Lines Matching +full:cache +full:- +full:time +full:- +full:ms

61 time is long.
63 it may be advantageous to cache its results,
73 that begins to dominate execution time.
91 Despite the long period of time that elapsed
113 We find that our general time sharing systems do about
117 representing 40% of the time processing system calls,
126 part	time	% of kernel
128 self	14.3 ms/call	11.3%
129 child	9.9 ms/call	7.9%
131 total	24.2 ms/call	19.2%
150 This caused us to modify \fInamei\fP to cache
155 begins.  Changing directories invalidates the cache, as
158 $N$ files, search time decreases from $O ( N sup 2 )$
161 The cost of the cache is about 20 lines of code
167 cache we ran ``ls \-l''
169 Before the per-process cache this command
170 used 22.3 seconds of system time.
171 After adding the cache the program used the same amount
172 of user time, but the system time dropped to 3.3 seconds.
176 The results showed that the time in \fInamei\fP
177 dropped by only 2.6 ms/call and
178 still accounted for 36% of the system call time,
180 This amounted to a drop in system time from 57% to about 55%.
187 part	time	% of kernel
189 self	11.0 ms/call	9.2%
190 child	10.6 ms/call	8.9%
192 total	21.6 ms/call	18.1%
195 Figure 4. Call times for \fInamei\fP with per-process cache.
200 was caused by a low cache hit ratio.
201 Although the cache was 90% effective when hit,
204 although the amount of time spent in \fInamei\fP itself
206 more time was spent in the routines that it called
213 Thus Robert Elz introduced a system wide cache of most recent
215 The cache is keyed on a name and the
221 \fInamei\fP first looks in its cache of recent translations
225 then the per-process cache may still be useful in
226 reducing the directory search time.
229 The cost of the name cache is about 200 lines of code
231 and 44 bytes per cache entry.
234 using 10-44 kilobytes of physical memory.
235 The name cache is resident in memory at all times.
237 After adding the system wide name cache we reran ``ls \-l''
239 The user time remained the same,
240 however the system time rose slightly to 3.7 seconds.
242 now had to maintain the cache,
247 showed a 6 ms/call decrease in \fInamei\fP, with
248 \fInamei\fP accounting for only 31% of the system call time,
249 16% of the time in the kernel,
251 System time dropped from 55% to about 49%.
258 part	time	% of kernel
260 self	9.5 ms/call	9.6%
261 child	6.1 ms/call	6.1%
263 total	15.6 ms/call	15.7%
274 the system wide cache.  This, coupled with the 25%
275 hit rate in the per-process offset cache yielded an
276 effective cache hit rate of 85%.
277 While the system wide cache reduces both the amount of time in
281 time spent in \fInamei\fP itself increases even though the
282 actual time per call decreases.
283 This is because less total time is being spent in the kernel,
284 hence a smaller absolute time becomes a larger total percentage.