I/O Devices Characterization: Dependability, Performance, and Storage Technologies - Prof., Study notes of Computer Architecture and Organization

Download I/O Devices Characterization: Dependability, Performance, and Storage Technologies - Prof. and more Study notes Computer Architecture and Organization in PDF only on Docsity! Chapter 6 Storage and Other I/O Topics Chapter 6 — Storage and Other I/O Topics — 2 Introduction  I/O devices can be characterized by  Behaviour: input, output, storage  Partner: human or machine  Data rate: bytes/sec, transfers/sec  Example: keyboard, sound output, network, magnetic tape  I/O bus connections § 6 .1 In tro d u c tio n Chapter 6 — Storage and Other I/O Topics — 5 Dependability Measures  Reliability: mean time to failure (MTTF)  Service interruption: mean time to repair (MTTR)  Mean time between failures  MTBF = MTTF + MTTR  Availability = MTTF / (MTTF + MTTR)  Improving Availability  Increase MTTF: fault avoidance, fault tolerance, fault forecasting  Reduce MTTR: improved tools and processes for diagnosis and repair Chapter 6 — Storage and Other I/O Topics — 6 Disk Storage  Nonvolatile, rotating magnetic storage § 6 .3 D is k S to ra g e Disk heads for each surface are connected together and move in conjunction Cylinder: all the tracks under the heads at a given point on all surfaces Chapter 6 — Storage and Other I/O Topics — 7 Disk Sectors and Access  Each sector records  Sector ID  Data (typically 512 bytes, 4096 bytes proposed)  Error correcting code (ECC)  Used to hide defects and recording errors  Synchronization fields and gaps  Access to a sector involves  Queuing delay if other accesses are pending  Seek: move the heads (position head over proper track)  Rotational latency  Data transfer  Controller overhead Chapter 6 — Storage and Other I/O Topics — 10 Flash Storage  Nonvolatile semiconductor storage  100× – 1000× faster than disk  Smaller, lower power, more robust  But more $/GB (between disk and DRAM) § 6 .4 F la s h S to ra g e Chapter 6 — Storage and Other I/O Topics — 11 Flash Types  NOR flash: bit cell like a NOR gate  Random read/write access  Used for instruction memory in embedded systems  NAND flash: bit cell like a NAND gate  Denser (bits/area), but block-at-a-time access  Cheaper per GB  Used for USB keys, media storage, …  Flash bits wears out after 1000’s of accesses  Not suitable for direct RAM or disk replacement  Wear leveling: remap data to less used blocks Chapter 6 — Storage and Other I/O Topics — 12 Interconnecting Components  Need interconnections between  CPU, memory, I/O controllers  Bus: shared communication channel  Parallel set of wires for data and synchronization of data transfer  Can become a bottleneck  Performance limited by physical factors  Wire length, number of connections  More recent alternative: high-speed serial connections with switches  Like networks § 6 .5 C o n n e c tin g P ro c e s s o rs , M e m o ry, a n d I/O D e v ic e s Chapter 6 — Storage and Other I/O Topics — 15 I/O Bus Examples Firewire USB 2.0 PCI Express Serial ATA Serial Attached SCSI Intended use External External Internal Internal External Devices per channel 63 127 1 1 4 Data width 4 2 2/lane 4 4 Peak bandwidth 50MB/s or 100MB/s 0.2MB/s, 1.5MB/s, or 60MB/s 250MB/s/lane 1×, 2×, 4×, 8×, 16×, 32× 300MB/s 300MB/s Hot pluggable Yes Yes Depends Yes Yes Max length 4.5m 5m 0.5m 1m 8m Standard IEEE 1394 USB Implementers Forum PCI-SIG SATA-IO INCITS TC T10 Chapter 6 — Storage and Other I/O Topics — 16 Typical x86 PC I/O System Chapter 6 — Storage and Other I/O Topics — 17 I/O Management  I/O is mediated by the OS  Multiple programs share I/O resources  Need protection and scheduling  I/O causes asynchronous interrupts  Same mechanism as exceptions  I/O programming is fiddly  OS provides abstractions to programs (requirements for correct device control are often very detailed) § 6 .6 In te rfa c in g I/O D e v ic e s … Chapter 6 — Storage and Other I/O Topics — 20 Polling  Periodically check I/O status register  If device ready, do operation  If error, take action  Common in small or low-performance real- time embedded systems  Predictable timing  Low hardware cost  In other systems, wastes CPU time Chapter 6 — Storage and Other I/O Topics — 21 Interrupts  When a device is ready or error occurs  Controller interrupts CPU  Interrupt is like an exception  But not synchronized to instruction execution  Can invoke handler between instructions  Cause information often identifies the interrupting device  Priority interrupts  Devices needing more urgent attention get higher priority  Can interrupt handler for a lower priority interrupt Chapter 6 — Storage and Other I/O Topics — 22 I/O Data Transfer  Polling and interrupt-driven I/O  CPU transfers data between memory and I/O data registers  Time consuming for high-speed devices  Direct memory access (DMA)  OS provides starting address in memory  I/O controller transfers to/from memory autonomously  Controller interrupts on completion or error Chapter 6 — Storage and Other I/O Topics — 25 Measuring I/O Performance  I/O performance depends on  Hardware: CPU, memory, controllers, buses  Software: operating system, database management system, application  Workload: request rates and patterns  I/O system design can trade-off between response time and throughput  Measurements of throughput often done with constrained response-time § 6 .7 I/O P e rfo rm a n c e M e a s u re s : … Chapter 6 — Storage and Other I/O Topics — 26 Transaction Processing Benchmarks  Transactions  Small data accesses to a DBMS  Interested in I/O rate, not data rate  Measure throughput  Subject to response time limits and failure handling  ACID (Atomicity, Consistency, Isolation, Durability)  Overall cost per transaction  Transaction Processing Council (TPC) benchmarks (www.tcp.org)  TPC-APP: B2B application server and web services  TCP-C: on-line order entry environment  TCP-E: on-line transaction processing for brokerage firm  TPC-H: decision support — business oriented ad-hoc queries Chapter 6 — Storage and Other I/O Topics — 27 File System & Web Benchmarks  SPEC System File System (SFS)  Synthetic workload for NFS server, based on monitoring real systems  Results  Throughput (operations/sec)  Response time (average ms/operation)  SPEC Web Server benchmark  Measures simultaneous user sessions, subject to required throughput/session  Three workloads: Banking, Ecommerce, and Support Chapter 6 — Storage and Other I/O Topics — 30 RAID 1 & 2  RAID 1: Mirroring  N + N disks, replicate data  Write data to both data disk and mirror disk  On disk failure, read from mirror  RAID 2: Error correcting code (ECC)  N + E disks (e.g., 10 + 4)  Split data at bit level across N disks  Generate E-bit ECC  Too complex, not used in practice Chapter 6 — Storage and Other I/O Topics — 31 RAID 3: Bit-Interleaved Parity  N + 1 disks  Data striped across N disks at byte level  Redundant disk stores parity  Read access  Read all disks  Write access  Generate new parity and update all disks  On failure  Use parity to reconstruct missing data  Not widely used Chapter 6 — Storage and Other I/O Topics — 32 RAID 4: Block-Interleaved Parity  N + 1 disks  Data striped across N disks at block level  Redundant disk stores parity for a group of blocks  Read access  Read only the disk holding the required block  Write access  Just read disk containing modified block, and parity disk  Calculate new parity, update data disk and parity disk  On failure  Use parity to reconstruct missing data  Not widely used Chapter 6 — Storage and Other I/O Topics — 35 RAID 6: P + Q Redundancy  N + 2 disks  Like RAID 5, but two lots of parity  Greater fault tolerance through more redundancy  Multiple RAID  More advanced systems give similar fault tolerance with better performance Chapter 6 — Storage and Other I/O Topics — 36 RAID Summary  RAID can improve performance and availability  High availability requires hot swapping  Assumes independent disk failures  Too bad if the building burns down!  See ―Hard Disk Performance, Quality and Reliability‖  http://www.pcguide.com/ref/hdd/perf/index.htm Chapter 6 — Storage and Other I/O Topics — 37 I/O System Design  Satisfying latency requirements  For time-critical operations  If system is unloaded  Add up latency of components  Maximizing throughput  Find ―weakest link‖ (lowest-bandwidth component)  Configure to operate at its maximum bandwidth  Balance remaining components in the system  If system is loaded, simple analysis is insufficient  Need to use queuing models or simulation § 6 .8 D e s ig n in g a n d I/O S y s te m Chapter 6 — Storage and Other I/O Topics — 40 Sun Fire x4150 1U server 4 cores each 16 x 4GB = 64GB DRAM Chapter 6 — Storage and Other I/O Topics — 41 I/O System Design Example  Given a Sun Fire x4150 system with  Workload: 64KB disk reads  Each I/O op requires 200,000 user-code instructions and 100,000 OS instructions  Each CPU: 109 instructions/sec  FSB: 10.6 GB/sec peak  DRAM DDR2 667MHz: 5.336 GB/sec  PCI-E 8× bus: 8 × 250MB/sec = 2GB/sec  Disks: 15,000 rpm, 2.9ms avg. seek time, 112MB/sec transfer rate  What I/O rate can be sustained?  For random reads, and for sequential reads Chapter 6 — Storage and Other I/O Topics — 42 Design Example (cont)  I/O rate for CPUs  Per core: 109/(100,000 + 200,000) = 3,333  8 cores: 26,667 ops/sec  Random reads, I/O rate for disks  Assume actual seek time is average/4  Time/op = seek + latency + transfer = 2.9ms/4 + 4ms/2 + 64KB/(112MB/s) = 3.3ms  303 ops/sec per disk, 2424 ops/sec for 8 disks  Sequential reads  112MB/s / 64KB = 1750 ops/sec per disk  14,000 ops/sec for 8 disks Chapter 6 — Storage and Other I/O Topics — 45 Fallacies  Disk failure rates are as specified  Studies of failure rates in the field  Schroeder and Gibson: 2% to 4% vs. 0.6% to 0.8%  Pinheiro, et al.: 1.7% (first year) to 8.6% (third year) vs. 1.5%  Why?  A 1GB/s interconnect transfers 1GB in one sec  But what’s a GB?  For bandwidth, use 1GB = 109 B  For storage, use 1GB = 230 B = 1.075×109 B  So 1GB/sec is 0.93GB in one second  About 7% error Chapter 6 — Storage and Other I/O Topics — 46 Pitfall: Offloading to I/O Processors  Overhead of managing I/O processor request may dominate  Quicker to do small operation on the CPU  But I/O architecture may prevent that  I/O processor may be slower  Since it’s supposed to be simpler  Making it faster makes it into a major system component  Might need its own coprocessors! Chapter 6 — Storage and Other I/O Topics — 47 Pitfall: Backing Up to Tape  Magnetic tape used to have advantages  Removable, high capacity  Advantages eroded by disk technology developments  Makes better sense to replicate data  E.g, RAID, remote mirroring