I/O Devices. Device. Lecture Notes Week 8

Transcription

1 I/O Devices CPU PC ALU System bus Memory bus Bus interface I/O bridge Main memory USB Graphics adapter I/O bus Disk other devices such as network adapters Mouse Keyboard Disk hello executable stored on disk Hierarchical organization: speed, cost External to system { real world constraints (real-time). External busbased connection. Device 1. External interface. Standards based. 2. Internal structure. Not externally visible, per se. CPU? Firmware? Internal buses? 1

2 Device Communications Communications protocol. Example uses simple registers. 1. Polling. CPU-intensive. aka Programmed I/O (PIO). Asking the same question over-and-over very quickly. Very inecient. Why? Similarity to spin locks. 2. Interrupts ISR / Interrupt handler. Real-time constraints more obvious: upper half / lower half. Concurrency Caution. Not always the better solution. Why? Hybrid approaches exist. 2

3 3. DMA. Direct memory access. CPU not needed to copy data to/from device. Better concurrency. Why? DMA Controller. Acts like an I/O processor. Figure 1: PIO with interrupts Figure 2: DMA: Direct Memory Access Figure 3: File system stack (layered) Hard Disk Drives (HDD) Electro-mechanical. 1. Geometry Platter(s) Spindle Disk arm + head 3

4 2. Very, very slow. Rotational speed. 3. Partial write 4. Logs (a bit later) 1. Track 2. Sector 3. Cylinder group 4. Rotational Delay. Single or multi-track, pretty much the same time. 5. Seek time big issue Multi-track. Many steps, none of them very fast. 4

5 6. Track skew. Improvement. Why? Disk writes 1. When does the disk controller acknowledge Load to buer Disk write completed 2. Controller caching. 3. I/O reordering 5

6 4. Access algorithm Calculations Disk Scheduling Recall that a disk and/or controller can have a pretty sophisticated local processor. 1. SSTF: shortest seek-time rst. Starvation. Why? 2. Elevator Primary above disk level. SCAN C-SCAN 3. SJF? goal: SSTF but fair 4. SPTF: shortest positioning time rst. Premise: seek and rotation times becoming the same. Starvation? Queue snapshots? Final Note MTTF (mean time to failure) is a very real concern for physical devices. Bit fatigue is also an issue. More about this under SSDs. Wikipedia article. Regarding terminology: MTTF vs. MTBF vs. MTTR. 6

7 Solid State Drives (SSD) Electronic only. No mechanical wear, but : : : wear still an issue. At the transistor level 1. single-level cell (SLC) ash, only a single bit is stored within a transistor (i.e., 1 or 0) 2. multi-level cell (MLC) ash, two bits are encoded into dierent levels of charge, e.g., 00, 01, 10, and 11 are represented by low, somewhat low, somewhat high, and high levels 3. triple-level cell (TLC) ash, which encodes 3 bits per cell. Each has its own MTTF statistics. Question: if more bits per transistor, what if the one transistor fails? Random transistor failures are more likely than whole device failures. Data recovery will be covered later. Old (1980s) technology. Starting to appear: 1. Memory prices dropping 2. Manufacturing technology improving Multiple approaches in hardware 3. Density soaring 4. Pricing (approx.) SSD: $0.60 per GB HDD: $0.05 per GB Figure 4: Flash-based SSD Interface is called Flash Translation Layer (FTL). Intelligent controller. 7

8 Benets 1. Faster read/write than traditional disks 2. Power / cooling / form factor 3. Multi-path read/write Downsides No seek or rotation costs! 1. Localized writes mean bit fatigue or wear leveling Pretty much need a log-structured le system (more on this under le systems) 2. Block-based mapping is troublesome for writes, but logs help Performance File systems Key Abstractions 1. inode 2. File 3. Directory Operations 1. open / create (creat) open(const char *pathname, int flags, mode t mode); creat is open with ags set to O CREAT O WRONLY O TRUNC. 8

9 2. read read(int fd, void *buf, size t count); 3. write write(int fd, const void *buf, size t count); 4. lseek(int fd, off t offset, int whence); whence can be measures from le start, end, or current oset. 5. fsync fsync(int fd); Force synchronization with associated storage 6. stat/fstat int stat(const char *pathname, struct stat *buf); int fstat(int fd, struct stat *buf); s t r u c t s t a t f dev t s t d e v ; / ID o f d e v i c e c o n t a i n i n g f i l e / i n o t s t i n o ; / inode number / mode t st mode ; / p r o t e c t i o n / n l i n k t s t n l i n k ; / number o f hard l i n k s / u i d t s t u i d ; / user ID o f owner / g i d t s t g i d ; / group ID o f owner / dev t s t r d e v ; / d e v i c e ID ( i f s p e c i a l f i l e ) / o f f t s t s i z e ; / t o t a l s i z e, in bytes / b l k s i z e t s t b l k s i z e ; / b l o c k s i z e f o r f i l e s y s t e m I /O / b l k c n t t s t b l o c k s ; / number o f 512B b locks a l l o c a t e d / g ; 7. unlink unlink(const char *pathname); Very useful for temporary les! 8. opendir / readdir / closedir 9. Links hard links \ln". Direct inode reference. Reference count. Restricted to shared device. symbolic or soft \ln -s". Uses path. Cross device. Dangling reference on delete. 9