I/O Design Considerations and Performance Metrics in Modern Systems

1. Chapter 8 Storage, Networks and Other Peripherals

2. Introduction • I/O Design affected by many factors (expandability, resilience) • Three characteristics for organizing I/O devices • Behavior: input (read once), output (write only), or storage • Partner: Either a human or a machine is at the other end of the I/O device • Data rate: the peak rate at which data can be transferred between the I/O device and the main memory or processor.

3. I/O Devices

4. I/O Performance • Performance: — access latency — throughput — connection between devices and the system — the memory hierarchy — the operating system • A variety of different users (e.g., banks, supercomputers, engineers)each has different requirements.

5. Typical Collection of I/O Devices

6. Importance of I/O in a Networked Society • Processors are being built from the same basic technology. • I/O becomes one of the most distinctive features of the machines. • As the importance of networking and information infrastructure grows, I/O plays an increasing important role.

7. Impact of I/O on System Performance • Suppose we have a benchmark that executes in 100 seconds of elapse time, where 90 seconds is CPU time and the rest is I/O time. If the CPU improves by 50% per year for the next five years but I/O time doesn’t improve, how much faster will our program run at the end of five years?Amdahl’s Law again!

8. Assessing I/O Performance • Depends on the application • System throughput • I/O bandwidth • how much data can we move through the system in a certain time? • How many I/O operations can we do per unit of time? • Response time

9. I/O Performance Measures • Examples from Disk and File Systems • affected by disk technology, how disk are connected, the memory, the processor, and the file system provided by the OS. • Benchmark relatively primitive compared with those for the CPU. • Note: transfer rate: 1 MB = 10^6 bytes, not 2^20 bytes • Supercomputer I/O benchmarks: dominated by access to large files on magnetic disks. Data throughput, # of bytes per second that can be transferred between a supercomputer’s main memory and disks. • Transaction Processing(TP) I/O benchmarks: • involve both response time requirement and a performance based on throughput. • Concerned with I/O rate, measured as # of disk accesses per second. • TPC has developed several benchmarks. • File System I/O benchmarks: five phases --> Makedir, Copy, ScanDir, ReadAll, Make

10. Disk Storage and Dependability • Disk storage is nonvolatile, meaning that the data remains even when power is removed. • Platters in hard disk are metal (or glass), offering several advantages over floppy disks: • can be larger because it is rigid • has higher density because it can be controlled more precisely • Has a higher data rate because it spins faster • can incorporate more platter

11. I/O Example: Disk Drives • To access data: — seek: position head over the proper track (3 to 14 ms. avg.) — rotational latency: wait for desired sector (.5 / RPM) — transfer: grab the data (one or more sectors) 30-80 MB/sec

12. Example • For a disk rotating at 5400 RPM, average rotational latency = 0.5 rotation / 5400 RPM = 0.5 rotation/(5400 RPM/ 60) = 5.6ms • For a disk rotating at 15,000 RPM, average rotational latency = 2.0ms • Note: detailed control of the disk and the transfer between the disk and the memory is usually handled by a disk controller. The controller adds the final component of disk access time, controller time. • The average time to perform an I/O operation will consist of these four times plus any wait time incurred because other processes are using the disk. • Many recent disks have included caches directly in the disk to speed up the access time.

13. Disk Read Time • What is the average time to read or write a 512-byte sector for a typical disk rotating at 10,000 RPM? The advertised average seek time is 6 ms, the transfer rate is 50MB/sec, and the control overhead is 0.2ms. (Assuming no waiting time)

14. Dependability, Reliability and Availability • A system alternating between states: • Service accomplishment: where the service is delivered as specified • Service interruption: where the service is different from the specified service • Transitions from state 1 to state 2 are caused by failures • Transitions from state 2 to state 1 are called restorations. • Reliability is a measure of the continuous service accomplishment. • Mean-time-between-failures = Mean-time-to-failure + Mean-time-to-repair • Availability = MTTF/(MTTF+MTTR)

15. How to Increase MTTF • Fault avoidance • Fault tolerance • Fault forecasting

16. RAID • Leveraging redundancy to improve the availability of disk storage is captured in the phrase: Redundant Array of Inexpensive Disks • No redundancy (RAID 0): allocation of logically sequential blocks to separate disks to allow higher performance than a single disk can deliver • Mirroring (RAID 1): writing the identical data to multiple disks to increase data availability. • Error Detecting and Correcting Code (RAID 2) • Bit-Interleaved Parity (RAID 3): Add enough redundant information to restore the lost information on a failure. • Block-interleaved Parity (RAID 4) • Distributed Block-interleaved Parity (RAID 5) • P+Q redundancy (RAID 6)

17. RAID 1-6

18. Small Write Update on Raid 3 vs. Raid 4

19. Networks • Key characteristics of typical networks: • distance: 0.01 to 10,000 kilometers • speed: 0.001 MB/sec to 1GBit/sec • topology: bus, ring, star, tree • shared lines: none (point-to-point) or shared • RS232: slow but cheap • LAN (ethernet): up to 1GBit/sec • Ethernet is a bus with multiple masters and a scheme for determining who gets bus control. • ATM: scalable network technology (155 Mbits/sec to 2.5 Gbits/sec) • Example: Performance of two networks

20. The OSI Model Layers

21. TCP/IP Packet Format

22. Performance of Two Networks • Bandwidth 100 Mbit/s vs. 1000 Mbit/s • Interconnect latency: 10us • HW latency from/to network: 2 us • SW overhead sending to network: 100 us • SW overhead receiving from network: 80 us • Question: Find the host-to-host latency for a 250 byte message using each network.

23. I/O Example: Buses • Shared communication link (one or more wires) • Difficult design: — may be bottleneck — length of the bus — number of devices — tradeoffs (buffers for higher bandwidth increases latency) — support for many different devices — cost • Types of buses: — processor-memory (short high speed, custom design) — backplane (high speed, often standardized, e.g., PCI) — I/O (lengthy, different devices, standardized, e.g., SCSI)

24. Bus: Connecting I/O Devices to Processor and Memory • A bus generally contains a set of control lines and a set of data lines. • Control lines are used to signal requests and acknowledges, and to indicate what type of information is on the data lines • Data lines carry information between the source and the destination. The information may consist of data, complex commands or addresses. • Bus transaction includes two parts: sending the address and receiving or sending the data. • Read transaction == transfers data from memory • Write transaction == writes data to the memory • Input operation: input to memory so the processor can read it • Output operation: output to device from memory

25. Input Operation

26. Different Machines using Different Types of Buses

27. Synchronous and Asynchronous Buses • Synchronous buses • use a clock and a synchronous protocol, fast and small • but every device must operate at same rate • clock skew requires the bus to be short • processor-memory buses are often synchronous • Asynchronous buses • don’t use a clock and instead use handshaking • can accommodate a wide variety of devices

28. Asynchronous Protocol • Let’s look at some examples from the text “Performance Analysis of Synchronous vs. Asynchronous” “Performance Analysis of Two Bus Schemes”