 |
The FreeRADIUS server
$Id: 15bac2a4c627c01d1aa2047687b3418955ac7f00 $
|
|
The FreeRADIUS server
$Id: 15bac2a4c627c01d1aa2047687b3418955ac7f00 $
|
Load generation. More...
#include <freeradius-devel/util/event.h>#include <freeradius-devel/util/talloc.h>
Include dependency graph for load.h:Go to the source code of this file.
Data Structures | |
| struct | fr_load_config_t |
| Load generation configuration. More... | |
| struct | fr_load_stats_t |
Typedefs | |
| typedef int(* | fr_load_callback_t) (fr_time_t now, void *uctx) |
| typedef struct fr_load_s | fr_load_t |
Enumerations | |
| enum | fr_load_reply_t { FR_LOAD_CONTINUE = 0 , FR_LOAD_DONE } |
| Whether or not the application should continue. More... | |
Functions | |
| fr_load_t * | fr_load_generator_create (TALLOC_CTX *ctx, fr_event_list_t *el, fr_load_config_t *config, fr_load_callback_t callback, void *uctx)) |
| fr_load_reply_t | fr_load_generator_have_reply (fr_load_t *l, fr_time_t request_time) |
| Tell the load generator that we have a reply to a packet we sent. More... | |
| int | fr_load_generator_start (fr_load_t *l) |
| Start the load generator. More... | |
| fr_load_stats_t const * | fr_load_generator_stats (fr_load_t const *l) |
| size_t | fr_load_generator_stats_sprint (fr_load_t *l, fr_time_t now, char *buffer, size_t buflen) |
| Print load generator statistics in CVS format. More... | |
| int | fr_load_generator_stop (fr_load_t *l) |
| Stop the load generation through the simple expedient of deleting the timer associated with it. More... | |
Load generation.
Definition in file load.h.
| struct fr_load_config_t |
Load generation configuration.
The load generator runs a callback periodically in order to generate load. The callback MUST do all of the work, and track all necessary state itself. The load generator simply provides a periodic signal.
The load begins with "start_pps", and ends after ramping up to "max_pps", no matter how long that takes. The ramp-up is done by "step" increments. Each step is run for "duration" seconds.
The callback is run "1/pps" times per second.
In order to send higher load, it is possible to run the callback "parallel" times per timeout. i.e. with "start_pps = 100", and "parallel = 10", the load generator will run the callback 10 times, wait 1/10s, run the callback another 10 times, and so on.
In order to prevent the load generator from overloading the backend, we have a configurable maximum backlog. i.e. packets sent without reply. This backlog is expressed in milliseconds of packets, not in numbers of packets. Expressing the backlog this way allows it to automatically scale to higher loads.
i.e. if the generator is senting 10K packets/s, and the "milliseconds" parameter is 1000, then the generator will allow 10K packets in the backlog.
Once the backlog limit is reached, the load generator will switch to a "gated" method of sending packets. It will only send one new packet when it has received a reply for one old packet.
If the generator receives many replies and the backlog is lower than the limit, the generator switches again to sending the configured "pps" packets
The generator will try to increase the packet rate after "duration" seconds, even if the maximum backlog is currently reached. This increase has the effect of also increasing the maximum backlog.
Collaboration diagram for fr_load_config_t:| Data Fields | ||
|---|---|---|
| fr_time_delta_t | duration | duration of each step |
| uint32_t | max_pps | max PPS, 0 for "no limit". |
| uint32_t | milliseconds | how many milliseconds of backlog to top out at |
| uint32_t | parallel | how many packets in parallel to send |
| uint32_t | start_pps | start PPS |
| uint32_t | step | how much to increase each load test by |
| struct fr_load_stats_t |
Collaboration diagram for fr_load_stats_t:| Data Fields | ||
|---|---|---|
| int | backlog | current backlog |
| bool | blocked | whether or not we're blocked |
| fr_time_t | end |
when the test started when the test ended, due to last reply received |
| fr_time_t | last_send | last packet we sent |
| int | max_backlog | maximum backlog we saw during the test |
| int | pps | current offered packets/s |
| int | pps_accepted | Accepted PPS for the last second. |
| int | received | total packets received (should be == sent) |
| fr_time_delta_t | rtt | smoothed round trip time |
| fr_time_delta_t | rttvar | RTT variation. |
| int | sent | total packets sent |
| int | skipped | we skipped sending this number of packets |
| fr_time_t | start | |
| int | times[8] | response time in microseconds to tens of seconds |
| enum fr_load_reply_t |
| fr_load_t* fr_load_generator_create | ( | TALLOC_CTX * | ctx, |
| fr_event_list_t * | el, | ||
| fr_load_config_t * | config, | ||
| fr_load_callback_t | callback, | ||
| void * | uctx | ||
| ) |
| fr_load_reply_t fr_load_generator_have_reply | ( | fr_load_t * | l, |
| fr_time_t | request_time | ||
| ) |
| int fr_load_generator_start | ( | fr_load_t * | l | ) |
| fr_load_stats_t const* fr_load_generator_stats | ( | fr_load_t const * | l | ) |
1.9.1