Complexity in Computer Systems Design

1. CGS 3863 Operating Systems Spring 2013 Dan C. Marinescu Office: HEC 439 B Office hours: M-Wd 11:30 AM – 1 PM

2. Last time: Discussion of the paper “Architecture of Complexity” by H. A. Simon Modularity, Abstractions, Layering, Hierarchy Today: Complexity of Computer Systems Resource Sharing and Complexity Bandwidth and Latency Iteration Names and The Basic Abstractions Memory. Critical Properties of Memory Systems. Read-Write Coherence Before or After Atomicity. RAID Next time Discussion of the paper “Hints for Computer Systems Design” by Butler Lamson. Lecture 4 – Thursday September 2, 2010 Lecture 4

3. Names and fundamental abstractions • The fundamental abstractions • Storage  mem, disk, data struct, File Systems, disk arrays • Interpreters cpu, programming language e.g. java VM • Communication wire, Ethernet rely on names. • Naming: • Flat • Hierarchical Lecture 4

4. Computers a distinct species of complex systems • The complexity of computer systems not limited by the laws of physics  distant bounds on composition • Digital systems are noise-free. • The hardware is controlled by software • The rate of change unprecedented • The cost of digital hardware has dropped in average 30% per year for the past 35 years Lecture 4

5. Analog, digital, and hybrid systems • Analog systems: • the noise from individual components accumulate and • the number of components is limited • Digital systems: • are noise-free • the number of components is not limited • regeneration  restoration of digital signal levels • static discipline the range of the analog values a device accepts for each input digital value should be wider than the range of analog output values • digital components could fail but big mistakes are easier to detect than small ones!! • Hybrid systems  e.g., quantum computers and quantum communication systems Lecture 4

6. Computers are controlled by software • Composition of hardware limited by laws of physics. • Composition of software is not physically constrained; • software packages of 107 lines of code exist • Abstractions hide the implementation beneath module interfaces and allow the • creation of complex software • modification of the modules • Abstractions can be leaky. Example, representation of integers, floating point numbers. Lecture 4

7. Exponential growth of computers • Unprecedented: • when a system is ready to be released it may already be obsolete. • when one of the parameters of a system changes by a factor of • 2 other components must be drastically altered due to the incommensurate scaling. • 10  the systems must be redesigned; E.g.; balance CPU, memory, and I/O bandwidth; • does not give pause to developers • to learn lessons from existing systems • find and correct all errors • negatively affects “human engineering”  ability to build reliable and user-friendly systems • the legal and social frameworks are not ready Lecture 4

8. Coping with complexity of computer systems • Modularity, abstraction, layering, and hierarchy are necessary but not sufficient. • An additional technique  iteration • Iteration • Design increasingly more complex functionality in the system • Test the system at each stage of the iteration to convince yourself that the design is sound • Easier to make changes during the design process Lecture 4

9. Iteration – design principles • Make it easy to change • the simplest version must accommodate all changes required by successive versions • do not deviate from the original design rationale • think carefully about modularity  it is very hard to change it. • Take small steps; rebuild the system every day, to discover design flaws and errors. Ask others to test it. • Don’t rush to implementation. Think hard before starting to program. • Use feedback judiciously • use alpha and beta versions • do not be overconfident from an early success • Study failures  understand that complex systems fail for complex reasons. Lecture 4

10. Curbing complexity • In absence of physical laws curb the complexity by good judgment. Easier said than done because: • tempted to add new features than in the previous generation • competitors have already incorporated the new features • the features seem easy to implement • the technology has improved • human behavior: arrogance, pride, overconfidence… Lecture 4

11. Critical elements of information revolution! Lecture 4

12. Lecture 4

13. The relation between homo sapiens and the computers • The feelings of the homo sapiens: • Hate • Frustration • Lack of understanding • The Operating System • A program to “domesticate” the computer. • Transforms a “bare machine” into a “user machine” • Controls and facilitates access to computing resources; optimizes the use of resources. • The relation went through several stages: • Many-to-one • One-to-one • Many-to-many • Peer-to-peer Lecture 4

14. Resource sharing and complexity • A main function of the OS is resource sharing. • Sharing computer resources went through several stages with different levels of complexity: • Many-to-one • One-to-one • Many-to-many • Peer-to-peer Lecture 4

15. Lecture 4

16. Lecture 4

17. Names and the three basic abstractions • Memory  stores named objects • write(name, value) value  READ(name) • file system: /dcm/classes/Fall09/Lectures/Lecture5.ppt • Interpreters  manipulates named objects • machine instructions ADD R1,R2 • modules  Variables call sort(table) • Communication Links connect named objects • HTTP protocol used by the Web and file systems Host: boticelli.cs.ucf.edu put /dcm/classes/Fall09/Lectures/Lecture5.ppt get /dcm/classes/Fall09/Lectures/Lecture5.ppt Lecture 4 17

18. Latency and Bandwidth Lecture 4 • Important concepts for physical characterization. • Applies to all three abstractions. • Informal • Bandwidth  number of operations per second! • Latency  to get there • The bandwidth of the CPU, Memory, and I/O sbsystems must be balanced. 18

19. Lecture 4 19

20. Memory • Hardware memory: • Devices • RAM (Random Access Memory) chip • Flash memory  non-volatile memory that can be erased and reprogrammed • Magnetic tape • Magnetic Disk • CD and DVD • Systems • RAID • File systems • DBMS (Data Base management Systems) Lecture 4 20

21. Attributes of the storage medium/system • Durability  the time it remembers • Stability  whether or not the data is changed during the storage • Persistence  property of data storage system, it keeps trying to preserve the data Lecture 4 21

22. Critical properties of a storage medium/system • Read/Write Coherence  the result of a READ of a memory cell should be the same as the most recent WRITE to that cell. • Before-or-after atomicity the result of every READ or WRITE is as if that READ or WRITE occurred either completely before or completely after any other READ or WRITE Lecture 4 22

23. Lecture 4 23

24. Why it is hard to guarantee the critical properties? • Concurrency  multiple threads could READ/WRITE to the same cell • Remote storage  The delay to reach the physical storage may not guarantee FIFO operation • Optimizations  data may be buffered to increase I/O efficiency • Cell size may be different than data size  data may be written to multiple cells. • Replicated storage  difficult to maintain consistency. Lecture 4 24

25. Access type; access time • Sequential access • Tapes • CD/DVD • Random access devices • Disk • Seek • Search time • Read/Write time • RAM Lecture 4 25

26. Physical memory organization Lecture 4 • RAM  two dimensional array. To select a flip-flop provide the x and y coordinates. • Tapes  blocks of a given length and gaps (special combination of bits. • Disk: • Multiple platters • Cylinders correspond to a particular position of the moving arm • Track  circular pattern of bits on a given platter and cylinder • Record  multiple records on a track 26

27. Names and physical addresses Figure 2.2 from textrbook Lecture 4 Location addressed memory  the hardware maps the physical coordinates to consecutive integers, addresses Associative memory  unrestricted mapping; the hardware does not impose any constraints in mapping the physical coordinates 27

28. RAID – Redundant Array of Inexpensive Disks Lecture 4 • The abstraction put to work to increase performance and durability. • Raid 0  allows concurrent reading and writing. Increases performance but does not improve reliability. • Raid 1  increases durability by replication the block of data on multiple disks (mirroring) • Raid 2  Disks are synchronized and striped in very small stripes, often in single bytes/words. • Error correction calculated across corresponding bits on disks, and is stored on multiple parity disks (Hamming codes). 28

29. Lecture 4 29

30. RAID (cont’d) Lecture 4 Raid 3 Striped set with dedicated parity or bit interleaved parity or byte level parity. Raid 4  improves reliability, it adds error correction 30

31. RAID (cont’d) Lecture 4 • Raid 5  striped disks with parity combines three or more disks to protect data against loss of any one disk. • The storage capacity of the array is reduced by one disk • Raid 6  striped disks with dual parity combines four or more disks to protect against loss of any two disks.  • Makes larger RAID groups more practical. • Large-capacity drives lengthen the time needed to recover from the failure of a single drive. Single parity RAID levels are vulnerable to data loss until the failed drive is rebuilt: the larger the drive, the longer the rebuild will take. Dual parity gives time to rebuild the array without the data being at risk if a (single) additional drive fails before the rebuild is complete. 31

32. Figure 2.3 from textbook Lecture 4 32