RPI The Relative Performance Index What is it?

1
RPIThe Relative Performance IndexWhat is it?
Mark Simmons, B.Sc(Hons), DIS, MBCS Chief Systems
Architect Fujitsu Computer Systems
2
Contents

• Definition of RPI and Standard Benchmarks
• RPI and OLTP TPM
• Standard benchmark results
• Standard benchmark relevance
• Benchmark activities in Linux
• RPI Load Categorization
• Benchmark Profiles
• Application Profiles
• RPI Values
• What is being Benchmarked on a Benchmark
• Benchmark Categories
• Testing Ranges
• Fujitsus TPM Number vs. TPC-C
• RPI Tool Demonstration
• Application Services
• AS Tool Demonstration

3
RPI and Standard Benchmarks

• RPI and OLTP TPM are -
• Internal use only
• Attempting to level set the server playing
  field
• Designed for system positioning and accurate
  system sizing
• First based on estimates, later adjustments are
  made if necessary
• Also provided for competitive/competitors
  products, though these are harder to verify
  fully

4
RPI and Standard Benchmarks

• Standard benchmark results are usually for-
• Sales and marketing, and for public awareness
• Analysts, press and media
• Designed for customer confidence
• Values are available on public domain web pages
• Underlines validity of internal sizing figures
• Referenced under diverse Web Pages such as those
  at- http//www.spec.org and http//www.tpc.org
• and diverse web pages of the analyst community

5
Standard Benchmarks - Relevance and test focus
6
Benchmark Activities of major vendors ( gt8 CPU
Cores Jan 03gtFeb04) Quantity
7
What is being Benchmarked when performing a
Benchmark?
Testing Range
Application and Benchmark Kit
Operating system, Compiler, Libraries
CPU
Cache
Memory
Multi- CPU
Disk- IO
DBMS
LAN
CPU
SPEC int/fp 2000
SPEC int/fp 2000 Rate
CPUs
Java VM
SPECjbb2000
Benchmark Category
SPECweb99
Webserver
TPC-C
OLTP
TPC-H, TPC-R
DSS
SAP
ERP
Oracle Applications
ERP
8
RPI - Load Categorization
Benchmark Profile
SPEC
TPC-C/H
OLTP
Application Profile

• simple - medium transactions
• high number of operations
• simple- medium operations
• short response time
• heavy use of physical I/O

no system calls little disk I/O one/few processes
well-balanced system load
Work Type
CPU Centric
I/O Centric
tpsA SAP tpsB Oracle Apps tpmC
SPECint92/95/2000 SPECrate_int92/95/2000
RPI Tool Category to use for sizing
9
Fujitsus Transactions Per MinuteRPI TPM vs.
TPC-C / tpmC

• Fujitsus measurement for OLTP performance
  (throughput of transactions per minute) in
  commercial environments is-
• based on an Oracle 9i(not RAC) database profile
• available for all vendors current systems and all
  configurations
• provides similar load characteristics as TPC-C
  and its tpmC results
• is not standardized, other competitors use
  similar absolutes or relative (e.g. mvalues
  (Sun) / rPERF (IBM) ) figures and the related
  database is not defined
• TPCs rules strongly restrict the usage of those
  proprietary values
• only with signed Non Disclosure Agreement(NDA)
• not in public disclosures
• no comparisons with published tpmC figures

10
RPI Tool Demonstration
11
RPI The Disadvantage

• RPI as a metric is not as easily applicable to
  the LINUX space as it is in the proprietary Unix
  space because-
• IA-32 is totally commoditized
• There is massive diversity in IA platforms
• Plethora of product offerings with very few
  constants
• I/O bandwidth delivery is highly variable
• Linux Kernels are generic in nature
• Systems running IA32/Linux differ widely in
  performance delivery
• Linux typically degrades faster than other Unix
  variants in scalability the heavier the load that
  is placed upon it
• Its Heavy Lifting capability is not yet mature
  !

12
The Solution is an Application Level Service

• Application Service is an abstract concept of an
  Application Unit of Work that must-
• Concurrently support multiple Linux application
  processes
• Oracle
• SAP
• Directory Services
• Web Services
• Monitor the quality of each application service
  and automatically adjusts capacity to meet
  businesss service level goals.
• Operate autonomously, with little or no
  intervention
• Make Linux based hard- and software resources
  shift automatically to meet changing demand.
• Allow dynamic capacity scaling mitigating
  impact of disparate hardware
• Offer users a self-management capability to
  assist in the administration of dramatically
  varying environments.

13
Application ServicesAdministrative Features

• Users must be able to specify application service
  priorities.
• Administrators should be able to change
  priorities dynamically, redirecting a higher
  service level toward other applications as
  business requirements change.
• Users must be able to
• Set min-max ranges or high/low water marks
• Directly influence the number of instances of an
  application that are available for use.
• Users can define Quality of Service (QoS) metrics
  based upon
• CPU load
• Network load.
• Application Services must automatically
  distribute the load using user-selected
  algorithms.
• Provide load balancing algorithms
• Balance servers for a given application
• Balance across all servers in total.

14
Application ServicesOperational Features

• Dynamically provisions servers with the resources
  necessary to support applications such as
• Dynamically adjusting the number of online
  application instances.
• Meeting defined Quality of Service (QoS) service
  levels.
• Provision applications and operating systems or
  both
• Automatically bring servers online for extra
  application support
• Re-provision servers used by lower-priority
  applications.
• Provision net-new servers that are added to the
  pool
• Takes excess systems offline if service levels
  are being met.
• Reduce license charges (must be negotiated with
  ISV!)
• Leaves offline servers in warm state
• Reduce the time to re-provision the server for
  same applications
• Reduce the time to re-provision for same
  operating system

15
Application ServicesCustomer Benefits

• Lowers total cost of ownership
• Server consolidation alone can reduce hardware
  and software costs
• Reduces the need to over-spec server capacity
• Improved utilization decreases costs and
  increases ROI
• Improves productivity
• Automated solution reduces manual administration
• Time spent in training
• Correcting human errors.
• Load balancing can optimize QoS
• Service level monitoring and capacity
  provisioning to meet needs as workloads change.
• Improved application response times
• Improves application availability
• Continuous availability of application services
  avoids costly downtime

16
Application Services Tool Demonstration