1. Microsoft Analytics Platform System
02 – Hardware/Software Architecture
Brian Walker | Microsoft ​Architect – Data Insights COE
Jesse Fountain| Microsoft ​WW TSP Lead
November 23, 2018

2. Agenda SMP and MPP differences What is APS? APS hardware components
High availability
PDW region overview
PDW tools of the trade

3. Differences between SMP and MPP

4. SQL upgrade prospect scenario: SQL Server 2014 or APS
SQL Server customers with scale and performance needs have a choice to make
Customer Application
SQL 2014
Solution? New H/W & back to Step1
New Requirements:
More Business, More Growth !
Step 2: Optimize
Step 1: Upgrade
Customer Application
SQL 2008 R2
Effects All Systems. Total Upgrade.
Upgrade 2 releases greater
Infrastructure Re-Architecture
Replace/Upgrade SAN
Regression Testing
Re-optimization is required for new version
Re-Optimization required
Unknown maintenance & cost requirements
Unknown gain in performance
Heavy testing and thought required
Start Here
Customer Application
APS
New Requirements?
Step 1: Migrate
Solution?
Add Capacity
Push-Button Upgrade
Automatic Data Distribution
Infrastructure Optimized
Data Center Rack Optimized
Migrate once. Quickly deliver value
Pre-Optimized Infrastructure
Best Practice Configuration Built-In
Automatic Data Distribution & Placement
Migrate once and upgrades are managed

5. Data warehousing comparison: SMP vs. MPP
Data Volume
Mixed
Workload
50 TB
100 TB
500 TB
10 TB
5 PB
Query Concurrency
Strategic, Tactical
Strategic
Loads
Loads, SLA
APS – Multi-dimensonal Scalability
SMP – Tunable in one dimension at cost of other dimensions
1.000
100
10.000
The spiderweb depicts important attributes to consider when evaluating Data Warehousing options.
Big Data support is newest dimension.
Data
Freshness
Query complexity
Near Real Time
Data Feeds
Daily
Load
Weekly
3-5 Way Joins
Joins +
OLAP operations +
Aggregation +
Complex “Where” constraints +
Views
Parallelism
5-10 Way Joins
Normalized
Multiple, Integrated Stars and Normalized
Simple Star
Multiple, Integrated Stars
Batch Reporting, Repetitive Queries
Ad Hoc Queries Data Analysis/Mining
TB’s
MB’s
GB’s
Query Freedom
Schema Sophistication
Query Data Volume

6. Management simplicity and lower operational costs
Database Administration Task
APS (MPP SQL Server)
SQL Server 2014 (SMP SQL Server)
Logical Data Modeling
High
Physical Data Modeling
Low
Data Partitioning Definition
Data Placement Definition
Auto
Free Space Management
Data Balancing Control
Data Reorganization
None
Moderate
Index Reorganization
Workspace Management
Query Tuning
Change Management
Rearchitect Environment
Never
Often
Database management in APS
Built-in high availability and failover
Linear scalability by adding nodes
Minimal tuning efforts and troubleshooting
Simple database and table definition
Unified administration console
Minimum ongoing maintenance
No need to manage disk or database subsystems
No detailed space management needed
No memory/cache management needed
No optimizer hints needed
No need to manage parallelism
No need to manage physical computing nodes
No index reorgs needed
No index rebuilds needed
Use DBAs for higher value activity, not low-level system management

7. What is APS?

8. Microsoft Analytics Platform System
Low-Cost, Rapid Value
Appliance: System is pre-configured at factory
Industry Standard: Co-engineered with HP/Dell/Quanta
Automatic Compression: 5X-15X with Columnar Technology
Insight on All Data
POLYBASE-SQL: easy-access to both relational and Hadoop data
Near Real-Time: Access multiple data sources quickly
Shared-Nothing: allow linear scalability to store all historical data
High-Performance Analytics
MPP: Powerful: Massively Parallel Processing (MPP) Engine
Mature: Parallel Cost-Based Optimizer from SQL Server
Dedicated: Direct-attached high-speed servers and storage

9. The foundation for data warehousing and advanced analytics
Combines Hardware & Software to provide a turn-key, balanced platform specific to data warehouse & analytical workloads
Built for easy scale-out as Data Warehouse capacity requirements grow
Deep, native integration with Hadoop
High Performance & Concurrent Data Warehouse Workloads for simultaneous data Loading & Query
Built-in Development Engineering Best Practices
Integrated Systems Monitoring & Management

10. New features come first to APS
Updateable Column Store
Agnostic Hadoop Integration via Polybase
Cardinality Estimation
Cost-Based Distributed SQL Query Engine
Hub and Spoke Architecture Support
Analytical Functions (e.g. Lag and Lead)
Incremental functional releases each year

11. APS hardware components

12. Rack and network Contains Also added Rack Ethernet Switches
InfiniBand Switches
Also added
Power Units (PDU)

13. PDW Base Scale Unit Contains Orchestration Host Passive Host
Optional Passive Host
Data Scale Unit

14. Hadoop Base Scale Unit Contains Rack & Network PDW Base Scale Unit
Orchestration Host
Passive Host
Data Scale Unit

15. Data Scale Unit Servers “active” in WFC Unit of growth
Used by both regions
Varies in size
By Vendor
By Appliance Size
Uses Existing Switches

16. Details HP configuration 2 – 56 compute nodes 1 – 7 racks
MGXFY13
11/23/2018
HP configuration
Base Unit (6U):
Redundant Infiniband
Redundant Ethernet
Mgmt & Control (Active)
Rack Failover Node (Passive)
Extension Base Unit (5U):
Redundant Infiniband
Redundant Ethernet
Rack Failover Node (Passive)
Extension Base Unit (5U):
Redundant Infiniband
Redundant Ethernet
Rack Failover Node (Passive)
Infiniband
Ethernet
Control Node
Failover Node
Infiniband
Ethernet
Failover Node
Infiniband
Ethernet
Failover Node
2 – 56 compute nodes
1 – 7 racks
1, 2, or 3 TB drives
15.1 – TB raw
53 – 6342 TB User data
Up to 7 spare nodes available across the entire appliance
Details
Reserved
Customer Space (9U)
ETL Servers
Backup Servers
Passive Unit (Additional spares)
Reserved
Reserved Space (9U)
Reserved
Customer Space (8U)
ETL Servers
Backup Servers
Passive Unit (Additional spares)
JBOD 4
Compute Node 7
Compute Node 8
Scale Unit (7U):
2 HP 1U Servers
(16 Cores/Ea. Total: 32)
JBOD 5U
1TB Drives
User Data Capacity: 75TB
JBOD 8
Compute Node 15
Compute Node 16
Scale Unit (7U):
2 HP 1U Servers
(16 Cores/Ea. Total: 32)
JBOD 5U
1TB Drives
User Data Capacity: 75TB
JBOD 12
Compute Node 23
Compute Node 24
60TB (Raw)
Full Rack
Scale Unit (7U):
2 HP 1U Servers
(16 Cores/Ea. Total: 32)
JBOD 5U
1TB Drives
User Data Capacity: 75TB
120.8TB (Raw)
2 Rack
181.2TB (Raw)
3 Rack
JBOD 3
Compute Node 5
Compute Node 6
Scale Unit (7U):
2 HP 1U Servers
(16 Cores/Ea. Total: 32)
JBOD 5U
1TB Drives
User Data Capacity: 75TB
JBOD 7
Compute Node 13
Compute Node 14
Scale Unit (7U):
2 HP 1U Servers
(16 Cores/Ea. Total: 32)
JBOD 5U
1TB Drives
User Data Capacity: 75TB
JBOD 11
Compute Node 21
Compute Node 22
Scale Unit (7U):
2 HP 1U Servers
(16 Cores/Ea. Total: 32)
JBOD 5U
1TB Drives
User Data Capacity: 75TB
JBOD 2
Compute Node 3
Compute Node 4
Scale Unit (7U):
2 HP 1U Servers
(16 Cores/Ea. Total: 32)
JBOD 5U
1TB Drives
User Data Capacity: 75TB
JBOD 6
Compute Node 11
Compute Node 12
Scale Unit (7U):
2 HP 1U Servers
(16 Cores/Ea. Total: 32)
JBOD 5U
1TB Drives
User Data Capacity: 75TB
JBOD 10
Compute Node 19
Compute Node 20
1/2 Rack
30TB (Raw)
Scale Unit (7U):
2 HP 1U Servers
(16 Cores/Ea. Total: 32)
JBOD 5U
1TB Drives
User Data Capacity: 75TB
90.6TB (Raw)
1 1/2 Rack
JBOD 1
Compute Node 1
Compute Node 2
Base Unit (7U):
2 HP 1U Servers
(16 Cores/Ea. Total: 32)
JBOD 5U
1TB Drives
User Data Capacity: 75TB
JBOD 5
Compute Node 9
Compute Node 10
Extension Base Unit (7U):
2 HP 1U Servers
(16 Cores/Ea. Total: 32)
JBOD 5U
1TB Drives
User Data Capacity: 75TB
JBOD 9
Compute Node 17
Compute Node 18
15TB (Raw)
¼ Rack
Extension Base Unit (7U):
2 HP 1U Servers
(16 Cores/Ea. Total: 32)
JBOD 5U
1TB Drives
User Data Capacity: 75TB
1¼ Rack
75.5TB (Raw)
© 2012 Microsoft Corporation. All rights reserved. Microsoft, Windows, and other product names are or may be registered trademarks and/or trademarks in the U.S. and/or other countries.
The information herein is for informational purposes only and represents the current view of Microsoft Corporation as of the date of this presentation. Because Microsoft must respond to changing market conditions, it should not be interpreted to be a commitment on the part of Microsoft, and Microsoft cannot guarantee the accuracy of any information provided after the date of this presentation. MICROSOFT MAKES NO WARRANTIES, EXPRESS, IMPLIED OR STATUTORY, AS TO THE INFORMATION IN THIS PRESENTATION.

17. Dell and Quanta configuration
MGXFY13
11/23/2018
Dell and Quanta configuration
Infiniband
Ethernet
Control Node
Failover Node
PDW Backplane (6U):
Redundant Infiniband
Redundant Ethernet
Mgmt & Control (Active)
Rack Failover Node (Passive)
2 – 54 compute nodes
1 – 6 racks
1, 2, or 3 TB drives
22.65 – TB raw
79 – 6116 TB User data
Up to 6 spare nodes available across the entire appliance
Details
Reserved
Reserved (6U)
JBOD 5
Compute Node 8
Compute Node 9
JBOD 6
Compute Node 7
Base Unit (10U):
3 Servers in 2U enclosure
(16 Cores/Ea. Total: 48)
2 JBOD 4U ea.
1TB Drives
User Data Capacity: 79TB
67.9TB (Raw)
Full Rack
JBOD 3
Compute Node 5
Compute Node 6
JBOD 4
Compute Node 4
Base Unit (10U):
3 Servers in 2U enclosure
(16 Cores/Ea. Total: 48)
2 JBOD 4U ea.
1TB Drives
User Data Capacity: 79TB
45.3TB (Raw)
2/3 Rack
JBOD 1
Compute Node 2
Compute Node 3
JBOD 2
Compute Node 1
Base Unit (10U):
3 Servers in 2U enclosure
(16 Cores/Ea. Total: 48)
2 JBOD 4U ea.
1TB Drives
User Data Capacity: 79TB
22.6TB (Raw)
1/3 Rack
JBOD 2
Compute Node 2
Compute Node 3
JBOD 1
Compute Node 1
© 2012 Microsoft Corporation. All rights reserved. Microsoft, Windows, and other product names are or may be registered trademarks and/or trademarks in the U.S. and/or other countries.
The information herein is for informational purposes only and represents the current view of Microsoft Corporation as of the date of this presentation. Because Microsoft must respond to changing market conditions, it should not be interpreted to be a commitment on the part of Microsoft, and Microsoft cannot guarantee the accuracy of any information provided after the date of this presentation. MICROSOFT MAKES NO WARRANTIES, EXPRESS, IMPLIED OR STATUTORY, AS TO THE INFORMATION IN THIS PRESENTATION.

18. High availability

19. Failover in Action: Control host failure
HST01 node marked as failed
(AD02 Persists on HST02)
Cluster fails over to HST02
HST02 already “warm” so Failover is very Fast
WFOHST01
HST01
CTL01
AD01
VMM
HST02
CTL01
AD02
VMM
HSA01
CMP01
ISCSI01
DAS01
HSA02
CMP02
ISCSI02
HSA03
CMP03
ISCSI03
DAS02
HSA04
CMP04
ISCSI04
HSA05
CMP05
ISCSI05
DAS03
HSA06
CMP06
ISCSI06

20. Failover in Action: Compute node failure
Compute node marked as failed
PDW Cluster restarts compute node on a passive server
ISCSI VM does not fail over
WFOHST01
HST01
CTL01
AD01
VMM
HST02
CMP01
HSA01
CMP01
ISCSI01
DAS01
HSA02
CMP02
ISCSI02
HSA03
CMP03
ISCSI03
DAS02
HSA04
CMP04
ISCSI04
HSA05
CMP05
ISCSI05
DAS03
HSA06
CMP06
ISCSI06

21. APS disk layout: LUNs and filegroups/files
Each LUN is composed of 2 drives in RAID1 mirroring configuration
Distributions are now split into 2 files
TempDB and Log are across all 16 LUNs
No fixed TempDB or log size allocation
VHDXs are on JBODs to ensure high availability
Disk I/O further parallelized relative to V1: bandwidth to increase by ~70% in V2 RTM
Design Details
Disk 1
Disk 2
Node 1: Distribution A – file 1
Node 1: Distribution A – file 2
Temp DB
Log
Disk 3
Disk 4
Disk 5
Disk 6
Node 1: Distribution B – file 2
Node 1: Distribution B – file 1
Disk 7
Disk 8 . .
Disk 29
Disk 30
Node 1: Distribution H – file 1
Node 1: Distribution H – file 2
Disk 31
Disk 32
Disk 33
Disk 34
Node 2: Distribution A – file 1
Temp DB
Log
Disk 35
Disk 36 . .
Disk 65
Disk 66
Fabric storage (VHDXs for node)
Disk 67
Disk 68
Hot spares
Disk 69
Disk 70
JBOD

22. Hadoop region HA #Scale Units Replication Factor Polybase =1 2 3 >1
Head/Controlling Nodes behave exactly the same as for PDW
Data Nodes are different
APS relies on Hadoop data replication for data availability
Disks are not Mirrored
Data Nodes do not failover
Replication Factor is configurable
#Scale Units
Replication Factor
Polybase =1 2 3
>1

23. PDW region overview

24. PDW Region Hadoop Region Appliance WFOHST01 HST01 HST02 HSA01 HSA02
CTL01
AD01
VMM
ISCSI01
ISCSI02
ISCSI04
ISCSI03
ISCSI05
ISCSI06
CMP01
CMP02
CMP03
CMP04
CMP05
CMP06
DAS01
DAS02
DAS03
AD02

25. Virtual Machine Manager Fabric Active Directory
PDW region nodes
WFOHST01
HST01
HST02
HSA01
HSA02
HSA03
HSA04
HSA05
HSA06
CTL01
AD01
VMM
ISCSI01
ISCSI02
ISCSI04
ISCSI03
ISCSI05
ISCSI06
CMP01
CMP02
CMP03
CMP04
CMP05
CMP06
DAS01
DAS02
DAS03
AD02
PDW Nodes
Control
Compute (>1)
Infrastructure Nodes
Virtual Machine Manager
Fabric Active Directory

26. Control and Compute workload nodes
Control Node
Compute Node
Compute Node
Compute Node
Compute Node
Compute Node
Compute Node

27. Services inside the Control node
PDW Services
Responsibilities
PDW Engine
DMS Core
PDW Agent
SQL Server
Admin Console (IIS)
Parse SQL / Syntax Check
Validate & Authorize
Generate D-SQL Plan
Orchestrate D-SQL Plan
Collate Diagnostic Info
Admin Console Web App

28. Services inside the Compute node
PDW Services
Responsibilities
DMS Core
PDW Agent
SQL Server
Hold User Data
Process Queries
Move Data
Load Data

29. Virtual Machine Manager (VMM)
Deployment of Virtual Machines
Configuration of Virtual Machines
Hosts Windows Update Services (WSUS)
Sits in the Fabric domain

30. PDW tools of the trade

31. Tools of the trade PDW Development Management SSDT (Visual Studio)
SQLCMD
SSIS / SSAS / SSRS
Power BI
SSIS adapters
dwloader.exe
Management Console
dwconfig.exe
pav.exe
PowerShell
System Center

32. Tool distribution with PDW
Location
Download
us/download/details.aspx?id=45294
Tools Distributed
SSIS Destinations
Clienttools.msi
dwloader
Adventureworks
Help File

33. Connecting from SQL 2012? Use SNAC 11 Connecting from SQL 2008 R2?
Connecting to PDW
Management Console
Development tools
TCP Port 17001
Security
Connecting from SQL 2012?
Use SNAC 11
Connecting from SQL 2008 R2?
Use SNAC 10

34. SQL Server Data Tools (SSDT)
Used for writing queries against PDW
SSMS is not a supported tool
SSDT for PDW part of standard SSDT deployment model

35. SQL Server Integration Services
Destination Adapters available for
SSIS 2008R2
SSIS 2012
SSIS 2014

36. BI solution development
Power Query
Power Pivot
Power Map
Power View
SQL Server Analysis Services
ROLAP Facts
MOLAP Dimensions
SQL Server Reporting Services

37. SQLCMD.exe -I SQLCMD support ..\Microsoft SQL Server\110\Tools\Binn\
SNAC R2
SNAC 11 for 2012+
QUOTED IDENTIFIER ON  Mandatory
Must be set at SQLCMD Invocation
SQLCMD.exe -I

38. Third-party tools Attunity Replicate MicroDesigner (MicroERD)
PDW Region supported target
warehousing/microsoft-pdw
MicroDesigner (MicroERD)
Data Modelling tool

39. PDW region configuration

42. Region must be restarted once reset

44. Management & monitoring

45. Management console Read-Only View of Management Information
Can cancel User Sessions, Queries, Loads
Can easily visualize DSQL structures
Requires View Server State
All data accessible via DMVs

47. Demo | Admin Console

48. Microsoft Analytics Platform System
11/23/2018
© 2014 Microsoft Corporation. All rights reserved. Microsoft, Windows, Windows Vista and other product names are or may be registered trademarks and/or trademarks in the U.S. and/or other countries.
The information herein is for informational purposes only and represents the current view of Microsoft Corporation as of the date of this presentation. Because Microsoft must respond to changing market conditions, it should not be interpreted to be a commitment on the part of Microsoft, and Microsoft cannot guarantee the accuracy of any information provided after the date of this presentation. MICROSOFT MAKES NO WARRANTIES, EXPRESS, IMPLIED OR STATUTORY, AS TO THE INFORMATION IN THIS PRESENTATION.
© 2014 Microsoft Corporation. All rights reserved. Microsoft, Windows, and other product names are or may be registered trademarks and/or trademarks in the U.S. and/or other countries.