17
Computer Systems Performance Engineering in Trondheim [email protected]

ComputerSystems Performance Engineering in Trondheim/menu/standard/… · group (IDI): Sp with queuing models 1989… Network systems H/w timing & synchron‐ isation •ICL •STC

  • Upload
    others

  • View
    1

  • Download
    0

Embed Size (px)

Citation preview

Page 1: ComputerSystems Performance Engineering in Trondheim/menu/standard/… · group (IDI): Sp with queuing models 1989… Network systems H/w timing & synchron‐ isation •ICL •STC

Computer Systems Performance Engineering in Trondheim

[email protected]

Page 2: ComputerSystems Performance Engineering in Trondheim/menu/standard/… · group (IDI): Sp with queuing models 1989… Network systems H/w timing & synchron‐ isation •ICL •STC

Overview

• 1970 – 1983   Mainframes and mini computers.

• Performance measurement tools & techniques. 

• 1983 – 1995  Networked systems, multi‐layered software. Performance modelling tools and graphical workstations

• Conclusion: Industrial & academic impact

HiNC3 October 2010 2p.h.hughes IDI NTNU

Page 3: ComputerSystems Performance Engineering in Trondheim/menu/standard/… · group (IDI): Sp with queuing models 1989… Network systems H/w timing & synchron‐ isation •ICL •STC

Era Problemarea

Partners Techniques

1970..UnivacMainframesExec 8

Capacityplanning

•Univac Norway•Other computer centres

Software monitors;Benchmarks

1977.. NordMinisSINTRAN III

CPU speed;Interactiveresponse

•Norsk Data A/S  •state institutions

h/w monitors; Interactor

1980..DMS1100SIBAS 

Databaseapplications

•Bibsys group;•Norsk Data

Pilotsyntheticprototypes

1970 – 1983  Measurement‐based work

HiNC3 October 2010 3p.h.hughes IDI NTNU

Page 4: ComputerSystems Performance Engineering in Trondheim/menu/standard/… · group (IDI): Sp with queuing models 1989… Network systems H/w timing & synchron‐ isation •ICL •STC

1970...Measurement context:

HiNC3 October 2010 4p.h.hughes IDI NTNU

Page 5: ComputerSystems Performance Engineering in Trondheim/menu/standard/… · group (IDI): Sp with queuing models 1989… Network systems H/w timing & synchron‐ isation •ICL •STC

Measurement context:

Controlledsystem change

Controlledworkload

”what if…..?” questions

HiNC3 October 2010 5p.h.hughes IDI NTNU

Page 6: ComputerSystems Performance Engineering in Trondheim/menu/standard/… · group (IDI): Sp with queuing models 1989… Network systems H/w timing & synchron‐ isation •ICL •STC

1. Monitor the Production Workload2. Analyse the log files• H/w resource usage• Cluster analysis

3. Construct scaled‐down workload• Selection/tuning of programs & data• job sequence

4. Validate using software monitor• Average device utilisations: OK• Throughput: optimistic (+25%)

Benchmark as a workload model

HiNC3 October 2010 6p.h.hughes IDI NTNU

Page 7: ComputerSystems Performance Engineering in Trondheim/menu/standard/… · group (IDI): Sp with queuing models 1989… Network systems H/w timing & synchron‐ isation •ICL •STC

HiNC3 October 2010 7p.h.hughes IDI NTNU

Page 8: ComputerSystems Performance Engineering in Trondheim/menu/standard/… · group (IDI): Sp with queuing models 1989… Network systems H/w timing & synchron‐ isation •ICL •STC

1977...Interactive Computing: ‐ testing a system with many

simultaneous users

HiNC3 October 2010 8p.h.hughes IDI NTNU

Page 9: ComputerSystems Performance Engineering in Trondheim/menu/standard/… · group (IDI): Sp with queuing models 1989… Network systems H/w timing & synchron‐ isation •ICL •STC

A small database system with test‐harness

HiNC3 October 2010 9p.h.hughes IDI NTNU

Page 10: ComputerSystems Performance Engineering in Trondheim/menu/standard/… · group (IDI): Sp with queuing models 1989… Network systems H/w timing & synchron‐ isation •ICL •STC

3 stages of system tuningRe‐entrantCode

Reorganisedsystem disc

RESPONSE TIMEsec.

Interactions per second

% Utilisation

Initial System

HiNC3 October 2010 10p.h.hughes IDI NTNU

Page 11: ComputerSystems Performance Engineering in Trondheim/menu/standard/… · group (IDI): Sp with queuing models 1989… Network systems H/w timing & synchron‐ isation •ICL •STC

era Application Partners Modelling tools

1983.. System sizing •Norsk Data•ICL

•Sp Structure & Performance

Graphicwork‐stations

Informationsystems

Systemering group (IDI):

Sp with queuingmodels

1989…Network systems

H/w timing& synchron‐isation

•ICL•STC commsnetworks

DEMOS activitydiagrams, PIT processinteraction tool.

Modellingenvironment

IMSE consortium

IMSE object store & user interface. 

1983 – 1995  model – based work

HiNC3 October 2010 11p.h.hughes IDI NTNU

Page 12: ComputerSystems Performance Engineering in Trondheim/menu/standard/… · group (IDI): Sp with queuing models 1989… Network systems H/w timing & synchron‐ isation •ICL •STC

1983...Modelling in SP ‐ the Abstract Virtual Machine

operations data

Input /

output

processing

memory

storage

HiNC3 October 2010 12p.h.hughes IDI NTNU

Page 13: ComputerSystems Performance Engineering in Trondheim/menu/standard/… · group (IDI): Sp with queuing models 1989… Network systems H/w timing & synchron‐ isation •ICL •STC

SP at the hardware interface:Streaming of Video on Demand

memory

servercpu

SCSIbus

ATMnet

disk transferlogic

S‐bus

AAL transferlogic

SCSI controller

M‐bus

ATM adapterprocessor

Deliverythread

DiskThread

segmentscontainers

Legend

processingcommunications

memory

diskunits

Architecture: Telenor FoU   circa 1995

HiNC3 October 2010 13p.h.hughes IDI NTNU

Page 14: ComputerSystems Performance Engineering in Trondheim/menu/standard/… · group (IDI): Sp with queuing models 1989… Network systems H/w timing & synchron‐ isation •ICL •STC

PIT Slotted Ring Model

1989... Simulation with PIT

HiNC3 October 2010 14p.h.hughes IDI NTNU

Page 15: ComputerSystems Performance Engineering in Trondheim/menu/standard/… · group (IDI): Sp with queuing models 1989… Network systems H/w timing & synchron‐ isation •ICL •STC

IMSE schematic

Experimenter

HiNC3 October 2010 15p.h.hughes IDI NTNU

Page 16: ComputerSystems Performance Engineering in Trondheim/menu/standard/… · group (IDI): Sp with queuing models 1989… Network systems H/w timing & synchron‐ isation •ICL •STC

Conclusion – technology impact• Industry applications:  Interactor, PIT• Mutually beneficial industrial partnerships• Limited take‐up *:– Ahead of its time  (SP ? )– Made irrelevant by other developments (PILOT)– Fell foul of emerging standards (Graphical interface in SP, PIT)

• Ongoing relevance:  SP, PIT

*Ref: Three causes of technology failure Museum of computing, Boston, Mass.

HiNC3 October 2010 16p.h.hughes IDI NTNU

Page 17: ComputerSystems Performance Engineering in Trondheim/menu/standard/… · group (IDI): Sp with queuing models 1989… Network systems H/w timing & synchron‐ isation •ICL •STC

Conclusion – academic impactResearch:   contributed with experimental techniques, new paradigms and concepts

Development: innovative tools and techniques

Teaching: well‐developed pensum, good experience with research‐driven teaching.  

Future of the field of Performance Engineering: Visibility too low for prospective students. Important research challenges remain.

HiNC3 October 2010 17p.h.hughes IDI NTNU