1 <!-- Copyright (C) 2003 Red Hat, Inc. -->
2 <!-- This material may be distributed only subject to the terms -->
3 <!-- and conditions set forth in the Open Publication License, v1.0 -->
4 <!-- or later (the latest version is presently available at -->
5 <!-- http://www.opencontent.org/openpub/). -->
6 <!-- Distribution of the work or derivative of the work in any -->
7 <!-- standard (paper) book form is prohibited unless prior -->
8 <!-- permission is obtained from the copyright holder. -->
12 >Kernel Real-time Characterization</TITLE
13 ><meta name="MSSmartTagsPreventParsing" content="TRUE">
16 CONTENT="Modular DocBook HTML Stylesheet Version 1.76b+
19 TITLE="eCos Reference Manual"
20 HREF="ecos-ref.html"><LINK
22 TITLE="The eCos Kernel"
23 HREF="kernel.html"><LINK
25 TITLE="Interrupt Handling"
26 HREF="kernel-interrupts.html"><LINK
28 TITLE="RedBoot™ User's Guide"
29 HREF="redboot.html"></HEAD
40 SUMMARY="Header navigation table"
49 >eCos Reference Manual</TH
57 HREF="kernel-interrupts.html"
82 NAME="KERNEL-CHARACTERIZATION">Kernel Real-time Characterization</H1
90 >tm_basic -- Measure the performance of the eCos kernel</DIV
94 NAME="KERNEL-CHARACTERIZATION-DESCRIPTION"
99 >When building a real-time system, care must be taken to ensure that
100 the system will be able to perform properly within the constraints of
101 that system. One of these constraints may be how fast certain
102 operations can be performed. Another might be how deterministic the
103 overall behavior of the system is. Lastly the memory footprint (size)
104 and unit cost may be important.
107 >One of the major problems encountered while evaluating a system will
108 be how to compare it with possible alternatives. Most manufacturers of
109 real-time systems publish performance numbers, ostensibly so that
110 users can compare the different offerings. However, what these numbers
111 mean and how they were gathered is often not clear. The values are
112 typically measured on a particular piece of hardware, so in order to
113 truly compare, one must obtain measurements for exactly the same set
114 of hardware that were gathered in a similar fashion.
117 >Two major items need to be present in any given set of measurements.
118 First, the raw values for the various operations; these are typically
119 quite easy to measure and will be available for most systems. Second,
120 the determinacy of the numbers; in other words how much the value
121 might change depending on other factors within the system. This value
122 is affected by a number of factors: how long interrupts might be
123 masked, whether or not the function can be interrupted, even very
124 hardware-specific effects such as cache locality and pipeline usage.
125 It is very difficult to measure the determinacy of any given
126 operation, but that determinacy is fundamentally important to proper
127 overall characterization of a system.
130 >In the discussion and numbers that follow, three key measurements are
131 provided. The first measurement is an estimate of the interrupt
132 latency: this is the length of time from when a hardware interrupt
133 occurs until its Interrupt Service Routine (ISR) is called. The second
134 measurement is an estimate of overall interrupt overhead: this is the
135 length of time average interrupt processing takes, as measured by the
136 real-time clock interrupt (other interrupt sources will certainly take
137 a different amount of time, but this data cannot be easily gathered).
138 The third measurement consists of the timings for the various kernel
145 NAME="KERNEL-CHARACTERIZATION-METHODOLOGY"
150 >Key operations in the kernel were measured by using a simple test
151 program which exercises the various kernel primitive operations. A
152 hardware timer, normally the one used to drive the real-time clock,
153 was used for these measurements. In most cases this timer can be read
154 with quite high resolution, typically in the range of a few
155 microseconds. For each measurement, the operation was repeated a
156 number of times. Time stamps were obtained directly before and after
157 the operation was performed. The data gathered for the entire set of
158 operations was then analyzed, generating average (mean), maximum and
159 minimum values. The sample variance (a measure of how close most
160 samples are to the mean) was also calculated. The cost of obtaining
161 the real-time clock timer values was also measured, and was subtracted
162 from all other times.
165 >Most kernel functions can be measured separately. In each case, a
166 reasonable number of iterations are performed. Where the test case
167 involves a kernel object, for example creating a task, each iteration
168 is performed on a different object. There is also a set of tests which
169 measures the interactions between multiple tasks and certain kernel
170 primitives. Most functions are tested in such a way as to determine
171 the variations introduced by varying numbers of objects in the system.
172 For example, the mailbox tests measure the cost of a 'peek' operation
173 when the mailbox is empty, has a single item, and has multiple items
174 present. In this way, any effects of the state of the object or how
175 many items it contains can be determined.
178 >There are a few things to consider about these measurements. Firstly,
179 they are quite micro in scale and only measure the operation in
180 question. These measurements do not adequately describe how the
181 timings would be perturbed in a real system with multiple interrupting
182 sources. Secondly, the possible aberration incurred by the real-time
183 clock (system heartbeat tick) is explicitly avoided. Virtually all
184 kernel functions have been designed to be interruptible. Thus the
185 times presented are typical, but best case, since any particular
186 function may be interrupted by the clock tick processing. This number
187 is explicitly calculated so that the value may be included in any
188 deadline calculations required by the end user. Lastly, the reported
189 measurements were obtained from a system built with all options at
190 their default values. Kernel instrumentation and asserts are also
191 disabled for these measurements. Any number of configuration options
192 can change the measured results, sometimes quite dramatically. For
193 example, mutexes are using priority inheritance in these measurements.
194 The numbers will change if the system is built with priority
195 inheritance on mutex variables turned off.
198 >The final value that is measured is an estimate of interrupt latency.
199 This particular value is not explicitly calculated in the test program
200 used, but rather by instrumenting the kernel itself. The raw number of
201 timer ticks that elapse between the time the timer generates an
202 interrupt and the start of the timer ISR is kept in the kernel. These
203 values are printed by the test program after all other operations have
204 been tested. Thus this should be a reasonable estimate of the
205 interrupt latency over time.
211 NAME="KERNEL-CHARACTERIZATION-USING-MEASUREMENTS"
214 >Using these Measurements</H2
216 >These measurements can be used in a number of ways. The most typical
217 use will be to compare different real-time kernel offerings on similar
218 hardware, another will be to estimate the cost of implementing a task
219 using eCos (applications can be examined to see what effect the kernel
220 operations will have on the total execution time). Another use would
221 be to observe how the tuning of the kernel affects overall operation.
227 NAME="KERNEL-CHARACTERIZATION-INFLUENCES"
230 >Influences on Performance</H2
232 >A number of factors can affect real-time performance in a system. One
233 of the most common factors, yet most difficult to characterize, is the
234 effect of device drivers and interrupts on system timings. Different
235 device drivers will have differing requirements as to how long
236 interrupts are suppressed, for example. The eCos system has been
237 designed with this in mind, by separating the management of interrupts
238 (ISR handlers) and the processing required by the interrupt
239 (DSR—Deferred Service Routine— handlers). However, since
240 there is so much variability here, and indeed most device drivers will
241 come from the end users themselves, these effects cannot be reliably
242 measured. Attempts have been made to measure the overhead of the
243 single interrupt that eCos relies on, the real-time clock timer. This
244 should give you a reasonable idea of the cost of executing interrupt
245 handling for devices.
251 NAME="KERNEL-CHARACTERIZATION-MEASURED-ITEMS"
256 >This section describes the various tests and the numbers presented.
257 All tests use the C kernel API (available by way of
260 >cyg/kernel/kapi.h</TT
261 >). There is a single main thread
262 in the system that performs the various tests. Additional threads may
263 be created as part of the testing, but these are short lived and are
264 destroyed between tests unless otherwise noted. The terminology
265 “lower priority” means a priority that is less important,
266 not necessarily lower in numerical value. A higher priority thread
267 will run in preference to a lower priority thread even though the
268 priority value of the higher priority thread may be numerically less
269 than that of the lower priority thread.
274 NAME="KERNEL-CHARACTERIZATION-MEASURE-THREADS"
277 >Thread Primitives</H3
287 >This test measures the <TT
289 >cyg_thread_create()</TT
291 Each call creates a totally new thread. The set of threads created by
292 this test will be reused in the subsequent thread primitive tests.
299 >This test measures the <TT
301 >cyg_thread_yield()</TT
303 For this test, there are no other runnable threads, thus the test
304 should just measure the overhead of trying to give up the CPU.
308 >Suspend [suspended] thread</DT
311 >This test measures the <TT
313 >cyg_thread_suspend()</TT
315 A thread may be suspended multiple times; each thread is already
316 suspended from its initial creation, and is suspended again.
323 >This test measures the <TT
325 >cyg_thread_resume()</TT
327 All of the threads have a suspend count of 2, thus this call does not
328 make them runnable. This test just measures the overhead of resuming a
336 >This test measures the <TT
338 >cyg_thread_set_priority()</TT
340 call. Each thread, currently suspended, has its priority set to a new
348 >This test measures the <TT
350 >cyg_thread_get_priority()</TT
356 >Kill [suspended] thread</DT
359 >This test measures the <TT
361 >cyg_thread_kill()</TT
363 Each thread in the set is killed. All threads are known to be
364 suspended before being killed.
368 >Yield [no other] thread</DT
371 >This test measures the <TT
373 >cyg_thread_yield()</TT
375 again. This is to demonstrate that the
378 >cyg_thread_yield()</TT
379 > call has a fixed overhead,
380 regardless of whether there are other threads in the system.
384 >Resume [suspended low priority] thread</DT
387 >This test measures the <TT
389 >cyg_thread_resume()</TT
391 again. In this case, the thread being resumed is lower priority than
392 the main thread, thus it will simply become ready to run but not be
393 granted the CPU. This test measures the cost of making a thread ready
398 >Resume [runnable low priority] thread</DT
401 >This test measures the <TT
403 >cyg_thread_resume()</TT
405 again. In this case, the thread being resumed is lower priority than
406 the main thread and has already been made runnable, so in fact the
407 resume call has no effect.
411 >Suspend [runnable] thread</DT
414 >This test measures the <TT
416 >cyg_thread_suspend()</TT
418 again. In this case, each thread has already been made runnable (by
423 >Yield [only low priority] thread</DT
426 >This test measures the <TT
428 >cyg_thread_yield()</TT
430 In this case, there are many other runnable threads, but they are all
431 lower priority than the main thread, thus no thread switches will take
436 >Suspend [runnable->not runnable] thread</DT
439 >This test measures the <TT
441 >cyg_thread_suspend()</TT
443 again. The thread being suspended will become non-runnable by this
448 >Kill [runnable] thread</DT
451 >This test measures the <TT
453 >cyg_thread_kill()</TT
455 again. In this case, the thread being killed is currently runnable,
456 but lower priority than the main thread.
460 >Resume [high priority] thread</DT
463 >This test measures the <TT
465 >cyg_thread_resume()</TT
467 The thread being resumed is higher priority than the main thread, thus
468 a thread switch will take place on each call. In fact there will be
469 two thread switches; one to the new higher priority thread and a
470 second back to the test thread. The test thread exits immediately.
477 >This test attempts to measure the cost of switching from one thread to
478 another. Two equal priority threads are started and they will each
479 yield to the other for a number of iterations. A time stamp is
480 gathered in one thread before the
483 >cyg_thread_yield()</TT
484 > call and after the call in the
494 NAME="KERNEL-CHARACTERIZATION-MEASURE-SCHEDULER"
497 >Scheduler Primitives</H3
507 >This test measures the <TT
509 >cyg_scheduler_lock()</TT
514 >Scheduler unlock [0 threads]</DT
517 >This test measures the <TT
519 >cyg_scheduler_unlock()</TT
521 call. There are no other threads in the system and the unlock happens
522 immediately after a lock so there will be no pending DSR’s to
527 >Scheduler unlock [1 suspended thread]</DT
530 >This test measures the <TT
532 >cyg_scheduler_unlock()</TT
534 call. There is one other thread in the system which is currently
539 >Scheduler unlock [many suspended threads]</DT
542 >This test measures the <TT
544 >cyg_scheduler_unlock()</TT
546 call. There are many other threads in the system which are currently
547 suspended. The purpose of this test is to determine the cost of having
548 additional threads in the system when the scheduler is activated by
551 >cyg_scheduler_unlock()</TT
556 >Scheduler unlock [many low priority threads]</DT
559 >This test measures the <TT
561 >cyg_scheduler_unlock()</TT
563 call. There are many other threads in the system which are runnable
564 but are lower priority than the main thread. The purpose of this test
565 is to determine the cost of having additional threads in the system
566 when the scheduler is activated by way of
569 >cyg_scheduler_unlock()</TT
579 NAME="KERNEL-CHARACTERIZATION-MEASURE-MUTEX"
582 >Mutex Primitives</H3
592 >This test measures the <TT
594 >cyg_mutex_init()</TT
596 number of separate mutex variables are created. The purpose of this
597 test is to measure the cost of creating a new mutex and introducing it
602 >Lock [unlocked] mutex</DT
605 >This test measures the <TT
607 >cyg_mutex_lock()</TT
609 purpose of this test is to measure the cost of locking a mutex which
610 is currently unlocked. There are no other threads executing in the
611 system while this test runs.
615 >Unlock [locked] mutex</DT
618 >This test measures the <TT
620 >cyg_mutex_unlock()</TT
622 The purpose of this test is to measure the cost of unlocking a mutex
623 which is currently locked. There are no other threads executing in the
624 system while this test runs.
628 >Trylock [unlocked] mutex</DT
631 >This test measures the <TT
633 >cyg_mutex_trylock()</TT
635 The purpose of this test is to measure the cost of locking a mutex
636 which is currently unlocked. There are no other threads executing in
637 the system while this test runs.
641 >Trylock [locked] mutex</DT
644 >This test measures the <TT
646 >cyg_mutex_trylock()</TT
648 The purpose of this test is to measure the cost of locking a mutex
649 which is currently locked. There are no other threads executing in the
650 system while this test runs.
657 >This test measures the <TT
659 >cyg_mutex_destroy()</TT
661 The purpose of this test is to measure the cost of deleting a mutex
662 from the system. There are no other threads executing in the system
663 while this test runs.
667 >Unlock/Lock mutex</DT
670 >This test attempts to measure the cost of unlocking a mutex for which
671 there is another higher priority thread waiting. When the mutex is
672 unlocked, the higher priority waiting thread will immediately take the
673 lock. The time from when the unlock is issued until after the lock
674 succeeds in the second thread is measured, thus giving the round-trip
675 or circuit time for this type of synchronizer.
684 NAME="KERNEL-CHARACTERIZATION-MEASURE-MAILBOX"
687 >Mailbox Primitives</H3
697 >This test measures the <TT
699 >cyg_mbox_create()</TT
701 number of separate mailboxes is created. The purpose of this test is
702 to measure the cost of creating a new mailbox and introducing it to
707 >Peek [empty] mbox</DT
710 >This test measures the <TT
714 attempt is made to peek the value in each mailbox, which is currently
715 empty. The purpose of this test is to measure the cost of checking a
716 mailbox for a value without blocking.
720 >Put [first] mbox</DT
723 >This test measures the <TT
727 item is added to a currently empty mailbox. The purpose of this test
728 is to measure the cost of adding an item to a mailbox. There are no
729 other threads currently waiting for mailbox items to arrive.
733 >Peek [1 msg] mbox</DT
736 >This test measures the <TT
740 attempt is made to peek the value in each mailbox, which contains a
741 single item. The purpose of this test is to measure the cost of
742 checking a mailbox which has data to deliver.
746 >Put [second] mbox</DT
749 >This test measures the <TT
753 second item is added to a mailbox. The purpose of this test is to
754 measure the cost of adding an additional item to a mailbox. There are
755 no other threads currently waiting for mailbox items to arrive.
759 >Peek [2 msgs] mbox</DT
762 >This test measures the <TT
766 attempt is made to peek the value in each mailbox, which contains two
767 items. The purpose of this test is to measure the cost of checking a
768 mailbox which has data to deliver.
772 >Get [first] mbox</DT
775 >This test measures the <TT
779 first item is removed from a mailbox that currently contains two
780 items. The purpose of this test is to measure the cost of obtaining an
781 item from a mailbox without blocking.
785 >Get [second] mbox</DT
788 >This test measures the <TT
792 last item is removed from a mailbox that currently contains one item.
793 The purpose of this test is to measure the cost of obtaining an item
794 from a mailbox without blocking.
798 >Tryput [first] mbox</DT
801 >This test measures the <TT
803 >cyg_mbox_tryput()</TT
805 single item is added to a currently empty mailbox. The purpose of this
806 test is to measure the cost of adding an item to a mailbox.
810 >Peek item [non-empty] mbox</DT
813 >This test measures the <TT
815 >cyg_mbox_peek_item()</TT
817 A single item is fetched from a mailbox that contains a single item.
818 The purpose of this test is to measure the cost of obtaining an item
819 without disturbing the mailbox.
823 >Tryget [non-empty] mbox</DT
826 >This test measures the <TT
828 >cyg_mbox_tryget()</TT
830 single item is removed from a mailbox that contains exactly one item.
831 The purpose of this test is to measure the cost of obtaining one item
832 from a non-empty mailbox.
836 >Peek item [empty] mbox</DT
839 >This test measures the <TT
841 >cyg_mbox_peek_item()</TT
843 An attempt is made to fetch an item from a mailbox that is empty. The
844 purpose of this test is to measure the cost of trying to obtain an
845 item when the mailbox is empty.
849 >Tryget [empty] mbox</DT
852 >This test measures the <TT
854 >cyg_mbox_tryget()</TT
856 attempt is made to fetch an item from a mailbox that is empty. The
857 purpose of this test is to measure the cost of trying to obtain an
858 item when the mailbox is empty.
862 >Waiting to get mbox</DT
865 >This test measures the <TT
867 >cyg_mbox_waiting_to_get()</TT
869 call. The purpose of this test is to measure the cost of determining
870 how many threads are waiting to obtain a message from this mailbox.
874 >Waiting to put mbox</DT
877 >This test measures the <TT
879 >cyg_mbox_waiting_to_put()</TT
881 call. The purpose of this test is to measure the cost of determining
882 how many threads are waiting to put a message into this mailbox.
889 >This test measures the <TT
891 >cyg_mbox_delete()</TT
893 The purpose of this test is to measure the cost of destroying a
894 mailbox and removing it from the system.
901 >In this round-trip test, one thread is sending data to a mailbox that
902 is being consumed by another thread. The time from when the data is
903 put into the mailbox until it has been delivered to the waiting thread
904 is measured. Note that this time will contain a thread switch.
913 NAME="KERNEL-CHARACTERIZATION-MEASURE-SEMAPHORE"
916 >Semaphore Primitives</H3
926 >This test measures the <TT
928 >cyg_semaphore_init()</TT
930 A number of separate semaphore objects are created and introduced to
931 the system. The purpose of this test is to measure the cost of
932 creating a new semaphore.
936 >Post [0] semaphore</DT
939 >This test measures the <TT
941 >cyg_semaphore_post()</TT
943 Each semaphore currently has a value of 0 and there are no other
944 threads in the system. The purpose of this test is to measure the
945 overhead cost of posting to a semaphore. This cost will differ if
946 there is a thread waiting for the semaphore.
950 >Wait [1] semaphore</DT
953 >This test measures the <TT
955 >cyg_semaphore_wait()</TT
957 The semaphore has a current value of 1 so the call is non-blocking.
958 The purpose of the test is to measure the overhead of
959 “taking” a semaphore.
963 >Trywait [0] semaphore</DT
966 >This test measures the <TT
968 >cyg_semaphore_trywait()</TT
970 call. The semaphore has a value of 0 when the call is made. The
971 purpose of this test is to measure the cost of seeing if a semaphore
972 can be “taken” without blocking. In this case, the answer
977 >Trywait [1] semaphore</DT
980 >This test measures the <TT
982 >cyg_semaphore_trywait()</TT
984 call. The semaphore has a value of 1 when the call is made. The
985 purpose of this test is to measure the cost of seeing if a semaphore
986 can be “taken” without blocking. In this case, the answer
994 >This test measures the <TT
996 >cyg_semaphore_peek()</TT
998 The purpose of this test is to measure the cost of obtaining the
999 current semaphore count value.
1003 >Destroy semaphore</DT
1006 >This test measures the <TT
1008 >cyg_semaphore_destroy()</TT
1010 call. The purpose of this test is to measure the cost of deleting a
1011 semaphore from the system.
1015 >Post/Wait semaphore</DT
1018 >In this round-trip test, two threads are passing control back and
1019 forth by using a semaphore. The time from when one thread calls
1022 >cyg_semaphore_post()</TT
1023 > until the other thread
1026 >cyg_semaphore_wait()</TT
1028 Note that each iteration of this test will involve a thread switch.
1037 NAME="KERNEL-CHARACTERIZATION-MEASURE-COUNTERS"
1044 CLASS="VARIABLELIST"
1050 >This test measures the <TT
1052 >cyg_counter_create()</TT
1054 A number of separate counters are created. The purpose of this test is
1055 to measure the cost of creating a new counter and introducing it to
1060 >Get counter value</DT
1063 >This test measures the
1066 >cyg_counter_current_value()</TT
1068 value of each counter is obtained.
1072 >Set counter value</DT
1075 >This test measures the <TT
1077 >cyg_counter_set_value()</TT
1079 call. Each counter is set to a new value.
1086 >This test measures the <TT
1088 >cyg_counter_tick()</TT
1090 Each counter is “ticked” once.
1097 >This test measures the <TT
1099 >cyg_counter_delete()</TT
1101 Each counter is deleted from the system. The purpose of this test is
1102 to measure the cost of deleting a counter object.
1111 NAME="KERNEL-CHARACTERIZATION-MEASURE-ALARMS"
1118 CLASS="VARIABLELIST"
1124 >This test measures the <TT
1126 >cyg_alarm_create()</TT
1128 number of separate alarms are created, all attached to the same
1129 counter object. The purpose of this test is to measure the cost of
1130 creating a new counter and introducing it to the system.
1134 >Initialize alarm</DT
1137 >This test measures the <TT
1139 >cyg_alarm_initialize()</TT
1141 call. Each alarm is initialized to a small value.
1148 >This test measures the <TT
1150 >cyg_alarm_disable()</TT
1152 Each alarm is explicitly disabled.
1159 >This test measures the <TT
1161 >cyg_alarm_enable()</TT
1163 Each alarm is explicitly enabled.
1170 >This test measures the <TT
1172 >cyg_alarm_delete()</TT
1174 Each alarm is destroyed. The purpose of this test is to measure the
1175 cost of deleting an alarm and removing it from the system.
1179 >Tick counter [1 alarm]</DT
1182 >This test measures the <TT
1184 >cyg_counter_tick()</TT
1186 counter is created that has a single alarm attached to it. The purpose
1187 of this test is to measure the cost of “ticking” a counter
1188 when it has a single attached alarm. In this test, the alarm is not
1193 >Tick counter [many alarms]</DT
1196 >This test measures the <TT
1198 >cyg_counter_tick()</TT
1200 counter is created that has multiple alarms attached to it. The
1201 purpose of this test is to measure the cost of “ticking” a
1202 counter when it has many attached alarms. In this test, the alarms are
1203 not activated (fired).
1207 >Tick & fire counter [1 alarm]</DT
1210 >This test measures the <TT
1212 >cyg_counter_tick()</TT
1214 counter is created that has a single alarm attached to it. The purpose
1215 of this test is to measure the cost of “ticking” a counter
1216 when it has a single attached alarm. In this test, the alarm is
1217 activated (fired). Thus the measured time will include the overhead of
1218 calling the alarm callback function.
1222 >Tick & fire counter [many alarms]</DT
1225 >This test measures the <TT
1227 >cyg_counter_tick()</TT
1229 counter is created that has multiple alarms attached to it. The
1230 purpose of this test is to measure the cost of “ticking” a
1231 counter when it has many attached alarms. In this test, the alarms are
1232 activated (fired). Thus the measured time will include the overhead of
1233 calling the alarm callback function.
1237 >Alarm latency [0 threads]</DT
1240 >This test attempts to measure the latency in calling an alarm callback
1241 function. The time from the clock interrupt until the alarm function
1242 is called is measured. In this test, there are no threads that can be
1243 run, other than the system idle thread, when the clock interrupt
1244 occurs (all threads are suspended).
1248 >Alarm latency [2 threads]</DT
1251 >This test attempts to measure the latency in calling an alarm callback
1252 function. The time from the clock interrupt until the alarm function
1253 is called is measured. In this test, there are exactly two threads
1254 which are running when the clock interrupt occurs. They are simply
1255 passing back and forth by way of the
1258 >cyg_thread_yield()</TT
1259 > call. The purpose of this test
1260 is to measure the variations in the latency when there are executing
1265 >Alarm latency [many threads]</DT
1268 >This test attempts to measure the latency in calling an alarm callback
1269 function. The time from the clock interrupt until the alarm function
1270 is called is measured. In this test, there are a number of threads
1271 which are running when the clock interrupt occurs. They are simply
1272 passing back and forth by way of the
1275 >cyg_thread_yield()</TT
1276 > call. The purpose of this test
1277 is to measure the variations in the latency when there are many
1290 SUMMARY="Footer navigation table"
1301 HREF="kernel-interrupts.html"
1310 HREF="ecos-ref.html"
1329 >Interrupt Handling</TD
1343 >RedBoot™ User's Guide</TD