Feder, J. The Evolution of UNIX System Performance. AT&T Bell Laboratories Technical Journal 63, 8 (October 1984), 1791--1814.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....and finally the process will terminate with an exit system call. Table 20 gives average times for each of these phases as well as the average time taken between fork and exit, that is, the total lifetime of a particular process. It has been noted that UNIX processes are, on average, short lived [35, 52]. This is partially because process creation has a low overhead and as a result processes have become a resource that is highly used by operating systems developers. Figure 27 shows a cumulative distribution of the time taken in each phase of the process s life. It is noteworthy that while more ....

Feder, J. The Evolution of UNIX System Performance. AT&T Bell Laboratories Technical Journal 63, 8 (October 1984), 1791--1814.

....of various problems. The two process approach is very inefficient. Every token needs to pass through the operating system two times (one write and one read) in order to go from one process to the other. This involves two system calls and a context switch. A typical time for a system call is 350 [Fed84, p. 1796] and for a context switch 600 [Fed84, p. 1798] The total overhead for a single token would be 1 3 . The performance finally obtained (see section 1.5.6 was 128 per token) so this approach would make the scanner ten times slower. Both approaches are not portable. There is no standard C ....

....is very inefficient. Every token needs to pass through the operating system two times (one write and one read) in order to go from one process to the other. This involves two system calls and a context switch. A typical time for a system call is 350 [Fed84, p. 1796] and for a context switch 600 [Fed84, p. 1798] The total overhead for a single token would be 1 3 . The performance finally obtained (see section 1.5.6 was 128 per token) so this approach would make the scanner ten times slower. Both approaches are not portable. There is no standard C library for coroutines and switching the stack ....

J. Feder. The evolution of UNIX system performance. Bell System Technical Journal, 63(8):1791--1814, October 1984.

....significant performance advantage over having to re fetch those blocks a short time after the close. 3.5.3. Block size We also evaluated the effects of differing block sizes. The original UNIX file system used 512 byte blocks. The block size has since been expanded to 1024 bytes in AT T System V [20] and 4096 bytes in most 4.2BSD systems. Figure 3 3 and Table 3 8 show the results of varying the block size and cache size. 0 32 0 40 4 8 12 16 20 24 28 Line Size (Kbytes) 10 20 30 I O Ratio (percent) 400 Kbytes 1 Mbyte 2 Mbytes 4 Mbytes 8 Mbytes Figure 3 3: I O ratio vs. block ....

J. Feder. The Evolution of UNIX System Performance. Bell Laboratories Technical Journal 63(8):1791-1814, October, 1984.

....exhibited large invocation overheads. Due to its popularity and wide availability, Unix is one of the more studied systems, and I use it here as a baseline for performance comparisons. In one study, Feder compares the evolution of Unix system performance over time and over different machines [13]. He studies the AT T releases of Unix System 3 and System 5 spanning a time period from the mid 70 s to late 1982 and shows that Unix performance had improved roughly 25 during this time. Among the measurements shown is the time taken to execute the getpid (get process id) system call. ....

J. Feder. The Evolution of Unix System Performance. AT&T Bell Laboratories Technical Journal, 63(8):1791--1814, October 1984.

....but these investigations were not primarily concerned with quantitative comparisons of file buffer cache access methods, or the impact of large physical memories on their performance. Giving more memory to a file buffer cache can boost system performance, but it can also have negative effects: [Feder84] reported that increasing the buffer cache size (and hence the number of cache blocks) sometimes resulted in lower overall performance, because the cost of finding data in the cache increased as the cache got larger. Measurements reported in [Ousterhout85] showed that using a delayed write ....

J. Feder. The evolution of UNIX system performance. AT&T Bell Laboratories Technical Journal, 63(8 part 2):1791--814, October 1984.

....measurement of the lifetime of information in files, and are similar to the results of Figure 4. 6.3. Block Size We also evaluated the effectiveness of different block sizes. The original UNIX system used 512 byte blocks, but the block size has grown since then to 1024 bytes in AT T s System V [1] and 4096 bytes in most 4.2 BSD systems. Figure 6 and Table VII show the results of varying the block size and cache size. For a 4 Mbyte cache, a block size of 16 kbytes reduces disk accesses by about 25 over a 4 kbyte block size and by a factor of 3 over 1 kbyte blocks. Even for a cache size of ....

Feder, J. "The Evolution of UNIX System Performance." Bell Laboratories Technical Journal, Vol. 63, No. 8, October 1984, pp. 1791-1814.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC