1 =head1 NAME
3 rrdcreate - Set up a new Round Robin Database
5 =head1 SYNOPSIS
7 B<rrdtool> B<create> I<filename>
8 S<[B<--start>|B<-b> I<start time>]>
9 S<[B<--step>|B<-s> I<step>]>
10 S<[B<DS:>I<ds-name>B<:>I<DST>B<:>I<dst arguments>]>
11 S<[B<RRA:>I<CF>B<:>I<cf arguments>]>
13 =head1 DESCRIPTION
15 The create function of RRDtool lets you set up new Round Robin
16 Database (B<RRD>) files. The file is created at its final, full size
17 and filled with I<*UNKNOWN*> data.
19 =head2 I<filename>
21 The name of the B<RRD> you want to create. B<RRD> files should end
22 with the extension F<.rrd>. However, B<RRDtool> will accept any
23 filename.
25 =head2 B<--start>|B<-b> I<start time> (default: now - 10s)
27 Specifies the time in seconds since 1970-01-01 UTC when the first
28 value should be added to the B<RRD>. B<RRDtool> will not accept
29 any data timed before or at the time specified.
31 See also AT-STYLE TIME SPECIFICATION section in the
32 I<rrdfetch> documentation for other ways to specify time.
34 =head2 B<--step>|B<-s> I<step> (default: 300 seconds)
36 Specifies the base interval in seconds with which data will be fed
37 into the B<RRD>.
39 =head2 B<DS:>I<ds-name>B<:>I<DST>B<:>I<dst arguments>
41 A single B<RRD> can accept input from several data sources (B<DS>),
42 for example incoming and outgoing traffic on a specific communication
43 line. With the B<DS> configuration option you must define some basic
44 properties of each data source you want to store in the B<RRD>.
46 I<ds-name> is the name you will use to reference this particular data
47 source from an B<RRD>. A I<ds-name> must be 1 to 19 characters long in
48 the characters [a-zA-Z0-9_].
50 I<DST> defines the Data Source Type. The remaining arguments of a
51 data source entry depend on the data source type. For GAUGE, COUNTER,
52 DERIVE, and ABSOLUTE the format for a data source entry is:
54 B<DS:>I<ds-name>B<:>I<GAUGE | COUNTER | DERIVE | ABSOLUTE>B<:>I<heartbeat>B<:>I<min>B<:>I<max>
56 For COMPUTE data sources, the format is:
58 B<DS:>I<ds-name>B<:>I<COMPUTE>B<:>I<rpn-expression>
60 In order to decide which data source type to use, review the
61 definitions that follow. Also consult the section on "HOW TO MEASURE"
62 for further insight.
64 =over
66 =item B<GAUGE>
68 is for things like temperatures or number of people in a room or the
69 value of a RedHat share.
71 =item B<COUNTER>
73 is for continuous incrementing counters like the ifInOctets counter in
74 a router. The B<COUNTER> data source assumes that the counter never
75 decreases, except when a counter overflows. The update function takes
76 the overflow into account. The counter is stored as a per-second
77 rate. When the counter overflows, RRDtool checks if the overflow
78 happened at the 32bit or 64bit border and acts accordingly by adding
79 an appropriate value to the result.
81 =item B<DERIVE>
83 will store the derivative of the line going from the last to the
84 current value of the data source. This can be useful for gauges, for
85 example, to measure the rate of people entering or leaving a
86 room. Internally, derive works exactly like COUNTER but without
87 overflow checks. So if your counter does not reset at 32 or 64 bit you
88 might want to use DERIVE and combine it with a MIN value of 0.
90 B<NOTE on COUNTER vs DERIVE>
92 by Don Baarda E<lt>don.baarda@baesystems.comE<gt>
94 If you cannot tolerate ever mistaking the occasional counter reset for a
95 legitimate counter wrap, and would prefer "Unknowns" for all legitimate
96 counter wraps and resets, always use DERIVE with min=0. Otherwise, using
97 COUNTER with a suitable max will return correct values for all legitimate
98 counter wraps, mark some counter resets as "Unknown", but can mistake some
99 counter resets for a legitimate counter wrap.
101 For a 5 minute step and 32-bit counter, the probability of mistaking a
102 counter reset for a legitimate wrap is arguably about 0.8% per 1Mbps of
103 maximum bandwidth. Note that this equates to 80% for 100Mbps interfaces, so
104 for high bandwidth interfaces and a 32bit counter, DERIVE with min=0 is
105 probably preferable. If you are using a 64bit counter, just about any max
106 setting will eliminate the possibility of mistaking a reset for a counter
107 wrap.
109 =item B<ABSOLUTE>
111 is for counters which get reset upon reading. This is used for fast counters
112 which tend to overflow. So instead of reading them normally you reset them
113 after every read to make sure you have a maximum time available before the
114 next overflow. Another usage is for things you count like number of messages
115 since the last update.
117 =item B<COMPUTE>
119 is for storing the result of a formula applied to other data sources
120 in the B<RRD>. This data source is not supplied a value on update, but
121 rather its Primary Data Points (PDPs) are computed from the PDPs of
122 the data sources according to the rpn-expression that defines the
123 formula. Consolidation functions are then applied normally to the PDPs
124 of the COMPUTE data source (that is the rpn-expression is only applied
125 to generate PDPs). In database software, such data sets are referred
126 to as "virtual" or "computed" columns.
128 =back
130 I<heartbeat> defines the maximum number of seconds that may pass
131 between two updates of this data source before the value of the
132 data source is assumed to be I<*UNKNOWN*>.
134 I<min> and I<max> define the expected range values for data supplied by a
135 data source. If I<min> and/or I<max> any value outside the defined range
136 will be regarded as I<*UNKNOWN*>. If you do not know or care about min and
137 max, set them to U for unknown. Note that min and max always refer to the
138 processed values of the DS. For a traffic-B<COUNTER> type DS this would be
139 the maximum and minimum data-rate expected from the device.
141 I<If information on minimal/maximal expected values is available,
142 always set the min and/or max properties. This will help RRDtool in
143 doing a simple sanity check on the data supplied when running update.>
145 I<rpn-expression> defines the formula used to compute the PDPs of a
146 COMPUTE data source from other data sources in the same <RRD>. It is
147 similar to defining a B<CDEF> argument for the graph command. Please
148 refer to that manual page for a list and description of RPN operations
149 supported. For COMPUTE data sources, the following RPN operations are
150 not supported: COUNT, PREV, TIME, and LTIME. In addition, in defining
151 the RPN expression, the COMPUTE data source may only refer to the
152 names of data source listed previously in the create command. This is
153 similar to the restriction that B<CDEF>s must refer only to B<DEF>s
154 and B<CDEF>s previously defined in the same graph command.
156 =head2 B<RRA:>I<CF>B<:>I<cf arguments>
158 The purpose of an B<RRD> is to store data in the round robin archives
159 (B<RRA>). An archive consists of a number of data values or statistics for
160 each of the defined data-sources (B<DS>) and is defined with an B<RRA> line.
162 When data is entered into an B<RRD>, it is first fit into time slots
163 of the length defined with the B<-s> option, thus becoming a I<primary
164 data point>.
166 The data is also processed with the consolidation function (I<CF>) of
167 the archive. There are several consolidation functions that
168 consolidate primary data points via an aggregate function: B<AVERAGE>,
169 B<MIN>, B<MAX>, B<LAST>.
171 =over
173 =item AVERAGE
175 the average of the data points is stored.
177 =item MIN
179 the smallest of the data points is stored.
181 =item MAX
183 the largest of the data points is stored.
185 =item LAST
187 the last data points is used.
189 =back
191 Note that data aggregation inevitably leads to loss of precision and
192 information. The trick is to pick the aggregate function such that the
193 I<interesting> properties of your data is kept across the aggregation
194 process.
197 The format of B<RRA> line for these
198 consolidation functions is:
200 B<RRA:>I<AVERAGE | MIN | MAX | LAST>B<:>I<xff>B<:>I<steps>B<:>I<rows>
202 I<xff> The xfiles factor defines what part of a consolidation interval may
203 be made up from I<*UNKNOWN*> data while the consolidated value is still
204 regarded as known. It is given as the ratio of allowed I<*UNKNOWN*> PDPs
205 to the number of PDPs in the interval. Thus, it ranges from 0 to 1 (exclusive).
208 I<steps> defines how many of these I<primary data points> are used to build
209 a I<consolidated data point> which then goes into the archive.
211 I<rows> defines how many generations of data values are kept in an B<RRA>.
212 Obviously, this has to be greater than zero.
214 =head1 Aberrant Behavior Detection with Holt-Winters Forecasting
216 In addition to the aggregate functions, there are a set of specialized
217 functions that enable B<RRDtool> to provide data smoothing (via the
218 Holt-Winters forecasting algorithm), confidence bands, and the
219 flagging aberrant behavior in the data source time series:
221 =over
223 =item *
225 B<RRA:>I<HWPREDICT>B<:>I<rows>B<:>I<alpha>B<:>I<beta>B<:>I<seasonal period>[B<:>I<rra-num>]
227 =item *
229 B<RRA:>I<MHWPREDICT>B<:>I<rows>B<:>I<alpha>B<:>I<beta>B<:>I<seasonal period>[B<:>I<rra-num>]
231 =item *
233 B<RRA:>I<SEASONAL>B<:>I<seasonal period>B<:>I<gamma>B<:>I<rra-num>[B<:smoothing-window=>I<fraction>]
235 =item *
237 B<RRA:>I<DEVSEASONAL>B<:>I<seasonal period>B<:>I<gamma>B<:>I<rra-num>[B<:smoothing-window=>I<fraction>]
239 =item *
241 B<RRA:>I<DEVPREDICT>B<:>I<rows>B<:>I<rra-num>
243 =item *
245 B<RRA:>I<FAILURES>B<:>I<rows>B<:>I<threshold>B<:>I<window length>B<:>I<rra-num>
247 =back
249 These B<RRAs> differ from the true consolidation functions in several ways.
250 First, each of the B<RRA>s is updated once for every primary data point.
251 Second, these B<RRAs> are interdependent. To generate real-time confidence
252 bounds, a matched set of SEASONAL, DEVSEASONAL, DEVPREDICT, and either
253 HWPREDICT or MHWPREDICT must exist. Generating smoothed values of the primary
254 data points requires a SEASONAL B<RRA> and either an HWPREDICT or MHWPREDICT
255 B<RRA>. Aberrant behavior detection requires FAILURES, DEVSEASONAL, SEASONAL,
256 and either HWPREDICT or MHWPREDICT.
258 The predicted, or smoothed, values are stored in the HWPREDICT or MHWPREDICT
259 B<RRA>. HWPREDICT and MHWPREDICT are actually two variations on the
260 Holt-Winters method. They are interchangeable. Both attempt to decompose data
261 into three components: a baseline, a trend, and a seasonal coefficient.
262 HWPREDICT adds its seasonal coefficient to the baseline to form a prediction, whereas
263 MHWPREDICT multiplies its seasonal coefficient by the baseline to form a
264 prediction. The difference is noticeable when the baseline changes
265 significantly in the course of a season; HWPREDICT will predict the seasonality
266 to stay constant as the baseline changes, but MHWPREDICT will predict the
267 seasonality to grow or shrink in proportion to the baseline. The proper choice
268 of method depends on the thing being modeled. For simplicity, the rest of this
269 discussion will refer to HWPREDICT, but MHWPREDICT may be substituted in its
270 place.
272 The predicted deviations are stored in DEVPREDICT (think a standard deviation
273 which can be scaled to yield a confidence band). The FAILURES B<RRA> stores
274 binary indicators. A 1 marks the indexed observation as failure; that is, the
275 number of confidence bounds violations in the preceding window of observations
276 met or exceeded a specified threshold. An example of using these B<RRAs> to graph
277 confidence bounds and failures appears in L<rrdgraph>.
279 The SEASONAL and DEVSEASONAL B<RRAs> store the seasonal coefficients for the
280 Holt-Winters forecasting algorithm and the seasonal deviations, respectively.
281 There is one entry per observation time point in the seasonal cycle. For
282 example, if primary data points are generated every five minutes and the
283 seasonal cycle is 1 day, both SEASONAL and DEVSEASONAL will have 288 rows.
285 In order to simplify the creation for the novice user, in addition to
286 supporting explicit creation of the HWPREDICT, SEASONAL, DEVPREDICT,
287 DEVSEASONAL, and FAILURES B<RRAs>, the B<RRDtool> create command supports
288 implicit creation of the other four when HWPREDICT is specified alone and
289 the final argument I<rra-num> is omitted.
291 I<rows> specifies the length of the B<RRA> prior to wrap around. Remember
292 that there is a one-to-one correspondence between primary data points and
293 entries in these RRAs. For the HWPREDICT CF, I<rows> should be larger than
294 the I<seasonal period>. If the DEVPREDICT B<RRA> is implicitly created, the
295 default number of rows is the same as the HWPREDICT I<rows> argument. If the
296 FAILURES B<RRA> is implicitly created, I<rows> will be set to the I<seasonal
297 period> argument of the HWPREDICT B<RRA>. Of course, the B<RRDtool>
298 I<resize> command is available if these defaults are not sufficient and the
299 creator wishes to avoid explicit creations of the other specialized function
300 B<RRAs>.
302 I<seasonal period> specifies the number of primary data points in a seasonal
303 cycle. If SEASONAL and DEVSEASONAL are implicitly created, this argument for
304 those B<RRAs> is set automatically to the value specified by HWPREDICT. If
305 they are explicitly created, the creator should verify that all three
306 I<seasonal period> arguments agree.
308 I<alpha> is the adaption parameter of the intercept (or baseline)
309 coefficient in the Holt-Winters forecasting algorithm. See L<rrdtool> for a
310 description of this algorithm. I<alpha> must lie between 0 and 1. A value
311 closer to 1 means that more recent observations carry greater weight in
312 predicting the baseline component of the forecast. A value closer to 0 means
313 that past history carries greater weight in predicting the baseline
314 component.
316 I<beta> is the adaption parameter of the slope (or linear trend) coefficient
317 in the Holt-Winters forecasting algorithm. I<beta> must lie between 0 and 1
318 and plays the same role as I<alpha> with respect to the predicted linear
319 trend.
321 I<gamma> is the adaption parameter of the seasonal coefficients in the
322 Holt-Winters forecasting algorithm (HWPREDICT) or the adaption parameter in
323 the exponential smoothing update of the seasonal deviations. It must lie
324 between 0 and 1. If the SEASONAL and DEVSEASONAL B<RRAs> are created
325 implicitly, they will both have the same value for I<gamma>: the value
326 specified for the HWPREDICT I<alpha> argument. Note that because there is
327 one seasonal coefficient (or deviation) for each time point during the
328 seasonal cycle, the adaptation rate is much slower than the baseline. Each
329 seasonal coefficient is only updated (or adapts) when the observed value
330 occurs at the offset in the seasonal cycle corresponding to that
331 coefficient.
333 If SEASONAL and DEVSEASONAL B<RRAs> are created explicitly, I<gamma> need not
334 be the same for both. Note that I<gamma> can also be changed via the
335 B<RRDtool> I<tune> command.
337 I<smoothing-window> specifies the fraction of a season that should be
338 averaged around each point. By default, the value of I<smoothing-window> is
339 0.05, which means each value in SEASONAL and DEVSEASONAL will be occasionally
340 replaced by averaging it with its (I<seasonal period>*0.05) nearest neighbors.
341 Setting I<smoothing-window> to zero will disable the running-average smoother
342 altogether.
344 I<rra-num> provides the links between related B<RRAs>. If HWPREDICT is
345 specified alone and the other B<RRAs> are created implicitly, then
346 there is no need to worry about this argument. If B<RRAs> are created
347 explicitly, then carefully pay attention to this argument. For each
348 B<RRA> which includes this argument, there is a dependency between
349 that B<RRA> and another B<RRA>. The I<rra-num> argument is the 1-based
350 index in the order of B<RRA> creation (that is, the order they appear
351 in the I<create> command). The dependent B<RRA> for each B<RRA>
352 requiring the I<rra-num> argument is listed here:
354 =over
356 =item *
358 HWPREDICT I<rra-num> is the index of the SEASONAL B<RRA>.
360 =item *
362 SEASONAL I<rra-num> is the index of the HWPREDICT B<RRA>.
364 =item *
366 DEVPREDICT I<rra-num> is the index of the DEVSEASONAL B<RRA>.
368 =item *
370 DEVSEASONAL I<rra-num> is the index of the HWPREDICT B<RRA>.
372 =item *
374 FAILURES I<rra-num> is the index of the DEVSEASONAL B<RRA>.
376 =back
378 I<threshold> is the minimum number of violations (observed values outside
379 the confidence bounds) within a window that constitutes a failure. If the
380 FAILURES B<RRA> is implicitly created, the default value is 7.
382 I<window length> is the number of time points in the window. Specify an
383 integer greater than or equal to the threshold and less than or equal to 28.
384 The time interval this window represents depends on the interval between
385 primary data points. If the FAILURES B<RRA> is implicitly created, the
386 default value is 9.
388 =head1 The HEARTBEAT and the STEP
390 Here is an explanation by Don Baarda on the inner workings of RRDtool.
391 It may help you to sort out why all this *UNKNOWN* data is popping
392 up in your databases:
394 RRDtool gets fed samples/updates at arbitrary times. From these it builds Primary
395 Data Points (PDPs) on every "step" interval. The PDPs are
396 then accumulated into the RRAs.
398 The "heartbeat" defines the maximum acceptable interval between
399 samples/updates. If the interval between samples is less than "heartbeat",
400 then an average rate is calculated and applied for that interval. If
401 the interval between samples is longer than "heartbeat", then that
402 entire interval is considered "unknown". Note that there are other
403 things that can make a sample interval "unknown", such as the rate
404 exceeding limits, or a sample that was explicitly marked as unknown.
406 The known rates during a PDP's "step" interval are used to calculate
407 an average rate for that PDP. If the total "unknown" time accounts for
408 more than B<half> the "step", the entire PDP is marked
409 as "unknown". This means that a mixture of known and "unknown" sample
410 times in a single PDP "step" may or may not add up to enough "known"
411 time to warrent for a known PDP.
413 The "heartbeat" can be short (unusual) or long (typical) relative to
414 the "step" interval between PDPs. A short "heartbeat" means you
415 require multiple samples per PDP, and if you don't get them mark the
416 PDP unknown. A long heartbeat can span multiple "steps", which means
417 it is acceptable to have multiple PDPs calculated from a single
418 sample. An extreme example of this might be a "step" of 5 minutes and a
419 "heartbeat" of one day, in which case a single sample every day will
420 result in all the PDPs for that entire day period being set to the
421 same average rate. I<-- Don Baarda E<lt>don.baarda@baesystems.comE<gt>>
423 time|
424 axis|
425 begin__|00|
426 |01|
427 u|02|----* sample1, restart "hb"-timer
428 u|03| /
429 u|04| /
430 u|05| /
431 u|06|/ "hbt" expired
432 u|07|
433 |08|----* sample2, restart "hb"
434 |09| /
435 |10| /
436 u|11|----* sample3, restart "hb"
437 u|12| /
438 u|13| /
439 step1_u|14| /
440 u|15|/ "swt" expired
441 u|16|
442 |17|----* sample4, restart "hb", create "pdp" for step1 =
443 |18| / = unknown due to 10 "u" labled secs > 0.5 * step
444 |19| /
445 |20| /
446 |21|----* sample5, restart "hb"
447 |22| /
448 |23| /
449 |24|----* sample6, restart "hb"
450 |25| /
451 |26| /
452 |27|----* sample7, restart "hb"
453 step2__|28| /
454 |22| /
455 |23|----* sample8, restart "hb", create "pdp" for step1, create "cdp"
456 |24| /
457 |25| /
459 graphics by I<vladimir.lavrov@desy.de>.
462 =head1 HOW TO MEASURE
464 Here are a few hints on how to measure:
466 =over
469 =item Temperature
471 Usually you have some type of meter you can read to get the temperature.
472 The temperature is not really connected with a time. The only connection is
473 that the temperature reading happened at a certain time. You can use the
474 B<GAUGE> data source type for this. RRDtool will then record your reading
475 together with the time.
477 =item Mail Messages
479 Assume you have a method to count the number of messages transported by
480 your mailserver in a certain amount of time, giving you data like '5
481 messages in the last 65 seconds'. If you look at the count of 5 like an
482 B<ABSOLUTE> data type you can simply update the RRD with the number 5 and the
483 end time of your monitoring period. RRDtool will then record the number of
484 messages per second. If at some later stage you want to know the number of
485 messages transported in a day, you can get the average messages per second
486 from RRDtool for the day in question and multiply this number with the
487 number of seconds in a day. Because all math is run with Doubles, the
488 precision should be acceptable.
490 =item It's always a Rate
492 RRDtool stores rates in amount/second for COUNTER, DERIVE and ABSOLUTE
493 data. When you plot the data, you will get on the y axis
494 amount/second which you might be tempted to convert to an absolute
495 amount by multiplying by the delta-time between the points. RRDtool
496 plots continuous data, and as such is not appropriate for plotting
497 absolute amounts as for example "total bytes" sent and received in a
498 router. What you probably want is plot rates that you can scale to
499 bytes/hour, for example, or plot absolute amounts with another tool
500 that draws bar-plots, where the delta-time is clear on the plot for
501 each point (such that when you read the graph you see for example GB
502 on the y axis, days on the x axis and one bar for each day).
504 =back
507 =head1 EXAMPLE
509 rrdtool create temperature.rrd --step 300 \
510 DS:temp:GAUGE:600:-273:5000 \
511 RRA:AVERAGE:0.5:1:1200 \
512 RRA:MIN:0.5:12:2400 \
513 RRA:MAX:0.5:12:2400 \
514 RRA:AVERAGE:0.5:12:2400
516 This sets up an B<RRD> called F<temperature.rrd> which accepts one
517 temperature value every 300 seconds. If no new data is supplied for
518 more than 600 seconds, the temperature becomes I<*UNKNOWN*>. The
519 minimum acceptable value is -273 and the maximum is 5'000.
521 A few archive areas are also defined. The first stores the
522 temperatures supplied for 100 hours (1'200 * 300 seconds = 100
523 hours). The second RRA stores the minimum temperature recorded over
524 every hour (12 * 300 seconds = 1 hour), for 100 days (2'400 hours). The
525 third and the fourth RRA's do the same for the maximum and
526 average temperature, respectively.
528 =head1 EXAMPLE 2
530 rrdtool create monitor.rrd --step 300 \
531 DS:ifOutOctets:COUNTER:1800:0:4294967295 \
532 RRA:AVERAGE:0.5:1:2016 \
533 RRA:HWPREDICT:1440:0.1:0.0035:288
535 This example is a monitor of a router interface. The first B<RRA> tracks the
536 traffic flow in octets; the second B<RRA> generates the specialized
537 functions B<RRAs> for aberrant behavior detection. Note that the I<rra-num>
538 argument of HWPREDICT is missing, so the other B<RRAs> will implicitly be
539 created with default parameter values. In this example, the forecasting
540 algorithm baseline adapts quickly; in fact the most recent one hour of
541 observations (each at 5 minute intervals) accounts for 75% of the baseline
542 prediction. The linear trend forecast adapts much more slowly. Observations
543 made during the last day (at 288 observations per day) account for only
544 65% of the predicted linear trend. Note: these computations rely on an
545 exponential smoothing formula described in the LISA 2000 paper.
547 The seasonal cycle is one day (288 data points at 300 second intervals), and
548 the seasonal adaption parameter will be set to 0.1. The RRD file will store 5
549 days (1'440 data points) of forecasts and deviation predictions before wrap
550 around. The file will store 1 day (a seasonal cycle) of 0-1 indicators in
551 the FAILURES B<RRA>.
553 The same RRD file and B<RRAs> are created with the following command,
554 which explicitly creates all specialized function B<RRAs>.
556 rrdtool create monitor.rrd --step 300 \
557 DS:ifOutOctets:COUNTER:1800:0:4294967295 \
558 RRA:AVERAGE:0.5:1:2016 \
559 RRA:HWPREDICT:1440:0.1:0.0035:288:3 \
560 RRA:SEASONAL:288:0.1:2 \
561 RRA:DEVPREDICT:1440:5 \
562 RRA:DEVSEASONAL:288:0.1:2 \
563 RRA:FAILURES:288:7:9:5
565 Of course, explicit creation need not replicate implicit create, a
566 number of arguments could be changed.
568 =head1 EXAMPLE 3
570 rrdtool create proxy.rrd --step 300 \
571 DS:Total:DERIVE:1800:0:U \
572 DS:Duration:DERIVE:1800:0:U \
573 DS:AvgReqDur:COMPUTE:Duration,Requests,0,EQ,1,Requests,IF,/ \
574 RRA:AVERAGE:0.5:1:2016
576 This example is monitoring the average request duration during each 300 sec
577 interval for requests processed by a web proxy during the interval.
578 In this case, the proxy exposes two counters, the number of requests
579 processed since boot and the total cumulative duration of all processed
580 requests. Clearly these counters both have some rollover point, but using the
581 DERIVE data source also handles the reset that occurs when the web proxy is
582 stopped and restarted.
584 In the B<RRD>, the first data source stores the requests per second rate
585 during the interval. The second data source stores the total duration of all
586 requests processed during the interval divided by 300. The COMPUTE data source
587 divides each PDP of the AccumDuration by the corresponding PDP of
588 TotalRequests and stores the average request duration. The remainder of the
589 RPN expression handles the divide by zero case.
591 =head1 AUTHOR
593 Tobias Oetiker E<lt>tobi@oetiker.chE<gt>