IWA works with a snapshot of the data in Informix data mart. Once you have defined the data mart on IWA and loaded the data, you need to periodically refresh the data. You can choose to refresh all of the data or just the partitions that were added, dropped or modified. Whether you have hundreds of gigabytes or many terabytes, faster refresh will help you analyze recent data more rapidly and get closer to real-time business. This session will explain the options for data refresh, review their performance and explain how to correctly implement a refresh plan. IBM and Intel will demonstrate live data refresh on the Intel Xeon platform and examine the impact on performance.
3. Intel - Legal Disclaimers Performance
• Performance tests and ratings are measured using specific computer systems and/or components and reflect the approximate
performance of Intel products as measured by those tests. Any difference in system hardware or software design or
configuration may affect actual performance. Buyers should consult other sources of information to evaluate the performance of
systems or components they are considering purchasing. For more information on performance tests and on the performance of
Intel products, Go to: http://www.intel.com/performance/resources/benchmark_limitations.htm.
• Intel does not control or audit the design or implementation of third party benchmarks or Web sites referenced in this document.
Intel encourages all of its customers to visit the referenced Web sites or others where similar performance benchmarks are
reported and confirm whether the referenced benchmarks are accurate and reflect performance of systems available for
purchase.
• Relative performance is calculated by assigning a baseline value of 1.0 to one benchmark result, and then dividing the actual
benchmark result for the baseline platform into each of the specific benchmark results of each of the other platforms, and
assigning them a relative performance number that correlates with the performance improvements reported.
• INFORMATION IN THIS DOCUMENT IS PROVIDED “AS IS”. NO LICENSE, EXPRESS OR IMPLIED, BY ESTOPPEL OR
OTHERWISE, TO ANY INTELLECTUAL PROPERTY RIGHTS IS GRANTED BY THIS DOCUMENT. INTEL ASSUMES NO
LIABILITY WHATSOEVER AND INTEL DISCLAIMS ANY EXPRESS OR IMPLIED WARRANTY, RELATING TO THIS
INFORMATION INCLUDING LIABILITY OR WARRANTIES RELATING TO FITNESS FOR A PARTICULAR PURPOSE,
MERCHANTABILITY, OR INFRINGEMENT OF ANY PATENT, COPYRIGHT OR OTHER INTELLECTUAL PROPERTY RIGHT.
• Performance tests and ratings are measured using specific computer systems and/or components and reflect the approximate
performance of Intel products as measured by those tests. Any difference in system hardware or software design or configuration
may affect actual performance. Buyers should consult other sources of information to evaluate the performance of systems or
components they are considering purchasing. For more information on performance tests and on the performance of Intel
products, reference www.intel.com/software/products.
#ibmiod
4. Call to Action:
• For more information on the content covered in this session,
o go to the Demo Room and see demoes.
o download this analyst paper, etc
o go to this website, etc.
• Stop by the Data Mgmt Demo Room and see deep dive demos on more
than 25 products including DB2, Data Warehousing and Informix. The
Data Mgmt Demo Room is located in the far back left corner of the Expo
area.
• Don’t miss this Special Analyst Session!
In 2022, What Is a Database? Bloor, Forrester and IDC Analysts Discuss the Future-
Philip Howard of Bloor, Noel Yuhanna of Forrester and Carl Olofson of IDC peer
into the future of database software. Database software operates under many
paradigms today, from relational and hierarchical to cloud to NoSQL and
NewSQL to in-memory to columnar to the many aspects of Big Data and more.
Where will this lead? Will databases be code-centric or data-centric? Don't miss
this chance to hear where these veterans think we might be headed. (Session:
IDB4230a, Wednesday, 12:00pm-1:15pm, South Pacific F)
4
#ibmiod
5. Data Management Demo Room
DB2, Data Warehouse & Tools: Stop by the Data Mgmt Demo Room and
Adaptive Compression
Continuous Data Ingest see deep dive demos on more than 25
Cubing and Dynamic Cubes products including DB2, Data
Database Admin & Development Solutions
Data Studio Warehousing & Informix. The Data Mgmt
DB2 PureScale
DB2 Merge Backup
Demo Room is located in the far back left
DB2 Recovery Expert corner of the Expo area.
Ease of App. Migration
HADR Multiple Standby
InfoSphere Data Architect Informix:
InfoSphere Federation Server Cloud solutions
Multi-temperature Data Mgmt Flexible Grid
Optim High Perf. Unload Genero
Optim Performance Manager TimeSeries
Optim Query Workload Tuner’ Informix clustering
Optim Configuration Manager Informix Warehouse Accelerator
Optim pureQuery Runtime OpenAdmin Tool
Real-time Operational Warehousing PureSystems:
Row & Column Access Control
IBM Database Patterns
Time Travel Query
PureData System for Operational Analytics
Workload Mgmt
PureData System for Transactions
Zig-Zag Join
IBM Mobile Database
5
#ibmiod
6. Agenda
• Overview of Informix Warehouse Accelerator.
• Data Life Cycle scenarios and Performance
• Using Informix high availability to scale out
• Intel Inside® : Intel® Technology & Roadmap
o Scaling on Xeon® E7 Platform
6
#ibmiod
8. Motivation
• Data Warehouse query Performance without Perspiration
• Consistent query performance without tuning efforts.
• More questions, faster answers, better data driven decisions & business insights
• SKECHERS: Acceleration from 60x to 1400x – average acceleration of 450x
8
#ibmiod
9. Informix Ultimate Warehouse edition IBM Smart Analytics
Studio
Step 1. Install, configure,
start Informix
Step 2. Install, configure, Step 3
start Accelerator
Step 1
Step 3. Connect Studio to
Informix & add accelerator
Step 4
Informix Database Server
Step 4. Design, validate,
Deploy Data mart
Step 5
Step 5. Load data to
accelerator
Ready for Queries
BI Applications
Step 2
Ready
Informix warehouse Accelerator
9
#ibmiod
13. Distributing data from IDS (Fact tables)
Fact Table IDS Stored Procedures
Data Fragment Coordinator
UNLOAD
UNLOAD
UNLOAD
Process
Data Fragment UNLOAD
Data Fragment
Data Fragment
A copy of the IDS data is now
transferred over to the Worker Copy
process. The Worker process
holds a subset of the data Worker Worker Worker
(compressed) in main memory Process Process Process
and is able to execute queries
on this subset. The data is
evenly distributed (no value
based partitioning) across the
CPUs. Compressed Data Compressed Data Compressed Data
Compressed Data Compressed Data Compressed Data
13
#ibmiod
14. Distributing data from IDS (Dimension tables)
IDS Stored Procedure
IDS
Dimension Table
Dimension Table Coordinator
UNLOAD
UNLOAD
UNLOAD
Process
Dimension Table UNLOAD
Dimension Table
Worker Worker
Worker
All dimension tables are Process Process
Process
transferred to the worker
process.
14
#ibmiod
15. Data Transfer from Informix to IWA – First time
Design the Data Mart Deploy the Data Mart LOAD the mart
-- ISAO Studio Ready
-- ISAO Studio -- ISAO Studio for Queries
-- Workload Analysis -- Workload Analysis -- command line tool
Optionally lock the table Informix
Insert table data into external table
Send data over to IWA
Fact table – split into each worker
Dim table – copy to each worker
Compression frequency partitioning & encoding
IWA Write the memory image to disk
15
#ibmiod
16. Compression: Frequency Partitioning
Trade Info (volume, product, Column Partitions
origin country) Histogram
Occurrences
Number of
Vol Prod Origin on Origin
China GER,
USA FRA,
… Rest
Common Rare
Values values
Origin
Top 64
traded goods Cell Cell Cell 4
– 6 bit code 1 3
Product
Cell Cell Cell 6
Rest 2 5
Histogram
on Product Table partitioned
into Cells
• Field lengths vary between cells
• Higher Frequencies Shorter Codes (Approximate Huffman)
• Field lengths fixed within cells
16
#ibmiod
18. Case 1: Full refresh every time
Design the Data Mart Deploy the Data Mart LOAD the mart Ready
-- ISAO Studio -- ISAO Studio -- ISAO Studio (enabled)
-- Workload Analysis -- Workload Analysis -- command line tool for Queries
Disable the mart
• During the data load, queries cannot be accelerated.
• Work around: Create additional, duplicate mart.
• Informix always picks up the data mart that was last loaded.
18
#ibmiod
19. Time cyclic data management
Partitioned fact table, partitioned by week
working window
week1 week2 week3
19
#ibmiod
20. Time cyclic data management
Partitioned fact table, partitioned by week
working window
week1 week2 week3 week4
DETACH ATTACH
week1 week4
partition partition
20
#ibmiod
21. Case 2: Partition refresh: Updates to existing Partitions IBM Smart Analytics
Step 1. Create the Sales-Mart Studio or stored
and load it. Sales is the fact partitioned fact table procedures or
table -- range partitioned. command line tool
Step 2. Load jobs
update the fact table “sales” Step 1
Only updates existing partition customer
Step 2 sales
Step 3. Identify the partition,
execute dropPartMart(). Modified partition
SQL Script: call
Stored procedure
Step 4. for same partition,
execute loadPartMart(). stores
Step 3 Step 4
Informix Database Server
Ready for Queries
INSERT, UPDATE, DELETE
BI Applications
IWA
Sales-Mart Ready
OLTP Apps
21
#ibmiod
22. Time cyclic data management with IWA
Partitioned fact table, partitioned by week
working window
week1 week2 week3 week4 week6
DETACH ATTACH
partition partition
1. Execute dropPartMart()on IWA
2. DETACH partition from the table a. ATTACH the partition
b. Execute loadPartMart() on IWA
22
#ibmiod
23. Case 3: Partition refresh: Time Cyclic data management IBM Smart Analytics
Step 1. Create the Sales-Mart Studio or stored
and load it. Sales is the fact partitioned fact table procedures or
table -- range partitioned. command line tool
Need to move the Time
window to next range. Step 1
customer
ep 2. DETACH operation
sales
Execute dropPartMart()
Move the window.
DETACH the partition
ep 3. ATTACH operation
ATTACH the partition stores
Execute loadPartMart() Step 2 Step 3
Informix Database Server
Ready for Queries
BI Applications
IWA
Sales-Mart Ready
OLTP Apps
23
#ibmiod
24. dropPartMart() function
dropPartMart(accelerator name, mart name, table owner, table name, partition id or name)
Execute function dropPartMart(‘iwa’, ‘salesmart’, ‘dsusr’, ‘sales_fact’, ‘july_partition’);
1. Uses the accelerator name, datamart name, table name
and partition name.
2. Partition name can be the name of the partition or
partition number (sysfragments.partn)
The partition name or number should be a valid partition
for the table.
3. Call dropPartMart() first before doing the DEATCH
4. In IWA, the data is removed row by row; when all the
rows in a block is freed, the block and memory is freed.
5. No compression dictionary update on IWA.
24
#ibmiod
25. loadPartMart() function
loadPartMart(accelerator name, mart name, table owner, table name, partition id or name)
Execute function loadPartMart(‘iwa’, ‘salesmart’, ‘dsusr’, ‘sales_fact’, ‘july_partition’);
1. Uses the accelerator name, datamart name, table name and
partition name.
2. Partition name can be the name of the partition or partition
number (sysfragments.partn). The partition name or number
should be a valid partition for the table.
3. The partition should be in the table before loading to IWA. ATTACH
the partition first, before calling loadPartMart().
4. During every load, row is prefixed with the rowid and sent over to
IWA.
5. The data is encoded with existing compression dictionary and
25
hence very fast. #ibmiod
26. create table cust(k1 int primary key, name varchar(32),
age int); -- Simply update the data for existing partitions
create table store(k2 int primary key, sname varchar(18)); execute function dropPartMart('my_acc', 'salesmart', 'o1',
'salesfact', 'p3');
create table salesfact(sk1_cust int, sk2_store int, id int, val
decimal(9,2)) partition by expression execute function loadPartMart('my_acc', 'salesmart', 'o1',
'salesfact', 'p3');
partition p1 (id = 1) in rootdbs,
partition p2 (id = 2) in rootdbs, -- drop partition from the table
partition p3 (id = 3) in rootdbs; execute function dropPartMart('my_acc', 'salesmart', 'o1',
'salesfact', 'p1');
insert into cust values(1, "John Smith", 32);
alter fragment on table salesfact detach p1 salesfact_p1;
insert into cust values(2, "Joe Smith", 32);
insert into cust values(3, "John Doe", 32); -- attach a partition from the table.
create table p4(sk1_cust int, sk2_store int, id int, val
decimal(9,2), check (id = 4));
insert into store values(101, "San Jose");
insert into store values(102, "New Delhi"); alter fragment on table salesfact attach p4 as (id = 4) after
p3;
insert into store values(103, "Munich");
execute function loadPartMart('my_acc', 'salesmart', 'o1',
'salesfact', 'p4');
insert into salesfact values(1, 101, 1, 20.22);
insert into salesfact values(1, 101, 3, 80.24);
insert into salesfact values(1, 102, 2, 34.34);
insert into salesfact values(1, 103, 1, 23.28);
insert into salesfact values(1, 101, 1, 20.22);
insert into salesfact values(1, 102, 3, 80.24);
insert into salesfact values(1, 103, 2, 34.34);
26
insert into salesfact values(1, 101, 1, 23.28); #ibmiod
27. Performance
Web_sales data mart
• Fact table: web_sales with 4.1 billion rows
• Each partition has about 120 million rows
• Full Refresh: 3 hours 29 minutes: 287 GB/hour
• Dropping a partition from IWA: 46 seconds
• Loading a partition from IWA: 115 seconds
Total operation time on IWA is less than 3 minutes.
27
#ibmiod
28. Keeping Informix and IWA data in sync
Fact Table: out of sync Fact Table: out of sync
Dimensions: in-sync Dimensions: in-sync
On IWA, drop and On IWA, drop and
Reload fact partitions Reload fact partitions
Load transactions
to existing partitions Attach new partition,
in Fact table Fact Table: in-sync
Dimensions: in-sync
On IWA, drop and
First, detach the partition on IWA Reload dimension partitions
Then detach the partition on Informix
Fact Table: sync
Update customer info, Dimensions: out of sync
Intend to detach Store info, etc
Fact Table: In-sync Full Reload of the data mart can be done any time.
Dimensions: in-sync
28
#ibmiod
30. Informix Warehouse Accelerator – In 11.70.FC4
IBM Smart Analytics
Step 1. Install, configure, Studio
start Informix
Step 2. Install, configure, Step 3
start Accelerator
Step 1
Step 3. Connect Studio to
Informix & add accelerator
Step 4
Informix Database Server
Step 4. Design, validate,
Deploy Data mart
Step 5
Step 5. Load data to
accelerator
Ready for Queries
BI Applications
Step 2
Ready
Informix warehouse Accelerator
30
30 #ibmiod
31. Background
• Prior to 11.70.FC5, adding accelerator, create, deploy, load, enable, disable datamart,
accelerating queries – are all operations officially supported only on Standard server or
Primary node of MACH11/HA environment.
• We estimate about 50% of Informix customers use HDR secondary servers and growing
number of customers use MACH11 (SDS secondary) configurations and RSS nodes.
MACH11 is the Informix scale out solution.
• IWA itself supports a scale out solution (on a cluster) starting with 11.70.FC4.
• Reasons to support MACH11 and IWA together.
o This feature will enable partitioning a cluster or HA group between OLTP and BI
workload.
o This feature will give help to off-load the expensive LOAD functionality to secondary
servers
o We have customers now requesting support for HDR secondary to IWA
31
#ibmiod
32. Informix Warehouse Accelerator – 11.70.FC5. MACH11 Support
IBM Smart Analytics
Step 1. Install, configure, Studio
start Informix
Step 2. Install, configure, Step 3
Informix
start Accelerator Informix
Step 1 Informix
HDR
Informix RSS
Secondary
Step 3. Connect Studio to SDS1
Informix & add accelerator
Informix Primary SDS2 Step 4
Step 4. Design, validate,
Deploy Data mart from
Primary, SDS, HDR, RSS Step 5
Step 5. Add IWA to sqlhosts
Load data to
Accelerator from any node.
Ready for Queries
BI Applications
Step 2
Ready
Informix warehouse Accelerator
32
32 #ibmiod
33. Step 1: Install:
• Informix and IWA are installed just like before.
• Informix can be combination of standard, primary, SDS, HDR secondary and RSS
nodes.
• IWA can be installed on the same computer as any one of the nodes or on distinct
computer.
• IWA can also be installed on a cluster hardware with multiple worker nodes for
scale out performance.
33
#ibmiod
34. Step 2: Configure
• Informix and IWA are installed just like before.
• Informix can be combination of standard, primary, SDS, HDR secondary and RSS
nodes.
• IWA can be installed on the same computer as any one of the nodes or on distinct
computer.
• IWA can also be installed on a cluster hardware with multiple worker nodes for
scale out performance.
• Note: Informix MACH11 technology works with logged and ANSI databases only.
34
#ibmiod
35. Step 2: Configure
•The secondary servers should be updatable
secondary servers.
•Set this in $ONCONFIG
UPDATABLE_SECONDARY 10
35
#ibmiod
36. Step 3: Connect
• You can connect to IWA from Informix from any of the Informix servers using existing
method.
o Get the connection details via:
# ondwa getpin
o The output will be, ip address, port, pin for IWA connection.
o Use that information to create the connection.
• After successful connection from Informix to IWA, the SQLHOSTS will have something like
this
FAST group - -
c=1,a=484224232041684420473a283e612f74393e6025757159506a51344a6b4e2f2d2d47455e6b653f2f6c795f287d7b65
224d6c3c2f65722e6a2a4245397b3b447d572c3129696b306440
FAST_1 dwsoctcp 162.34.42.188 21022 g=FAST
• To use this connection on any of Informix nodes, copy these lines AS IS to the SQLHOSTS file of those servers.
• Make sure copy ALL the lines within the FAST group.
36
#ibmiod
37. Step 3: Connect...continued
• The name of the IWA will be used as the AQT site name in systables.sitename. So, it’s
important to have the right site name in SQLHOSTS entry for a successful connection.
• Changing ANY of the details of this SQLHOSTS entry will result in connection, query
matching and acceleration issues.
37
#ibmiod
38. Step 4. Design, Validate and Deploy
• The secondary servers should be updatable secondary servers.
• Set this in $ONCONFIG
UPDATABLE_SECONDARY 10
• The Design, validate and deploy would be identical.
• The partition refresh via dropPartMart() and loadPartMart() can be invoked from any
node of the MACH11 cluster.
38
#ibmiod
39. IWA: Road so far…
12.10: Now in Beta.
Sign up!
2012 IOD
2012 IIUG
11.7xC6
Partition refresh
11.7xC5 (Dimension tables)
Support TimeSeries
11.7xC4
Partition Refresh (fact)
11.7xC3 MACH11 support
11.7xC2 Solaris on Intel
SMB: IGWE
Scale out: IWA
Workload Analysis Tool on Blade Server
IWA 1 Release
st More Locales
On SMP Data Currency
39
39 #ibmiod
40. Informix Publications
Bulletin of the Technical Committee on Data Engineering: March 2012
Vol. 35 No. 1
Real Time Business Intelligence. September 2, 2011 - Seattle, United States
2012 Bloor Report: IBM Informix in hybrid workload
environments
2012 Ovum Analyst report: Informix Accelerates Analytic Integration
into OLTP
http://youtu.be/xJd8M-fbMI0
IBM Data management Magazine: Supercharging
the
data wharehouse while keeping the costs down.
DBTA Article: Empowering Business Analysts with Faster Insights
40
#ibmiod
42. Intel® Xeon® E7-8870: One Terabyte Scaling
performance.
• Hardware setup
o Intel® Xeon® E7-8870 processor – 4 socket (40C/80T) and 8 socket
(80C/160T) configurations
• 2.4 GHz, 30MB last level shared cache
o 10 TB storage
o 2 TB RAM
• Software Setup
o Informix and Informix Warehouse Accelerator: v11.70.FC5
o Both Informix and IWA on the same machine.
42
#ibmiod
43. Scaling in Westmere: Data Warehouse Setup.
• TPC-DS Schema; web_sales 73,049 66
• Mart Data size: 1 terabytes 22
• web_sales, 4.1 billion rows
360,000
o Fact with 34 partitions 86,400
4.1 billion
• Dimensions: 13, non partitioned. 1,800
3600
20
1.9 million
15 million 7,200
66
30 million
43
#ibmiod
46. INTEL/IWA: Breakthrough technologies for
performance
7. Multi-core, multi-node environment 1. Large memory support
Nehalem has 8 cores and Westmere 10 cores. This trend is 64-bit computing; System X with MAX5 supports up
expected to continue. IWA: Parallelize the scan, join, group to 6TB on a single SMP box; Up to 640GB on each
operations. Keep copies of dimensions to avoid cross-node node of blade center. IWA: Compress large dataset
synchronization. and keep it in memory; totally avoid IO.
6. Single Instruction Multiple Data
Specialized instructions for manipulating 2. Large on-chip Cache
128-bit data simultaneously. IWA: L1 cache 64KB per core, L2 cache is 256KB per
Compresses the data into deep columnar 7 1 core and L3 cache is about 4-12 MB.
fashion optimized to exploit SIMD. Used in Additional Translation lookaside buffer (TLB).
parallel predicate evaluation in scans. 6 2 IWA: New algorithms to avoid pipeline
flushing and cache hash tables in L2/L3 cache
5 3
5. Hyperthreading 4 3. Frequency Partitioning
2x logical processors; increases processor IWA: Enabler for the effective parallel access
throughput and overall performance of threaded of the compressed data for scanning.
software. IWA: Does not exploit this since the Horizontal and Vertical Partition Elimination.
software is written to avoid pipeline flushing.
4. Virtualization Performance
Lower overhead: Core micro-architecture
enhancements, EPT, VPID, and End-to-End
HW assist IWA: Helps informix and IWA to
seemlessly run and perform in virtualized
environment.
46
46 #ibmiod
47. Tick-Tock Development Model:
Sustained Microprocessor Leadership
Intel® Core™
® ™ Intel® Microarchitecture Intel® Microarchitecture Intel® Microarchitecture
Microarchitecture Codename Nehalem Codename Sandy Codename Haswell
Bridge
Merom Penryn Nehalem Westmere Sandy Ivy Haswell Future
65nm 45nm 45nm 32nm Bridge
32nm Bridge
22nm 22nm 14nm
New New New New New New New New
Micro- Process Micro- Process Micro- Process Micro- Process
architecture Technology architecture Technology architecture Technology architecture Technology
TOCK TICK TOCK TICK TOCK TICK TOCK TICK
47
#ibmiod
48. Intel Xeon Processor
® ®
Family for Business
Scalable
Enterprise
Mainstream Top-of-the-line performance,
Enterprise scalability, and reliability
Best combination of
performance, power efficiency,
and cost Mission Critical
Small
Business Enterprise Server Performance and reliability for the most
business critical workloads with outstanding
Versatility for infrastructure apps (up to 4S) economics
Economical and more Cloud Computing Cloud Computing
dependable vs. desktop Efficient, secure, and open platforms for Highest virtualization density and advanced
Internet datacenters and IAAS reliability for private cloud
Entry Servers and High Performance Computing & High Performance Computing
Workstations Workstations
More features and performance than Bandwidth-optimized for high Greater scaling and memory capacity
traditional desktop systems performance analytics & visualization
Increasing capability
48
48 #ibmiod
49. Intel® Xeon® Processor
E7-8800/4800/2800 Product Families
Building on Xeon® 7500 Leadership Capabilities
More Performance More Expandable
• 10 cores / 20 threads • Supports 32GB DDR3 DIMMs (2TB per 4-
socket system)1
• 30MB of last level cache
More Security & RAS More Efficient
E7-4800 E7-4800
SECURITY • More performance within same
• Intel® Advanced Encryption max CPU TDP as Xeon 7500
E7-4800 E7-4800 • Lower partial active & idle power
Standard-New Instructions
• Intel® Trusted Execution via Intel Intelligent Power
Technology2
Technology (TXT)
• Support for Low Voltage-DIMMs3
RELIABILITY, AVAILABILITY, SERVICEABILITY • Reduced power memory buffers4
• Enhanced DRAM Double Device Data Correction
• Fine Grained Memory Mirroring
Delivers more Performance, Expandability and RAS
Delivers more Performance,
while improving Energy Efficiency
while improving Energy Efficiency
1. Up to 64 slots per standard 4 socket system x 32GB/DIMM = 2TB
2. Uses similar core and package C6 power states enabled on Intel Xeon 5500/5600 series processors. Requires OS support.
49 3.
4.
Savings dependent on workload and configuration.
Memory buffer power savings of up to 1.3W active and 3W idle per buffer per Intel estimates. Slightly more savings when used with LV DIMMs
#ibmiod
50. Intel® Xeon® 7500/E7 8 Socket Configuration
4+4 (8S) IBM® System
x3850 X5
Up to 10 cores and 2.4 Ghz
per CPU
Support 8 socket mode by
combining 2 systems via
external QPI links
Memory Configuration
4TB in 8 socket server
6TB in 8 socket + MAX5
Continued 1066MHz
support
50
#ibmiod
51. Advanced Reliability Starts With Silicon
Intel® Xeon® processor E7 family RAS Capabilities
Memory I/O Hub CPU/Socket
• • Inter-socket Memory Mirroring
Inter-socket Memory Mirroring •• Physical IOH Hot Add
Physical IOH Hot Add • • Machine Check Architecture
•Machine Check Architecture
• Machine Check Architecture
Machine Check Architecture
• • Intel® ® Scalable Memory
Intel Scalable Memory •• OS IOH On-lining*
OS IOH On-lining* (MCA) recovery (MCA-R)
(MCA) recovery (MCA-R)
(MCA) recovery (MCA-R)
(MCA) recovery (MCA-R)
Interconnect (Intel® SMI) Lane
Interconnect (Intel® SMI) Lane •• PCI-E Hot Plug
PCI-E Hot Plug • • Corrected Machine Check
Corrected Machine Check
Failover
Failover Interrupt (CMCI)
Interrupt (CMCI)
• • Intel® ® SMI Clock Fail Over
Intel SMI Clock Fail Over • • Corrupt Data Containment
Corrupt Data Containment
• • Intel® ® SMIPacket Retry
Intel SMI Packet Retry Mode
Mode
• • Memory Address Parity
Memory Address Parity • • Viral Mode
Viral Mode
• • Failed DIMM Isolation
Failed DIMM Isolation • • OS Assisted Processor Socket
OS Assisted Processor Socket
• • Memory Board Hot Add/Remove
Memory Board Hot Add/Remove Migration*
Migration*
• • Dynamic Memory Migration*
Dynamic Memory Migration* • • OS CPU on-lining **
OS CPU on-lining
• • OS Memory On-lining **
OS Memory On-lining • • CPU Board Hot Add at QPI
CPU Board Hot Add at QPI
• • Recovery from Single DRAM
Recovery from Single DRAM • • Electronically Isolated (Static)
Electronically Isolated (Static)
Device Failure (SDDC) plus
Device Failure (SDDC) plus Partitioning
Partitioning
random bit error
random bit error • • Single Core Disable for Fault
Single Core Disable for Fault
• • Memory Thermal Throttling
Memory Thermal Throttling Resilient Boot
Resilient Boot
• • Demand and Patrol scrubbing
Demand and Patrol scrubbing
• • Fail Over from Single DRAM
Fail Over from Single DRAM Intel® QuickPath Interconnect
Device Failure (SDDC)
Device Failure (SDDC) • • Intel QPI Packet Retry
Intel QPI Packet Retry
• • Enhanced DRAM Double Device
Enhanced DRAM Double Device • • Intel QPI Protocol Protection via
Intel QPI Protocol Protection via
Data Correction
Data Correction CRC (8bit or 16bit rolling)
CRC (8bit or 16bit rolling)
• • Fine Grained Memory Mirroring
Fine Grained Memory Mirroring • • QPI Clock Fail Over
QPI Clock Fail Over
• • Memory DIMM and Rank Sparing
Memory DIMM and Rank Sparing • • QPI Self-Healing
QPI Self-Healing
• • Intra-socket Memory Mirroring
Intra-socket Memory Mirroring
• • Mirrored Memory Board Hot
Mirrored Memory Board Hot
Add/Remove
Add/Remove
Advanced reliability features work to maintain data integrity
reliability features
51
52. Intel® Xeon® E5 and E7 Family Roadmap
2012 2013/Future
Intel® Xeon® processor E7-8800/4800/2800
product families
Expandable 2-8 sockets, up to 10C/20T per socket, up to 30MB shared cache, “Westmere” microarchitecture
Future Intel®
Micro-
architecture
Intel® Xeon® processor E5-4600 product family codename
4 sockets, up to 8C/16T per sockets, up to 20MB shared cache, “Sandy Bridge” microarchitecture Ivy Bridge
4S Efficient
Performance
Intel® Xeon® processor E5-2600 product family
2 sockets, up to 8C/16T per sockets, up to 20MB shared cache, “Sandy Bridge” microarchitecture
2S Efficient
Performance
52
52
#ibmiod
53. Big Ridge* Introduction
Big Ridge Unleashes Platform and Application Performance with
Scalable,
Predictable, Efficient I/O Performance
Platform NVM Tier
Storage Tier Memory Tier Compute Tier
Big Ridge
HDD SSD Intelligent Storage Memory DRAM
Extension
Concurrently Supports NVM Direct Access and Fast Storage Usage Models
Why Big Ridge:
•NVM performance, cost/power create a significant inflection point for the platform/datacenter
•CPU/Server performance has grown significantly, storage/memory has not kept pace
•To unlock NVM potential, software optimization and new access methods are required
•Big Ridge offers new levels of platform performance, power and overall TCO improvements
•Architected for future NVM to further scale platform and application performance
•We are building an extensive ecosystem of support from OEMs, ISVs and End Users
*
Big Ridge is Intel’s codename for its first generation of application optimized NVM technology, Safford Peak is Intel’s codename for its first product
using this technology.
53
#ibmiod
54. IWA Resources
• IBM Informix Infocenter: http://ibm.co/fMcUDg
• Martin’s blog: http://ibm.co/Ts0cll
• Fred Ho’s blog: http://ibm.co/T9FaNy
• Keshav’s blog: http://ibm.co/RQXExL
Thank You
54
#ibmiod
Editor's Notes
IBM IOD 2011 10/25/12 Prensenter name here.ppt
IBM IOD 2011 10/25/12 Prensenter name here.ppt 10/25/12 18:58
IBM IOD 2011 10/25/12 Prensenter name here.ppt 10/25/12 18:58
IWA uses a technique called Frequency Partitioning. In the diagram above, once sees that the table Trade Info contains columns Volume, Product, Orgin Country. Histograms are built for each column to determine frequency of data value occurrences, as shown with Origin and Product. Then the system looks for the most frequently occuring values in each of the columns, in the example, Top 64 Traded Goods. It then encodes those values with the least number of bits that can adequately represent the data (Approximate Huffman encoding) Idea being that most accessed values will require the least number of bits to be manipulated. Now, these values are then intersected with values in other columns, Top Traded Goods from China/USA. These encoded values are then placed in memory cells across all available memory in the system used for subsequest scan operations. The next slide shows an example of this and further encoding used for IWA.
Slide Purpose: Show full systems and use as chance to highlight the Energy Efficiency enhancements in Intel® Xeon® processor E7 family The Xeon E7 family is designed and built upon Intel’s 32nm Nehalem micro-architecture, which allows us to deliver 25% more cores and cache providing more performance within same maximum TDP as the Xeon 7500 series. It also supports 16 DIMMs per socket, which equates to 2TB of memory for the 4-socket E7-4800 product family – allowing for increased expandability. The Xeon E7 family features energy efficiency technologies including the Intel® Intelligent Power Technology (IPT) which is a shared technology from Intel’s Efficient Performance product line. IPT reduces partial active and idle power in the CPU and memory. Xeon E7 also supports lower power memory as well as memory buffers which support both standard and LV-DIMMs. The Xeon processor E7 family not only includes all of the reliability, availability and serviceability (RAS) features of the previous generation such as machine check architecture-recovery but also includes additional memory error correction features such as Enhanced DRAM Double Device Data Correction (DDDC) and Fine Grained Memory Mirroring. DDDC is an improved memory RAS feature which allows for a 2nd memory error & replacement of DIMMs w/o crashing . Fine Grained Memory Mirroring provides protection against uncorrectable memory errors that would otherwise result in a platform failure and allows for more flexible memory mirroring configurations (allows memory mirroring of just a critical portion of memory, leaving the rest of memory un-mirrored). This enables more cost-effective mirroring by mirroring just the critical portion of memory versus the entire memory space. New security features such as Intel® Advanced Encryption Standard New Instructions (AES-NI) and Intel® Trusted Execution Technology (TXT) are also supported. These advanced security features within the Xeon processor E7 family work to maintain data integrity, accelerate encrypted transactions, and maximize business continuity.