
Distributed
Data

To support parallel query processing, PDW distributes fact table rows
across the Compute nodes and stores the table as many smaller physical
tables.
Within each Compute node the distributed data is stored into 8 physical
tables that are each stored on independent disk-pairs. Each independent
storage location is called a distribution. PDW runs queries in parallel on
each distribution. Since every Compute node has 8 distributions, the
degree of parallelism for a query is determined by the number of
Compute nodes. For example, if your appliance has 8 Compute nodes
your queries will run in parallel on 64 distributions across the appliance.
The following diagram shows data being distributed to four Compute
nodes and 32 distributions.


19

Figure 6: Fact table rows are stored across all of the Compute node distributions

When PDW distributes a fact table, it uses one of the columns as the key
for determining the distribution to which the row belongs. A hash
function assigns each row to a distribution according to the key value in
the distribution column. Every row in a table belongs to one and only
one distribution. If you don’t choose the best distribution column when
create the table, it is easy to re-create the table to use a different
distribution column.
PDW doesn’t require that all tables get distributed. Small dimension
tables are usually replicated to each Compute node. Replicating small
tables speeds query processing since the data is always available on
each Compute node and there is no need to waste time transferring the
data among the nodes in order to satisfy a query.

Queries Are
Highly
Parallelized

PDW’s cost-based query optimizer is the “secret sauce” that makes
parallel queries run fast and return accurate results. A result of
Microsoft’s extensive research and development efforts, the query
optimizer uses proprietary algorithms to successfully choose a high
performing parallel query plan.
The parallel query plan, also called the “dsql” plan, contains all of the
operations necessary to run the query in parallel. As a result, PDW
handles all the complexity that comes with parallel processing and
processes the query seamlessly in parallel behind the scenes. The results
are streamed back to the client as though only one instance of SQL
Server ran the query.


20

Here’s a look into how PDW query processing works “under the covers”.

Figure 7: PDW chooses a query plan that runs queries fast and accurately in parallel.

First, a query client submits a Transact-SQL user query to the Control
node, which coordinates the parallel query process. All query clients
connect with SQL Server Native Client, or .NET, so you can use SQL
Server Data Tools, sqlcmd, or your own application to submit the query.
Query Clients
 Customer LOB applications that connect with SQL Server Native
Client or .Net Framework


SQL Server Data Tools



SQL Server’s sqlcmd Command-Line SQL Client

After receiving a query, PDW’s cost-based parallel query optimizer uses
statistics to choose a query plan, from many options, for running the
user query in parallel across the Compute nodes. The Control node
sends the parallel query plan, called the dsql plan, to the Compute
nodes, and then the Compute nodes run the parallel query plan on their
portion of the data.
The Compute nodes each use SQL Server 2012 to run their portion of
the query. When the Compute nodes finish, the results are quickly
streamed back to the client through the Control node. All of this occurs
quickly without landing data on the Control node, and the data does not
bottleneck at the Control node.


21

Data
Movement
Service
Transfers
Data Fast

PDW data relies on colocated data, which means the right data must be
on the right Compute node at the right time before running a query.
When two tables use the same distribution column they can be joined
without moving data. Data movement is necessary, though, when a
distributed table is joined to another distributed table and the two
tables are not distributed on the same column.

In-Memory
Clustered
Columnstore
Indexes
Improve
Query
Performance

PDW uses in-memory clustered columnstore indexes to improve query
performance and to store data more efficiently. These indexes are
updateable, and are applied to the data after it is distributed.

PDW uses Data Movement Service (DMS) to move data, as necessary, for
parallel query processing. DMS only moves the minimum amount of
data necessary to satisfy the query. Since data movement takes time, the
query optimizer considers the cost of moving the data when it chooses a
query plan.

A clustered columnstore index is a technology for storing, retrieving and
managing data by using a columnar data format, called a columnstore.
The data is compressed, stored, and managed as a collection of partial
columns, called column segments.

Figure 8: Clustered columnstore indexes improve query speed and data compression


22

PDW uses the column-based data layout to significantly improve
compression rates and query execution time. For example:








Columns often have similar data, which results in high compression
rates. In turn, higher compression rates further improve query
performance by using a smaller in-memory and I/O footprint.
High compression rates improve query performance by using a
smaller in-memory footprint. In turn, query performance can
improve because SQL Server PDW can perform more query and data
operations in-memory.
Most queries select only a few columns from a table, which reduces
total I/O to and from the physical media. The I/O is reduced because
columnstore tables are stored and retrieved by column segments,
and not by B-tree pages.
Advanced query execution technology processes chunks of columns
called batches in a streamlined manner, which reduces CPU usage.

To prevent fragmented column segments, some of the clustered
columnstore index data is stored temporarily in a rowstore table, called a
deltastore, until it is compressed and moved into the columnstore. The
clustered columnstore index operates on both the columnstore and the
deltastore, to return the correct query results.

Why You
Can Trust
PDW

PDW’s design is built in Windows Server 2012 Storage Spaces, Failover
Clustering, and Hyper-V to keep your data safe and highly available.


23

PDW’s
Hardware
Redundancy
Keeps Your
Data Safe

The hardware is designed to make each Compute node run efficiently on
its own hardware without interfering with the performance of other
Compute nodes or storage. Each Compute node runs in a Virtual
machine on one server that is direct-attached to two Windows Storage
Spaces disk arrays.
The hardware is designed for redundancy too, so that you can trust PDW
to not lose your data.
Capacity is added to PDW by adding scale units. In the scale unit, one
Compute node runs on each server, and writes to 16 pairs of dedicated
mirrored disks. The number of Compute nodes per scale unit depends
on the hardware vendor. The following diagram shows how 2 Compute
nodes are grouped together to form a scale unit.

Figure 9: A scale unit with 2 Compute nodes

By using Windows Storage Spaces, each Compute node has read and
write access to all the disks in the Windows Storage Spaces disk arrays
for the scale unit. If the server fails, the Compute node fails over to a
spare server in the rack. It continues to run and uses the InfiniBand
network to access its disks through another server in the scale unit.

High
Availability

The hardware and software are architected together for high availability.
Using features in Windows Server 2012, PDW is able to run each node
within a virtual environment. With this ability, the nodes can technically
run on any server. When a server goes down, PDW uses Windows
Failover Clustering to fail the node, which runs in a Virtual Machine, over
to the spare server.
Using Windows Storage spaces, each Compute node can read and write
to all of the disks in the scale unit, if it needs to. If a Compute node fails
to the spare server, the Compute node can continue running and
accessing its data and disks through the InfiniBand connection to
another server in its scale unit.


24

Figure 10: Redundant and high availability design keeps your data safe and online.

PDW is Easy
to Use and
Manage

“We can’t afford to lose our current
employees and hire new ones to
support a new data warehouse
technology. Please make it simpler to
maintain a data warehouse!”


25

PDW has a minimal learning curve. There’s no need to hire new talent
in order to move from SQL Server to SQL Server PDW. DBAs who already
know Transact-SQL can easily transfer their knowledge to PDW. Client
queries are written with Transact-SQL statements. Some statements are
added or extended to accommodate the MPP architecture. Some
statements are not needed because they don’t apply to MPP, or PDW is
already performing the operations behind the scenes. There’s no need
to hire new talent in order to move from SQL Server to SQL Server PDW.
PDW is easy to configure. When the appliance arrives at your data
center, the hardware and software are already installed and configured.
PDW ships with a checklist of tasks for configuring your system, and has
a built-in Configuration Manager for perform many of the
configurations.

Figure 11: Administrators configure the appliance with the Configuration Manager Tool

PDW has built-in monitoring tools. There is a web-based Admin
Console that users and administrators can use to monitor the appliance.
The Admin Console has information about queries, loads, and backups.
It also has information about the appliance health, and storage
resources.
Information surfaced by the Admin Console can also be viewed through
scripts that query dynamic management views. In addition, PDW has
Management Packs so that you can use System Center Operations
Manager (SCOM) to monitor PDW.


26

Figure 12: PDW has a web-based Admin Console for monitoring the appliance.

Platform for
Data Mining
and Analysis

“It’s not enough to store data in
Hadoop. Businesses today need to
figure out how they can analyze
Hadoop data fast and seamlessly in
order to make more informed
business decisions.”
In today’s world, data is generated from many sources, some of which
are not suitable for relational database management systems. For
example, businesses could have clickstream data that is generated at a
rapid rate, social sentiment data with different schemas, sensor data, or
telemetry data feeds. The buzz about Big Data is not only about figuring
out how to store, manage, and analyze data from non-relational sources,
but also about mashing various non-relational data with an
organization’s relational data to gain business insight.


27

PolyBase for
Hadoop
Integration

Today, many businesses are exploring possibilities with Hadoop to store
their non-relational data. Hadoop is open source software that manages
non-relational data across many nodes. It is easy to add data to Hadoop,
but not as quick to extract and analyze the data you need. The idea is
that, if the data is there, it may take a while to retrieve it, but at least the
data is stored somewhere in the system. Hadoop is best for batchoriented processing.
PDW’s PolyBase is a breakthrough technology for analyzing Hadoop
data, and it opens up a new world of data analysis possibilities. Queries
that run too slow in Hadoop can now run quickly in PDW, data mining
queries can combine Hadoop and PDW data, Hadoop data can be
stored as relational data in PDW, and query results can be stored back to
Hadoop. By using the power of PDW to run queries on Hadoop data, it
is now possible to do more in-depth data mining, reporting, and analysis
without acquiring the skills to run MapReduce queries in Hadoop.
PolyBase gives you the flexibility to bring the Hadoop data you need,
when you need it, into PDW for fast analysis. You can seamlessly select
from both Hadoop data and PDW data in the same query, and join data
from both data sources. To satisfy a query, PolyBase transfers data
quickly and directly between PDW’s Compute nodes and Hadoop’s
DataNodes.

Figure 13: Data transfers in parallel between PDW Compute nodes and HDFS DataNodes.

PDW uses external tables to point to data stored on Hadoop. Once an
external table is created, the table can be used in a SELECT statement in
the same manner as a PDW table.
The following example creates an external table called
DimCustomer_Hadoop_Export and transfers data from the DimCustomer
table in PDW to the DimCustomerExport.txt file residing in the Hadoop
Cluster.


28

CREATE EXTERNAL TABLE DimCustomer_Hadoop_Export
WITH
(
LOCATION =
ˈhdfs://10.192.63.147:8020/DemoDb/DimCustomerExport.t
xt,
FORMAT_OPTIONS ( FIELD_TERMINATOR = ˈ|ˈ)
)
AS SELECT * FROM DimCustomer;
To create an external table without transferring data, the syntax is similar
to creating a PDW table. This example creates an external table called
ClickStream that points to data stored in a text file called employee.txt
on Hadoop.
CREATE EXTERNAL TABLE ClickStream (
url varchar(50),
event_date date,
user_IP varchar(50)
)
WITH (
LOCATION =
'hdfs://10.192.63.147:8020/DemoDb/employee.txt',
FORMAT_OPTIONS (
FIELD_TERMINATOR = '|'),
DATE_FORMAT = ꞌMM/dd/yyyyꞌ
)
)
[;]
To view this external table in PDW or to perform operations on the data,
simply SELECT from it as from any other table.
SELECT * FROM ClickStream;
For an example of combining Hadoop data and PDW data to solve a
problem, see the video Insight Through Integration: SQL Server 2012
Parallel Data Warehouse Demo - PolyBase Video
(http://go.microsoft.com/fwlink/?LinkId=320541).
PDW’s deep integration with Business Intelligence (BI) tools makes PDW
a comprehensive platform for building end-to-end data mining and
analysis solutions. PDW integrates with both the Microsoft BI Stack
(Reporting Services, Analysis Services, PowerPivot for Excel, PowerView).
It also integrates with a growing list of leading non-Microsoft BI
platforms, such as Business Objects, Cognos, SAP Data Integrator,
Tableau, MicroStrategy, QlikView, Oracle Business Intelligence, and
TIBCO Spotfire.


29

Business
Intelligence
Integration

BI tools connect to PDW by using the standard SQL Server connection
tools, SQL Server Native Client or the .NET Data Provider for SQL Server.
The fact that all these BI tools can access a PDW database as though it is
a standard SQL Server 2012 database allows you to quickly update or
modify BI solutions to point to PDW.
New in SQL Server 2012 PDW, you can use PDW as a high performance
relational data source for building multidimensional OR tabular models
with SQL Server Analysis Services (SSAS). For example, you can:








Summary

Use DirectQuery to perform real-time queries against a SQL Server
PDW data source from a tabular model.
Reduce ROLAP query times by using clustered columnstore indexes
on SQL Server PDW tables.
Use new features of SSAS. For example, use
EnableRolapDistinctCountOnDataSource to run distinct count
operations on PDW tables.
Use PDW as a data warehouse for aggregating Hadoop data for use
in multidimensional or tabular models.

Microsoft SQL Server 2012 Parallel Data Warehouse is a compelling and
innovative solution for next generation data challenges. It enables you
to make faster and more accurate business decisions by providing an
innovative platform for managing data warehousing workloads that
scales from a few terabytes to over 6 petabytes of data in a single
appliance. Queries commonly run 50 times faster than traditional
relational database systems. Significantly higher data compression ratios
allow more data to be stored cost efficiently. PDW has deep integration
with Microsoft’s BI tools and other leading non-Microsoft tools, making
it simple to use the BI tools you are familiar with to perform analysis.
PolyBase technology enables you to query Hadoop data and merge it
with your relational data by using Transact-SQL and without learning a
host of new skills. The management costs are minimal and the learning
curve for SQL Server DBAs is easy, making it possible to use your current
DBAs to manage PDW.
PDW’s breakthrough query performance and scalability result from
PDW’s purpose-built design for data warehousing workloads. PDW’s
hardware and software are designed for fast parallel processing,
distributed data storage, hardware redundancy, scalable capacity, and
high availability. All of these factors give PDW the ability to handle
growing storage and analysis needs for your data. As a result, upgrading
from SQL Server to SQL Server PDW is an important step forward
towards empowering your business to manage growing data needs and
towards performing complex analysis that enables you to make faster
and more accurate business decisions.

30

For additional information:
 SQL Server 2012 Parallel Data Warehouse Web site:
http://www.microsoft.com/PDW/


Microsoft data warehousing web site: http://www.microsoft.com/enus/sqlserver/solutions-technologies/data-warehousing.aspx

Did this paper help you? Please give us your feedback. Tell us on a scale
of 1 (poor) to 5 (excellent), how would you rate this paper and why have
you given it this rating? For example:




Are you rating it high due to having good examples, excellent screen
shots, clear writing, or another reason?
Are you rating it low due to poor examples, fuzzy screen shots, or
unclear writing?

This feedback will help us improve the quality of white papers we
release.
Send feedback.
About the Authors
Barbara Kess leads data warehousing content at Microsoft and is the
primary author responsible for SQL Server PDW content.
Dan Kogan leads Database Appliance marketing at Microsoft and is
responsible for the SQL Server PDW product.
Acknowledgements
A special thank you to the SQL Server PDW engineering, consulting, and
sales teams who have dedicated themselves to developing and shipping
a great product. They have relentlessly reviewed the PDW product
documentation, which much of this paper is based on.
Thanks to Murshed Zaman and John Hoang who are on the SQL Server
Customer Advisory Team and use their everyday experiences helping
customers to develop best practices for SQL Server PDW. Their deep
experience with PDW is embedded in the details of this paper.
Henk van der Valk is a Technical Solution Professional who works with
prospective customers who are interested in SQL Server PDW. His input
helped us to better explain what our customers want to know.
Thanks to Rick Byham, content author at Microsoft, for reviewing the
paper and giving insightful feedback.


31

Sql server 2012_parallel_data_warehouse_breakthrough_platform_white_paper

Recommended

More Related Content

What's hot (20)

Viewers also liked (13)

Similar to Sql server 2012_parallel_data_warehouse_breakthrough_platform_white_paper (20)

Recently uploaded (20)

Sql server 2012_parallel_data_warehouse_breakthrough_platform_white_paper