sap hana - uk & ireland sap users group · pdf filecrm ecc bw computers tablets smart...
TRANSCRIPT
SAP HANA
Lesson learned on production implementations
November 2014
Introduction
Why HANA?
Sizing
Lessons learned from customer
case-study
Wrap-up
Questions
CRM
ECC
BWCOMPUTERS
TABLETS
SMART PHONES
HANAA
SAP HANA PlatformROWS COLUMNS
CONSUMERS
(ODATA,HTML5…)
XS ENGINE
BO TOOLS
VIEWS
Why customers choose HANA
SIZING
SAP produced clear method for sizing SAP HANA based environments:
• decision tree,
• SAP Notes,
• scripts,
• formulas ,
• Hana Quicksizer tool.
This still raises some questions:
• any pitfalls?
• what is really sized?
• are optimized objects from the SAP BW application taken into account?
• what are the results?
The path to the box
S
I
Z
I
N
G
B
A
S
I
S
Initial Migration
X = n
Y = mZ = µ
Quick
Sizer
Result
Sizing
Report
Since sept 2014
SIZING
Sizing tools - results & remarks
SAPS TOT RAM TOT SAPS DB RAM DB SAPS AS RAM AS DBSIZE
HANA 5 ans 330 000 35 694 592 75 000 35 221 504 250 000 474 112 28 646
HANA 2 ans 330 000 35 567 616 75 000 35 095 562 250 000 473 088 28 561
NON HANA 280 000 527 360 25 000 53 248 250 000 474 112 57 267
SAPS TOT RAM TOT SAPS DB RAM DB SAPS AS RAM AS DBSIZE
HANA 5 ans 29 000 47 608 832 5 800 47 546 368 23 000 62 464 38 677
HANA 2 ans 29 000 19 030 016 5 800 18 968 576 23 000 61 440 15 437
NON HANA 25 000 68 608 2 000 6 144 23 000 62 464 77 328
User based sizing (50 000 High activity users)
Throughput sizing (Year: 100 Mio Obj. @ 100 items / Peak 1 hour: 100 KObj. @ 100 items)ECC FI
ECC FI
(Unités RAM: MB / DBSIZE: GB)
SAPS TOT RAM TOT SAPS DB RAM DB SAPS AS RAM AS DBSIZE
HANA 98 000 4 210 688 N/A 4 034 560 98 000 177 152 4 022
Non Hana 160 000 221 184 31 000 44 032 130 000 178 176 23 966
Initial sizing w/o planning functionsBW
• AS sizing result is the same for both Quicksizers
• => Hana-optimized transactions not taken to account
• Data retention time has a huge impact on the RAM sizing result
Qu
icks
izer
Scri
pts
• A couple of remarks:• A lot of updates (plenty of versions producing different results)• First sizing results were incorrect• => The results have to be “reworked” to allow margin of error.
SIZING
Application server sizing
Scenario:• Wall at 30 m• 3 balls• 1 x 54 years old player• Health: OK
Migration to HANA
withDB time / 3
Scenario:• Wall at 10 m• 3 balls• 1 x 54 years old player• Heart attack
CPU’ssaturated
=> SAP stuck
Hardware options: beef up application servers:• Add CPUs (Several 54 years old players)• Change CPU technology (replace with one 18 year old player)
! The only available option for user activity!
Customize: limit activity :• Decrease batch parallelization. Ex:
• Variant configuration• Settings in OMIQ (MRP)
• Limit the RFCs• 7.3x tune rfc_min_wait_dia_wp• 7.4x tune rdisp/scheduler/*/max_quota
Remark: No matter who is pushing the wall…. The problem is the same with Exadata..
SIZING
Application server sizing
AS DB
Processing time :• Total elapse : 1:00 • AS Time: 50%• DB Time: 50% • Think time: 0% !
Three users begin to enter 100 billing
documents at 10:26
Scenario:
1000 SAPS
1000 SAPS AS in 1 hour
Migration to HANA
withDB Time / 3
Three users (now powered by HANA)
begin to enter 100 billing documents at
10:26
Scenario:
Processing time :• Total elapse : 00:40• AS Time: 75%• DB Time: 25% • Think time: 0% !
1000 SAPS
1000 SAPS AS in 40 minutes
SIZING
Customer caseCUSTOMER CASE
– Business
• Sporting goods retail chains
• Sporting goods manufacturing
• Present in 21 countries
• 360 stores
• Turnover: $9.38 billion
– SAP Landscapes
• Retail: 2 ECC + 1 BW
• HR: 1 ECC + 1 BW
BW landscape
CUSTOMER CASE
– Usage
• Retail reporting on BEx
• Users : 3,000 Named / 1,500 Connected every Monday morning
OS DB MIGRATION
UPGRADE
+
SOU
RC
E
TAR
GET
6 Tb 1.2 Tb
BW landscape
CUSTOMER CASE
– Landscape configuration
• Production on dedicated box (scale out)
• DR, Quality and Development on secondary site
Production DR
QA
DEV
SECONDARY SITEPRIMARY SITE
STORAGE REPLICATION (GPFS)
– Implementation planning
BW landscape
CUSTOMER CASE
Kick OffDevelopment
MigrationFunctional
ProjectProduction Migration
User Acceptance
Production (new)
HA
NA
Q3 2013Q2 2013Q1 2013 Q4 2013 Q1 2014 Q2 2014
Production
OR
AC
LE
Dual Run
Licenses acquisition
Switch to new HANA Production
ECC landscape
CUSTOMER CASE
– Usage
• Module : Retail
• Users : 1,600 named / 1,000 connected each month
• Activity : 270 million SAP steps / month
OS DB MIGRATION
UPGRADE
+
SOU
RC
E
TAR
GET
5 .7 Tb 1.2 Tb
SLO for Production
ECC Landscape
CUSTOMER CASE
– Landscape configuration
SECONDARY SITEPRIMARY SITE
DEV
Production
Production
DEV
DEV
DEV
DEV
DEV
QA
QA
QA
QA
QA
QA
DR
DEV
DEV
QA
QA
QA
Pre-
Prod
HANA REPLICATION
DRPre-
Prod
– Implementation planning
ECC Landscape
CUSTOMER CASE
KickOff Landscape
Upgrade
Functional Project User Acceptance
Production (new)
HA
NA
Q3 2013Q2 2013 Q4 2013 Q1 2014 Q2 2014
Production
OR
AC
LE
24 hours downtime
Switch to new HANA Production
Q3 2014
Technical Acceptance
Stress test
Downtime:24 Hours
– Incidents, disruption and downtime
week week week week week week week week week week week
Post Go Live
CUSTOMER CASE
13 14 15 16 17 18 19 20 21 22 23GO LIVE
Inci
den
ts
Disruption:49 Hours
– Root cause breakdown
Post Go Live
CUSTOMER CASE
HANA Error
Human Error
Custom Queries
Hardware
ECC & 3rd Party
– Root cause breakdown
Post Go Live
CUSTOMER CASE
Custom Queries
Complex selection criteria
Symptom: high CPU consumption
Impact: system unavailable
Solution: limit usable CPU per query
Across large tables
Symptom: high memory consumption
Impact: memory overflow / database crash
Solution: limit usable memory (SP8)
Extensive testing is highly recommended beforethe Go Live to define relevant CPU and memorylimits.
– Root cause breakdown
Post Go Live
CUSTOMER CASE
AS/DB SAPS ratio changes
Symptom: high CPU Usage on ECC
Solution: increase available SAPS on the
Application side
Schedulers
Symptom: workload increased
Impact: application less stable
Solution: increase resources for the scheduler
With HANA there’s no database bottleneck Allconnected systems must interact faster Thewhole HANA ecosystem must be stress-tested.
ECC & 3rd Party
– Root cause breakdown
Post Go Live
CUSTOMER CASE
Hardware stability
Symptom: hardware failures, firmware and
driver issues
HANA has a low tolerance for hardware latencyand hardware failures.Every firmware, driver and OS patch must bevalidated through OSS support.
Hardware
SUPPORT PORTAL
– SAP: single point of contact
Communication channel
COMMUNICATION
ABAP
HARDWARE
O.S
HANA
Technical Layers Support Centers
Application Support
HANASupport
– Root cause breakdown
Post Go Live
CUSTOMER CASE
HANA errors / stability
Hana is still a young product but SAP is
improving it continuously:
SP6: Q1 2013 - 16 revisions
SP7: Q3 2013 - 8 revisions
SP8 : Q2 2014 - 3 revisions
Hana revisions are delivered every 2 or 3 months by SAP.It is recommended to perform 2 updates per year.
HANA Error
– Support pack update
REVISION
Update strategy
RECOMENDATIONS
SP 7
SP 8
REV 70 REV 71 REV 72 REV 73 REV 74
REV 80 REV 81 … REV 8x
REV 74.01 … REV 74.0x
DSP(Datacenter Service Point)
Support: 3 months
REVISION MAINTENANCE REVISION
– Support pack update
REVISION
Update strategy
RECOMENDATIONS
SP 7
SP 8
REV 70 REV 71 REV 72 REV 73 REV 74
REV 80 REV 81 … REV 8x
REV 74.01 … REV 74.0x
DSP(Datacenter Service Point)
Support: 3 months
REVISION MAINTENANCE REVISION
– BW performance comparisons
HANA performance
PERFORMANCES
PERFORMANCE RATIO MINIMUM MAXIMUM AVERAGE
Non-Optimized DATA LOADS 0.33 13 1.06
OPTIMIZED DATA LOADS 1.28 44 3.3
QUERIES 1 31 2.3
– Custom programs
• re-write queries
• re-engineer code
– BW: to optimize fully
• transfer rules must be converted from 3.x to 7.x (DTP)
• info-cubes must be converted to a “Hana Optimized" mode
• transformation finder: detect info-providers with duplicated data
• routine analyzer: look for optimization point in routine code and other process-
chain steps
– Table partitioning
• list the most demanding queries on biggest tables
• adapt partitioning keys regarding the queries
Functional work
PERFORMANCES
Goal: transfer the workload from
the ABAP stack to the Database
– Notes on sizing
• 1514966 - SAP HANA 1.0: Sizing SAP In-Memory Database
• 1872170 - Suite on HANA memory sizing report
• 1736976 - Sizing Report for BW on HANA
– Notes on table partitioning
• 2012533 – Not yet released
– Upgrade and maintenance strategy
• 2021789 - SAP HANA Revision and Maintenance Strategy
• 1600929 - SAP BW powered by SAP HANA DB: Information
• 1892593 - Preparing Support Services for SAP HANA Scenarios
• 1948334 - SAP HANA Database Update Paths for Maintenance Revisions
Wrap-up and additional information
WRAP-UP
Thank you for your participation!Any questions?