GitLab maintenance scheduled for Tomorrow, 2019-09-24, from 12:00 to 13:00 CT - Services will be unavailable during this time.

Commit 8438ee10 authored by Shane Snyder's avatar Shane Snyder

add first try at BG/Q logutil aggregation function

We don't currently have a system to try this out on, but just
wanted to make sure something was there as a failsafe. The current
implementation should be able to correctly aggregate all counters
in BGQ records, except for BGQ_INODES.
parent 6a6f306a
......@@ -267,7 +267,43 @@ static void darshan_log_print_bgq_rec_diff(void *file_rec1, char *file_name1,
static void darshan_log_agg_bgq_recs(void *rec, void *agg_rec, int init_flag)
{
/* TODO: how would aggregation work for the BG/Q module ? */
struct darshan_bgq_record *bgq_rec = (struct darshan_bgq_record *)rec;
struct darshan_bgq_record *agg_bgq_rec = (struct darshan_bgq_record *)agg_rec;
int i;
if(init_flag)
{
/* when initializing, just copy over the first record */
memcpy(agg_bgq_rec, bgq_rec, sizeof(struct darshan_bgq_record));
/* TODO: each record stores the ID of the corresponding rank's BG/Q
* inode. Currently, this log aggregation interface assumes we can
* aggregate logs one at a time, without having to know the value of
* a counter on all processes. What we need here is a way to determine
* the inode IDs used for every process, filter out duplicates, then
* count the total number to set this counter. Will have to think
* more about how we can calculate this value using this interface
*/
agg_bgq_rec->counters[BGQ_INODES] = -1;
}
else
{
/* for remaining records, just sanity check the records are identical */
for(i = 0; i < BGQ_NUM_INDICES; i++)
{
/* TODO: ignore BGQ_INODES counter since it might be different in
* each record (more details in note above)
*/
if(i == BGQ_INODES)
continue;
assert(bgq_rec->counters[i] == agg_bgq_rec->counters[i]);
}
/* NOTE: ignore BGQ_F_TIMESTAMP counter -- just use the value from the
* record that we initialized with
*/
}
return;
}
......
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment