Re: [RFD 3/5] tracing: Add option to quantize key values
On Thu, 2015-04-30 at 12:06 +0200, Daniel Wagner wrote:
> Let's group some values together. This avoids a too detailed
> histogram. Some sort of logarythmic scale could be useful
> for latency plots.
>
> Now we can write something like:
>
> 'hist:key=latency.bucket:val=hitcount:sort=latency'
>
> latency: 0 hitcount: 166440
> latency:256 hitcount: 21104
> latency:512 hitcount: 7754
> latency:768 hitcount: 3269
> latency: 1024 hitcount: 1647
> latency: 1280 hitcount:841
> latency: 1536 hitcount:524
> latency: 1792 hitcount:371
> latency: 2048 hitcount:302
> latency: 2304 hitcount:240
> latency: 2560 hitcount:207
> latency: 2816 hitcount:149
> latency: 3072 hitcount:123
> latency: 3328 hitcount:119
> latency: 3584 hitcount:102
> latency: 3840 hitcount: 94
> latency: 4096 hitcount: 89
> latency: 4352 hitcount: 79
> latency: 4608 hitcount: 88
>
Nice addition!
> One thing I struggled with the grammatic above is that I haven't found
> a nice way to pass in arguments, for example the bucket size. There a lot
> of options to do it. Just a couple random ideas, not necessarly consistent
> or clever:
>
> 'hist:key=latency.bucket[10,1.5]:val=hitcount:sort=latency'
>where [x,y]: x first bucket size, y scaling factor
>
I like this notation - it's consistent with the other uses of the dot
notation in that it's modifying the way things are displayed, in this
case displaying latency as a histogram with specific [non-default]
parameters.
Tom
> 'hist:key=latency:val=hitcount:sort=latency:bucket=latency,10,1.5'
>
> Not for inclusion!
>
> Not-Signed-off-by: Daniel Wagner
> ---
> kernel/trace/trace_events_hist.c | 8
> 1 file changed, 8 insertions(+)
>
> diff --git a/kernel/trace/trace_events_hist.c
> b/kernel/trace/trace_events_hist.c
> index fe06707..cac94a6 100644
> --- a/kernel/trace/trace_events_hist.c
> +++ b/kernel/trace/trace_events_hist.c
> @@ -84,6 +84,7 @@ enum hist_field_flags {
> HIST_FIELD_STRING = 8,
> HIST_FIELD_EXECNAME = 16,
> HIST_FIELD_SYSCALL = 32,
> + HIST_FIELD_BUCKET = 64,
> };
>
> struct hist_trigger_sort_key {
> @@ -400,6 +401,8 @@ static int create_key_field(struct hist_trigger_data
> *hist_data,
> flags |= HIST_FIELD_EXECNAME;
> else if (!strcmp(field_str, "syscall"))
> flags |= HIST_FIELD_SYSCALL;
> + else if (!strcmp(field_str, "bucket"))
> + flags |= HIST_FIELD_BUCKET;
> }
>
> field = trace_find_event_field(file->event_call, field_name);
> @@ -900,6 +903,9 @@ static void event_hist_trigger(struct event_trigger_data
> *data, void *rec)
> key = entries;
> } else {
> field_contents = hist_data->key->fn(hist_data->key, rec);
> + if (hist_data->key->flags & HIST_FIELD_BUCKET)
> + field_contents &= ~0xff;
> +
> if (hist_data->key->flags & HIST_FIELD_STRING)
> key = (void *)field_contents;
> else
> @@ -1343,6 +1349,8 @@ static const char *get_hist_field_flags(struct
> hist_field *hist_field)
> flags_str = "hex";
> else if (hist_field->flags & HIST_FIELD_SYSCALL)
> flags_str = "syscall";
> + else if (hist_field->flags & HIST_FIELD_BUCKET)
> + flags_str = "bucket";
> else if (hist_field->flags & HIST_FIELD_EXECNAME)
> flags_str = "execname";
>
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majord...@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
Re: [RFD 3/5] tracing: Add option to quantize key values
On Thu, 2015-04-30 at 12:06 +0200, Daniel Wagner wrote:
Let's group some values together. This avoids a too detailed
histogram. Some sort of logarythmic scale could be useful
for latency plots.
Now we can write something like:
'hist:key=latency.bucket:val=hitcount:sort=latency'
latency: 0 hitcount: 166440
latency:256 hitcount: 21104
latency:512 hitcount: 7754
latency:768 hitcount: 3269
latency: 1024 hitcount: 1647
latency: 1280 hitcount:841
latency: 1536 hitcount:524
latency: 1792 hitcount:371
latency: 2048 hitcount:302
latency: 2304 hitcount:240
latency: 2560 hitcount:207
latency: 2816 hitcount:149
latency: 3072 hitcount:123
latency: 3328 hitcount:119
latency: 3584 hitcount:102
latency: 3840 hitcount: 94
latency: 4096 hitcount: 89
latency: 4352 hitcount: 79
latency: 4608 hitcount: 88
Nice addition!
One thing I struggled with the grammatic above is that I haven't found
a nice way to pass in arguments, for example the bucket size. There a lot
of options to do it. Just a couple random ideas, not necessarly consistent
or clever:
'hist:key=latency.bucket[10,1.5]:val=hitcount:sort=latency'
where [x,y]: x first bucket size, y scaling factor
I like this notation - it's consistent with the other uses of the dot
notation in that it's modifying the way things are displayed, in this
case displaying latency as a histogram with specific [non-default]
parameters.
Tom
'hist:key=latency:val=hitcount:sort=latency:bucket=latency,10,1.5'
Not for inclusion!
Not-Signed-off-by: Daniel Wagner daniel.wag...@bmw-carit.de
---
kernel/trace/trace_events_hist.c | 8
1 file changed, 8 insertions(+)
diff --git a/kernel/trace/trace_events_hist.c
b/kernel/trace/trace_events_hist.c
index fe06707..cac94a6 100644
--- a/kernel/trace/trace_events_hist.c
+++ b/kernel/trace/trace_events_hist.c
@@ -84,6 +84,7 @@ enum hist_field_flags {
HIST_FIELD_STRING = 8,
HIST_FIELD_EXECNAME = 16,
HIST_FIELD_SYSCALL = 32,
+ HIST_FIELD_BUCKET = 64,
};
struct hist_trigger_sort_key {
@@ -400,6 +401,8 @@ static int create_key_field(struct hist_trigger_data
*hist_data,
flags |= HIST_FIELD_EXECNAME;
else if (!strcmp(field_str, syscall))
flags |= HIST_FIELD_SYSCALL;
+ else if (!strcmp(field_str, bucket))
+ flags |= HIST_FIELD_BUCKET;
}
field = trace_find_event_field(file-event_call, field_name);
@@ -900,6 +903,9 @@ static void event_hist_trigger(struct event_trigger_data
*data, void *rec)
key = entries;
} else {
field_contents = hist_data-key-fn(hist_data-key, rec);
+ if (hist_data-key-flags HIST_FIELD_BUCKET)
+ field_contents = ~0xff;
+
if (hist_data-key-flags HIST_FIELD_STRING)
key = (void *)field_contents;
else
@@ -1343,6 +1349,8 @@ static const char *get_hist_field_flags(struct
hist_field *hist_field)
flags_str = hex;
else if (hist_field-flags HIST_FIELD_SYSCALL)
flags_str = syscall;
+ else if (hist_field-flags HIST_FIELD_BUCKET)
+ flags_str = bucket;
else if (hist_field-flags HIST_FIELD_EXECNAME)
flags_str = execname;
--
To unsubscribe from this list: send the line unsubscribe linux-kernel in
the body of a message to majord...@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
