[v3,3/5] latencystats: do not use floating point
Checks
Commit Message
The cycle counts do not need to be stored as floating point.
Instead keep track of latency in cycles, and convert to
nanoseconds when read.
Change Exponential Weighted Moving Average weight from .2 to .25
to avoid use of floating point for that.
Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
---
lib/latencystats/rte_latencystats.c | 37 +++++++++++------------------
1 file changed, 14 insertions(+), 23 deletions(-)
Comments
On Wed, Apr 17, 2024 at 10:07:25AM -0700, Stephen Hemminger wrote:
> The cycle counts do not need to be stored as floating point.
> Instead keep track of latency in cycles, and convert to
> nanoseconds when read.
>
> Change Exponential Weighted Moving Average weight from .2 to .25
> to avoid use of floating point for that.
>
> Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
> ---
> lib/latencystats/rte_latencystats.c | 37 +++++++++++------------------
> 1 file changed, 14 insertions(+), 23 deletions(-)
>
> diff --git a/lib/latencystats/rte_latencystats.c b/lib/latencystats/rte_latencystats.c
> index fe8c3c563a..11bd0ea4ae 100644
> --- a/lib/latencystats/rte_latencystats.c
> +++ b/lib/latencystats/rte_latencystats.c
> @@ -47,10 +47,10 @@ static uint64_t timer_tsc;
> static uint64_t prev_tsc;
>
> struct rte_latency_stats {
> - float min_latency; /**< Minimum latency in nano seconds */
> - float avg_latency; /**< Average latency in nano seconds */
> - float max_latency; /**< Maximum latency in nano seconds */
> - float jitter; /** Latency variation */
> + uint64_t min_latency; /**< Minimum latency */
> + uint64_t avg_latency; /**< Average latency */
> + uint64_t max_latency; /**< Maximum latency */
> + uint64_t jitter; /** Latency variation */
> rte_spinlock_t lock; /** Latency calculation lock */
> };
>
> @@ -82,13 +82,12 @@ int32_t
> rte_latencystats_update(void)
> {
> unsigned int i;
> - float *stats_ptr = NULL;
> uint64_t values[NUM_LATENCY_STATS] = {0};
> int ret;
>
> for (i = 0; i < NUM_LATENCY_STATS; i++) {
> - stats_ptr = RTE_PTR_ADD(glob_stats,
> - lat_stats_strings[i].offset);
> + const uint64_t *stats_ptr = RTE_PTR_ADD(glob_stats,
> + lat_stats_strings[i].offset);
> values[i] = floor(*stats_ptr / latencystat_cycles_per_ns());
will this need a cast to uint64_t?
> }
>
> @@ -105,11 +104,10 @@ static void
> rte_latencystats_fill_values(struct rte_metric_value *values)
> {
> unsigned int i;
> - float *stats_ptr = NULL;
>
> for (i = 0; i < NUM_LATENCY_STATS; i++) {
> - stats_ptr = RTE_PTR_ADD(glob_stats,
> - lat_stats_strings[i].offset);
> + const uint64_t *stats_ptr = RTE_PTR_ADD(glob_stats,
> + lat_stats_strings[i].offset);
> values[i].key = i;
> values[i].value = floor(*stats_ptr / latencystat_cycles_per_ns());
> }
> @@ -156,15 +154,8 @@ calc_latency(uint16_t pid __rte_unused,
> void *_ __rte_unused)
> {
> unsigned int i;
> - uint64_t now;
> - float latency;
> - static float prev_latency;
> - /*
> - * Alpha represents degree of weighting decrease in EWMA,
> - * a constant smoothing factor between 0 and 1. The value
> - * is used below for measuring average latency.
> - */
> - const float alpha = 0.2;
> + uint64_t now, latency;
> + static uint64_t prev_latency;
>
if this is merged i can abandon this series
https://patchwork.dpdk.org/project/dpdk/list/?series=31747
Acked-by: Tyler Retzlaff <roretzla@linux.microsoft.com>
@@ -47,10 +47,10 @@ static uint64_t timer_tsc;
static uint64_t prev_tsc;
struct rte_latency_stats {
- float min_latency; /**< Minimum latency in nano seconds */
- float avg_latency; /**< Average latency in nano seconds */
- float max_latency; /**< Maximum latency in nano seconds */
- float jitter; /** Latency variation */
+ uint64_t min_latency; /**< Minimum latency */
+ uint64_t avg_latency; /**< Average latency */
+ uint64_t max_latency; /**< Maximum latency */
+ uint64_t jitter; /** Latency variation */
rte_spinlock_t lock; /** Latency calculation lock */
};
@@ -82,13 +82,12 @@ int32_t
rte_latencystats_update(void)
{
unsigned int i;
- float *stats_ptr = NULL;
uint64_t values[NUM_LATENCY_STATS] = {0};
int ret;
for (i = 0; i < NUM_LATENCY_STATS; i++) {
- stats_ptr = RTE_PTR_ADD(glob_stats,
- lat_stats_strings[i].offset);
+ const uint64_t *stats_ptr = RTE_PTR_ADD(glob_stats,
+ lat_stats_strings[i].offset);
values[i] = floor(*stats_ptr / latencystat_cycles_per_ns());
}
@@ -105,11 +104,10 @@ static void
rte_latencystats_fill_values(struct rte_metric_value *values)
{
unsigned int i;
- float *stats_ptr = NULL;
for (i = 0; i < NUM_LATENCY_STATS; i++) {
- stats_ptr = RTE_PTR_ADD(glob_stats,
- lat_stats_strings[i].offset);
+ const uint64_t *stats_ptr = RTE_PTR_ADD(glob_stats,
+ lat_stats_strings[i].offset);
values[i].key = i;
values[i].value = floor(*stats_ptr / latencystat_cycles_per_ns());
}
@@ -156,15 +154,8 @@ calc_latency(uint16_t pid __rte_unused,
void *_ __rte_unused)
{
unsigned int i;
- uint64_t now;
- float latency;
- static float prev_latency;
- /*
- * Alpha represents degree of weighting decrease in EWMA,
- * a constant smoothing factor between 0 and 1. The value
- * is used below for measuring average latency.
- */
- const float alpha = 0.2;
+ uint64_t now, latency;
+ static uint64_t prev_latency;
now = rte_rdtsc();
@@ -186,8 +177,7 @@ calc_latency(uint16_t pid __rte_unused,
* Reference: Calculated as per RFC 5481, sec 4.1,
* RFC 3393 sec 4.5, RFC 1889 sec.
*/
- glob_stats->jitter += (fabsf(prev_latency - latency)
- - glob_stats->jitter)/16;
+ glob_stats->jitter += ((prev_latency - latency) - glob_stats->jitter) / 16;
if (glob_stats->min_latency == 0)
glob_stats->min_latency = latency;
else if (latency < glob_stats->min_latency)
@@ -198,9 +188,10 @@ calc_latency(uint16_t pid __rte_unused,
* The average latency is measured using exponential moving
* average, i.e. using EWMA
* https://en.wikipedia.org/wiki/Moving_average
+ *
+ * Alpha is .25
*/
- glob_stats->avg_latency +=
- alpha * (latency - glob_stats->avg_latency);
+ glob_stats->avg_latency += (latency - glob_stats->avg_latency) / 4;
prev_latency = latency;
}
rte_spinlock_unlock(&glob_stats->lock);