The Design of a Meta-System for Measurement and Simulation of Time-Sharing Computers

University of Pennsylvania ScholarlyCommons

Technical Reports (CIS) Department of Computer & Information Science

February 1971

Andrew S. Noetzel University of Pennsylvania

Follow this and additional works at: https://repository.upenn.edu/cis_reports

Recommended Citation Andrew S. Noetzel, "The Design of a Meta-System for Measurement and Simulation of Time-Sharing Computers", . February 1971.

University of Pennsylvania Department of Computer and Information Science Technical Report No. MS-CIS-71-15.

This paper is posted at ScholarlyCommons. https://repository.upenn.edu/cis_reports/803 For more information, please contact [email protected]. The Design of a Meta-System for Measurement and Simulation of Time-Sharing Computers

Abstract This report presents the results of a study of a measurement-simulation technique for evaluating modifications ot time-sharing systems. The technique consists of using raw data, obtained by measurement of an operational system, as the input to a simulation model of the variant version of the system. The results of the research are reported in the form of a detailed design of a system that will perform this evaluation. It is called the Meta-System.

The Meta-System consists of three parts: event-recording mechanisms in the time sharing system that record a 'system event trace', a pre-processor of the system event trace that decomposes it into independent 'task event traces' and the simulator that accepts these as input.

The significant problem in the design of such a system is that of ensuring that the representations of the systems operation (the task event traces) are valid for use by the simulation model. Various solutions for the particular cases of timesharing system operation are provided in the design of each of the three stages of the Meta-System.

It is shown that the Meta-System may be designed to extract representations of the time-sharing systems' operation at any one of several levels. These are called the 'levels of Meta-System awareness of system operation'. The extent of the simulation model is specified by the level, as well as the class of modifications that may be evaluated in the model.

The particular Meta-System designed in this study was developed to operate with a model time-sharing system, whose features have been abstracted from those of several modern systems. These features include demand paging, multi-tasking, dynamic loading, and a virtual access method. The description of the model system is included in the report.

Keywords event analyzer, file structure, input/output simulators, measurement-simulaton, memory allocation, meta- system, random access, simulation, time sharing

Comments University of Pennsylvania Department of Computer and Information Science Technical Report No. MS- CIS-71-15.

This technical report is available at ScholarlyCommons: https://repository.upenn.edu/cis_reports/803 University of .L'eii~isylvania 'I!KE MOORE SCHOOL OF ~CTR1CA.LENGTNEEliING Philadelphia, Pennsylvania 19104

'l3CHNICA.L REPORT

TRE DESIGN OF A ME113CI-SYSTEM FOR MEASUREMENT AND Sj3WlXTION 01' TIME-SHARING COMPTJTEFiS

Andrew S. Noetzel

February 1971

Submitted to the Office of Naval Research Inf'ormation Systems Branch Arlington, Virginia

under Contract ~00014-67-A-0216-0014 Research Project NR 049-153

Reproduction in whole or in part is permitted for any purpose of the United States Government

Moore School Report No. 71-15 Security Classification I DOCUMENTCONTROLDATA- R&D 1 (Security classification of title, body of abstract and indexing annofalion must be entered when the overall report is rlnssilied) 1. ORIGINATING ACTIVITY (C~p~rat~d~thor) ZI. REPORT SECURITY CLASSIFICATION The Moore School of Electrical Engineering University of Pennsylvania UN~SSIFIED Phila., Pa. 19104 2b. GROUP I 3. REPORT TITLE I 'DIE DESIGN OF A MEE4-SYSTEM FOR MEASITREDENT AND SIMULClTION OF I TIME SHARING COMKJTERS 4 1 FSCRlPTlVE NOTES (5pe of laport and.inc1usIve dates) Technical Report I 8. AU THOR(S) (Firat name. middle inlttal. last name) Andrew S. Noetzel

6. REPORT DATE 7a. TOTAL NO. OF PAGES 76. NO. OF REFS February 1971 279 128 aa. CONTRACT OR GRANT NO. ga. ORIGINATOR'S REPORT NUMBER(SI I ~00014-67-A-0216-0014 Moore School Report No. 71-15 b. PROJECT NO.

9b. 0 THER REPORT NOIS) (Any other numbers that may be assigned this report)

10. DISTRIBUTION STATEMENT Reproduction in whole or in part is permitted for any purpose of the I U.S. Government. 11. SUPPLEMENTARY NOTES 12. SPONSORING MILITARY ACTIVITY Office of Naval Research Information Systems Branch Arlington, Virginia

13. ABSTRACT This report presents the results of a study of a measurement-simulation technique for evaluating modifications to time-sharing systems. The technique consists of using raw data, obtained by measurement of an operational system, as the input to a simulation model of the variant version of the system. The results of the research are reported in the form of a detailed design of a system that will perform this evaluation. It is called the Meta-System. The Meta-System consists of three parts: event-recording mechanisms in the time sharing system that record a 'system event trace', a pre-processor of the system event trace that decomposes it into independent 'task event traces' and the simulator that 1 accepts these as input. 'Ihe significant problem in the design of such a system is that of ensuring that 1 the representations of the systems operation (the task event traces) are valid for use by the simulation model. Various solutions for the particular cases of time- sharing system operation are provided in the design of each of the three stages of the Meta-System. It is shown that the Meta-System may be designed to extract representations of the time-sharing systems1 operation at any one of several levels. These are called the 'levels of Meta-System awareness of system operation'. The extent of the sitnula- tion model is specified by the level, as well as the class of modifications that may be evaluated in the model. I ( continued next pi@;e) I FORM 1473 (PAGE 1) DD I NOV 65 S/N 01 01 -807-681 1 Security Classification A-31408 S/N 0101-807-6821 Security Classification 4-31409 13. ABSTRACT (continued)

The particular Meta-System designed in this study was developed to operate with a model time-sharing system, whose features have been abstracted from those of several modern systems. These features include demand paging, multi-tasking, dynamic loading, and a virtual access method. 5e description of the model system is included in the report. Abstract

THE DESIGN OF A META_-SYSTEM FOR MEASLTiUDENT AND SIMJLA.TION 01" Tl.ME-:3.ARIN(; COMPUlCERS

!This report presents the ~~ecultsof a study of' a measurement- simulation technique for. cvaluatine rr~odificat ions to t i~ne-sharing systems. The technique consist:: 01' using raw data, ohtained by measurement of an operational system, a:: the input to rr simulation rnodcl of the variant version of the system. The results of the research are reported in the form of a detailed design of a system that will perform this evaluation. It is called the Meta-System.

The Meta-System cons,ists 02 three parts: event-recording mechanisms in the time-sharing system that record a 'system event trace', a preprocessor of the system event trace that decomposes it into independent 'task event traces' and the simulator that accepts these as input.

The significant problem in the design of such a system is t'rlat of ensuring that the representations of the systems operation (the task event traces) are valid for use by the simulation model. Various solutions for the particular cases of time-sharing cystem operation are provided in the design of each of the three stages of the Meta-System.

It is shown that the Meta-System may be designed. to extract representations of the time-sharing systems' opera ti or^ at any one of several levels. These are called the 'levels of Meta-:;ystern awareness of system operation'. The extent of the simulation model ic specified by the level, as well as the class of modifications that may be evaluaterl in the model.

Chapter 1: Introduction and Summary 1 The Meta-System 2

Summary of Research Reported 5

Chapter 2: Background, Motivation, and Previous Work 13

Evaluation and Measurement of Time-sharing System Performance

Criteria for Evaluating Hatch Processing Systems 13 Quantification of Criteria 15

Fffects of Evaluation in the Development of Batch Processing Systems 16

Difference Between Batch,and Time-Sharing Evaluation 18 Response Time Constraint 19

Intrinsic Problem of Time-sharing Systems 21

Sophistication of System Design 24

Isolation of User Requirements From Hardware 24

Criteria for Time- Sharing Systems 27 Measurement Techniquco

Mea ouremcnt of Utilixat iorr

Approach to thc Keu:;urcn~c.rlt l'roLlc.~n 33 Vie Meta-System 38

Chapter 3: Overview of Neta-System Desi en 39

The Measurement Function of tile I/lc.ta-Systcm 39 The Evaluation Function of the Ideta- Systern 41 TABLE OF CONTENTS (continued)

Functional Description of the Meta-System The Recording Mechanisms The Preprocessor The Simulator Levels of Meta-System Awareness of System Operation Design Problems of the Meta-System Task Identifications in the Event Pace

Representation of P~OC~ssor Time Requirements Specification of Memory Requirements Entrances and Exits to System Routines Volumes of Data Recorded: The Concept of Class of Subroutines Modeling System Routines Chapter.4: The Time-Sharing System Description The Hardware The Processor The Main Memory The Secondary Storage

I/O Channels and Devices The Generalized Operating System Resident Parts of the Operating System

The Operation of I/O Devices

Management of Channel Queues by the I/O Control Sub system

The Channel-Complete Interrupt Response ii TABLE OF' COI.JTENTS (continued) Page

The Task Initiator 87 The File System: Logical I/O 88

The File Structure 91 Address Transformation of User Files 9 3 Logical I/O Macros and Operations 97 The Virtual Access Method 104

VAM Files 105

Economization of Secondary Storage 109

The Loader 112

Hesolutiorl of Address Constants 11%

Loader Operation 11'1

Loader Operation with' the Virtual Access Method 120

The Page Loader 124

124

Higher Level ?.lacros 129 132

Chapter 5: Extracting the Event Traces 1311

lbe Hardware Event Tracc 1 36

Recording the SHET 131:

SHET ?3xample 1 33

Processor of the S1IF:T 111 3

FScnmple of Task IIard.~~rtrr:I:'vcnt 'l'rucc 150

?'he Phy:; ical Event Truce 152 TABLE OF CONTENTS (continued) Page Definition of the SPET 153 Extracting the SPET 155 Example of the SPET 156 The Processor of the SPET 160 Task Physical Event Trace 161 The Logical Event Trace 165 Definition of the SLET 166 Identification of Entry and Bit Points 170'

Recording the SIZT 175 Definition of Logical Level 176 . . Example of SLET Recording 178

Preprocessor of the SLET 181 Example of TLET 182

The Relocation Event Trace 187 Definition of the Relocation Event Trace 189 Recording the SRET 190 matuple of SRET Iiecording 190 Preprocessor of the SRET 193

'Example of the TRET 194 Chapter 6: Design of the Simulator In The Simulator 197

Input to the Simulator 198 Output of the Simulator 19 iv iI

WLF, OF CONTENTS (continued) . Page

The Basic Simulator - Overview 200

The Clockworks 201

The Event Analyzer 204

The Hardware Model 205

Model of Random Access Device Operations 208

The Models of Event Response Routines 211

Replacements of Hardware Instructions 213

Main Memory Allocation by the Model 213

Chapter 7: Examples of Meta-system Operation 219

Method of Rurming Examples 220

Iiesults from the Ac.tual Sybtem 222

Investigation of Hardware Device Change Using the Physical Event Trace

Investigation of Change of Scheduling Discipline Using the Physical Event Trace

Evaluatio~lof Alternative Logical I/O Routines Using the Logical Level Ideta-system

Evaluation of More Complex Scheduling Disciplines Using the Logical Level Meta-system

L-aluat ion or Sy stan Idcmorjr I4unuyemcnt Usil~~r,the Helocut ion IWer~t'I'ruce

Chapter 8: Results and Concluu~ons

Uniqueness of the Work

Future Research

~~bendkA Location of Event Recorcli~rg14echanisms

Appendix B Glossary v LIST OF FICURES Page

Figure 1-1 The Meta-system 2

1-2 Conceptuslization of Meta-system Level 2

Figure 2-1 Response Time as a Function of Utilization 20

2-2 Isolation of User Requirements from Hardware 26

2- 3 '&yes of Meta- systems 35

Figure 3-1 The Meta-system ' 43 3-2 Meta-system Levels due to Subroutine Structure 60

Figure 4-1 Resident Parts 02 -the Operating System

4-2 Page Fault Interrupt Response Routine 74

4- 3 Paging wnnel- complete Interrupt Response Routine 75

4-4 Physical I/O Initiation 79

4-5 Examples of Path Sets

4-6 I/O Channel Complete Interrupt Response 85

4-7 I/O Wait and Terminal Iiesponse Wait Macros 89

4-8 Terminal I/O Complete Interrupt Response 93

4-9 Two Level Directory of Index Sequential File 92

4-10 Example of Generalized File 9'+ 4-11 Abstract Struc1;ure of Syscat 94 4-12 Structure of Syscat 96 4-13 Logical 1/0-&t 99

4-14 Open Macro 101

4-15 physical Operations in File Open and Get 102

vi ' LIST OF FICUIiES (continued) Page 4-16 Vam Get Macro 108

4-17 Operation of Vam Get 110 4-18 Address-Specifying Information in Object Modules 116 4-19 Resolution of External Symbols 119 4-20 Program Load Operation 122

4-21 Physical Events in the Load Operation 123 4-22 The Page Loader 125

4-23 The Create Macro ' 128

4-24 Task Wait and Post Macros 130

Figure 5-1 Preprocessor of SliET 149

5-2 Preprocessor of SPET 162,163 5-3 Interface Choices for SLET 167 5-4 Subroutine -its 173

. 5-5 Recording LI/O Routines 177 5 &6 Preprocessor of SUT 183,1.84

5-7 Interface for Relocation Event Trace 188 Figure 6-1 Outline of Simulator 201

6-2 The Clochrorks 6-3 The Event Analysis Routine

6-4 Subroutines for :;.lmulal;cd IIar'dwar~ Ir~~I;ru.ctlon~

6-5 Software Model o:' P1/0 Iiequccl; Ikcro

6-6 Pa3c 13c-use Times in Tack Trace

6--( Modification of f.?etrtoryI)emr~r~d by tile :;irnulutor LIST OF FIGURES (continued)

Page

Figure 7-1 New Open 227

7-2 New Get 227 LIST OF 24BL;ES

Table 5-1 Example of SKET

5-2 Extracted Events of Tssk 1 5-3 'MET for Task 1 5-4 Esrample of SPET 5-5 TPET for Task 1 5-6 Example of SLET 5-7 TLZT for Task 1 .

5-8 Example of SRET 5-9 TR3T for Tssk 1 Table 7-1 Utilization Figures for the Ac-t;ua1 System 223

7-2 Results of Simulation & With Fast ~a~i& Device 224

7-3 Simulation Run With F'CFS Sc2ieduling 225

7-11 Simulatiorl Run Mith IJew Logical I/O Routines 227

7-5 Sirnu1ai;ion liun With Co~np1.e~Cchcduling Dicciplir~e 223 c~IAP~~II1

INTI{03XIC!I!.L ON AND ~JJMMAftY

The design and irnplementat ion of multipro~rummcd,time chnrin~ computer systems contirlues long after the system is put into use. A tool is needed that will measure and evaluate the computer system while it is in operation, as an aid to further development or optimiza- tion for a particular usage. The design of such a tool is described in detail in this report. It is called the Meta-system.

The uniqueness of the Meta-system is due to the coalescing of two widely used techniques - on-linc measurement, and simulation - into one system. Measurement is performed by extracting raw reprecentations of a computer system's operation (from that systcm) using software techniques only. Evaluation of the system is based on input of the measured performance characteristics to a simulation model that exer-

cises modified hardware-software versions of the system. A11 the poten- tial modifications to the system are evaluated in the context of the task load of the system, as extracted from the operational system.

This report presents the design of the system. When implemented, the Meta-system will have a wide range of applicability. System design-

ers will be able to optimize cystcm peri'omncc by measurirle the: cffcsts

of 'modificationo to hardware dcviccs, system corlrigurat iorlc;, :;oL't;tirtr.e

residences, ar~dopcrat i11g sy stern roul; 1 ncs. Usob:: of tilc: zy:: tern will

obtain benc1lnnr.k compari:;on:; l'or. vrrrmi ou:: corlrig~cr*ationr:trrltl :.cl~cduli IIC: policies.

-1- The Meta-system

The Meta-system consists of a set of routines which extract measurements from an operational time-sharing system, and a ~imulation model of that system. The raw data is extracted from the oy~temin the form of a laystem event trace1, A preprocessor decomposes this event trace into event traces that represent the operation of each task with- in the system. l%ese 'task event traces' become the input for the simulation model of the variant version of the system. Together with the system designers, the Meta-system represents the measurement and evaluation loop that results in design improvements to the system. The Meta-system loop is shown in Figure 1-1.

The detailed design of the Meta-cystem, that is'presented in this report and the feasibility of its implementation that may be inferred from the design, is the first impartant result of this research. The second result is the specification of means and procedures for obtaining performance measurements of one system that will be useful in the simulation of some other system. This result is the generaliza- tion of the detailed design, and could be the basis for a theory of execution-simulation of computer systems. A bricf d.iscusaion of the

Meta-system technique is given before the resulto are stated more explicitly. First, conoider the meusurcmcrits of one system that arc useful for simulating another system. Thq meaou~cmcnl;~rnuct not be the final results, such as the utilization factors of various hardware components,

since these will be obtained from the simulation model. Rather, they .