Storage 1: Caches, memory hierarchy, buffer cache, system calls

Overview

In this lecture, we introduce the memory hierarchy and discuss how files are read and written.

Full lecture notes on storage — Textbook readings

File I/O via system calls

• The C++ standard defines library functions for file access
  • C-like calls: fopen, fread, fwrite, printf, STDIN, STDOUT…
  • C++-like object interfaces: std::iostream, std::cin, std::cout…
• These functions are implemented in terms of system calls
  • open(filename, mode): Open a file; returns a file descriptor
    • man 2 open (system calls are defined in manual section 2)
  • read(fd, buf, sz): Read up to sz bytes from fd into buf; return number of bytes read
  • write(fd, buf, sz): Write up to sz bytes from buf into fd; return number of bytes written

Write data to a file one byte at a time

    size_t n = 0;
    while (n < size) {
        int ch = fputc(buf[0], f);
        if (ch == EOF) {
            perror("write");
            exit(1);
        }
        ++n;
    }

• fputc(buf[0], f) ≈ fwrite(buf, 1, 1, f), with different error return convention
  • Check the manual!

Write data to a file one byte at a time II

    size_t n = 0;
    while (n < size) {
        ssize_t r = write(fd, buf, 1);
        if (r != 1) {
            perror("write");
            exit(1);
        }
        ++n;
    }

Write data to memory one byte at a time

• ?

Question 1

• How long will it take each program to write data one byte at a time?
• Will the results differ by operating system? How much?

Why is membyte faster than osbyte?

Why is stdiobyte faster than osbyte?

Why is osbyte faster than syncbyte?

• w-syncbyte.cc opens the file in a mode where the kernel immediately completes every write * The write system call does not return until the drive’s data is modified

Storage hierarchy

• Also called “memory hierarchy”

Expense of storage

• Expense is a major factor in the construction of computer systems
  • Faster technologies are often more expensive
  • Older, slower technologies are often cheaper
• The relative and absolute expenses of storage technologies may be the single thing about computer systems that has changed the most dramatically over time

Question 2

• How much more did a megabyte of memory cost in 1955 relative to now?
• How much more did a megabyte of hard disk storage cost in 1955 relative to now?

Historical costs of storage

Absolute costs ($/MB):

Relative costs (relative to hard disk storage in 2018):

(Processor speed has also increased a lot—from 0.002 MIPS in 1951 to 750,000 MIPS now—but this is “only” 375,000,000x.)

Caching

• Cache: a small amount of fast storage used to speed up access to slower storage
• Computer systems are rife with caches
  • One of the most important patterns you learn in this class
  • Registers as a cache for primary memory
  • Memory as a cache for flash/hard disk
• Life is rife with caches

Investigating the standard I/O cache

• strace
  • Powerful Linux program snoops on another program’s system calls
  • Prints a human-readable summary of those system calls to a file
• strace -o strace.out PROGRAM ARGUMENTS…
  • -f: follow forks (subprocesses)
  • -s N: Print more bytes of string arguments/return values
  • man strace for more