Skip to content

Commit

Permalink
Update node utilization documentation (#1397)
Browse files Browse the repository at this point in the history
  • Loading branch information
jtpalmer authored Jul 23, 2020
1 parent 759286a commit 4ef28b7
Showing 1 changed file with 87 additions and 16 deletions.
103 changes: 87 additions & 16 deletions docs/howto-node-utilization.md
Original file line number Diff line number Diff line change
Expand Up @@ -8,24 +8,95 @@ it is not appropriate for all resource configurations. Specifically, if
your resource allows node sharing, this statistic will not be accurate.
All nodes are assumed to be exclusive when this statistic is calculated.

To enable the statistic add an object to the `statistics` array for the
`Jobs` realm in `datawarehouse.json`:
First, add the SQL queries used by the statistic.

`/etc/xdmod/datawarehouse.d/include/Jobs-node-util-agg.sql`:

```sql
100.0 * (
COALESCE(
SUM(agg.node_time)
/
(
SELECT
SUM(ra.percent * inner_days.hours * rs.q_nodes / 100.0)
FROM
modw.resourcespecs rs,
modw.resource_allocated ra,
modw.days inner_days
WHERE
inner_days.id BETWEEN YEAR(FROM_UNIXTIME(ra.start_date_ts)) * 100000 + DAYOFYEAR(FROM_UNIXTIME(ra.start_date_ts))
AND COALESCE(YEAR(FROM_UNIXTIME(ra.end_date_ts)) * 100000 + DAYOFYEAR(FROM_UNIXTIME(ra.end_date_ts)), 999999999)
AND inner_days.id BETWEEN YEAR(FROM_UNIXTIME(rs.start_date_ts)) * 100000 + DAYOFYEAR(FROM_UNIXTIME(rs.start_date_ts))
AND COALESCE(YEAR(FROM_UNIXTIME(rs.end_date_ts)) * 100000 + DAYOFYEAR(FROM_UNIXTIME(rs.end_date_ts)), 999999999)
AND inner_days.id BETWEEN YEAR(FROM_UNIXTIME(${START_DATE_TS})) * 100000 + DAYOFYEAR(FROM_UNIXTIME(${START_DATE_TS}))
AND YEAR(FROM_UNIXTIME(${END_DATE_TS})) * 100000 + DAYOFYEAR(FROM_UNIXTIME(${END_DATE_TS}))
AND ra.resource_id = rs.resource_id
AND FIND_IN_SET(
rs.resource_id,
GROUP_CONCAT(DISTINCT agg.task_resource_id)
) <> 0
),
0
) / 3600.0
)
```

`/etc/xdmod/datawarehouse.d/include/Jobs-node-util-time.sql`:

```sql
100.0 * (
COALESCE(
SUM(agg.node_time)
/
(
SELECT
SUM(ra.percent * inner_days.hours * rs.q_nodes / 100.0)
FROM
modw.resourcespecs rs,
modw.resource_allocated ra,
modw.days inner_days
WHERE
inner_days.id BETWEEN YEAR(FROM_UNIXTIME(ra.start_date_ts)) * 100000 + DAYOFYEAR(FROM_UNIXTIME(ra.start_date_ts))
AND COALESCE(YEAR(FROM_UNIXTIME(ra.end_date_ts)) * 100000 + DAYOFYEAR(FROM_UNIXTIME(ra.end_date_ts)), 999999999)
AND inner_days.id BETWEEN YEAR(FROM_UNIXTIME(rs.start_date_ts)) * 100000 + DAYOFYEAR(FROM_UNIXTIME(rs.start_date_ts))
AND COALESCE(YEAR(FROM_UNIXTIME(rs.end_date_ts)) * 100000 + DAYOFYEAR(FROM_UNIXTIME(rs.end_date_ts)), 999999999)
AND inner_days.id BETWEEN YEAR(FROM_UNIXTIME(duration.${AGGREGATION_UNIT}_start_ts)) * 100000 + DAYOFYEAR(FROM_UNIXTIME(duration.${AGGREGATION_UNIT}_start_ts))
AND YEAR(FROM_UNIXTIME(duration.${AGGREGATION_UNIT}_end_ts)) * 100000 + DAYOFYEAR(FROM_UNIXTIME(duration.${AGGREGATION_UNIT}_end_ts))
AND ra.resource_id = rs.resource_id
AND FIND_IN_SET(
rs.resource_id,
GROUP_CONCAT(DISTINCT agg.task_resource_id)
) <> 0
),
0
) / 3600.0
)
```

Then add the statistic to the Jobs realm statistics configuration.

`/etc/xdmod/datawarehouse.d/ref/Jobs-statistics.json`:

```json
[
{
"realm": "Jobs",
...
"statistics": [
...
{
"name": "node_utilization",
"class": "NodeUtilizationStatistic"
}
]
{
...
"node_utilization": {
"name": "Node Utilization",
"description_html": "The percentage of node time that a resource has been running jobs.<br/><i>Node Utilization:</i> The ratio of the total node hours consumed by jobs over a given time period divided by the maximum node hours that the system could deliver (based on the number of nodes present on the resources). This value does not take into account downtimes or outages. It is just calculated based on the number of nodes in the resource specifications.",
"precision": 2,
"unit": "%",
"aggregate_formula": {
"$include": "datawarehouse.d/include/Jobs-node-util-agg.sql"
},
"timeseries_formula": {
"$include": "datawarehouse.d/include/Jobs-node-util-time.sql"
}
}
]
}
```

After that just reload the portal and the Node Utilization statistic
will appear in the list of Job statistics.
Last, run `acl-config` and restart your web server.

Reload the portal and the "Node Utilization" statistic will appear in the list
of Job statistics.

0 comments on commit 4ef28b7

Please sign in to comment.