Lookout for Metrics quotas
Amazon Lookout for Metrics sets quotas for the amount of data that a detector can use to learn and detect anomalies. There are also quotas for data import intervals, records processing, and Amazon Lookout for Metrics API requests. Additionally, there are data requirements for data retention for re-training, coldstart anomaly detection, backtesting, time series, and record field key value pairs.
Quotas set by other services can impact operation. For information on quotas for other services such as Amazon CloudWatch, see AWS service quotas in the Amazon Web Services General Reference.
Topics
Quotas
| Resource | Default quota | Adjustable | |
|---|---|---|---|
| Quotas for detectors | Maximum number of detectors |
10 |
Yes |
Maximum number of alerts for a detector |
10 |
Yes |
|
| Maximum number of files you can upload per interval | 5-minute interval |
5 files |
Yes |
10-minute interval |
5 files |
Yes |
|
1-hour interval |
10 files |
Yes |
|
1-day interval |
10 files |
Yes |
|
| Maximum number of records Lookout for Metrics can process per interval | 5-minute interval |
15000 records |
Yes |
10-minute interval |
24000 records |
Yes |
|
1-hour interval |
150000 records |
Yes |
|
1-day interval |
150000 records |
Yes |
|
| Maximum number of API requests per second | Each second |
10 requests |
Yes |
Maximum number of datasets for a detector |
1 dataset |
Yes |
|
| Quotas for datasets | Maximum number of dimensions for a dataset |
10 dimensions |
No |
Maximum number of measures for a dataset |
10 measures |
No |
|
| Quotas for historical data | Maximum intervals (continuous mode) |
2500 |
No |
Maximum intervals (backtest mode) |
3000 |
No |
|
Maximum number of files in historical data |
3000 |
No |
|
| Maximum number of metrics Lookout for Metrics can process per detector | 5-minute interval |
5000 metrics |
No |
10-minute interval |
10000 metrics |
No |
|
1-hour interval |
50000 metrics |
Yes |
|
1-day interval |
50000 metrics |
Yes |
To request a quota increase, use the Service Quotas console
Data retention time periods for re-training
The amount of time you should retain your data for re-training depends on the interval of your data collection as follows:
| Interval | Maximum time | Average time |
|---|---|---|
| 5 minutes | 9 days | 4.5 days |
| 10 minutes | 18 days | 9 days |
| 1 hour | 3.4 months | 2 months |
| 1 day | 5 years | 2 years |
Coldstart anomaly detection
For coldstart anomaly detection (no historical data), the amount of time it takes for Lookout for Metrics to detect anomalies depends on the interval of your data collection as follows:
-
5 minute interval – 25 hours
-
10 minute interval – 50 hours
-
1 hour interval – 4 days
-
1 day interval – 14 days
File size
The amount of data that the detector can process for an interval is limited. You can reduce the size of files by aggregating records or by removing fields that are not used as measures or dimensions.
-
5 minute interval – 200 MB
-
10 minute interval – 200 MB
-
1 hour interval – 200 MB
-
1 day interval – 200 MB
Backtesting data requirements
The ratio of training and testing data used for backtesting is 70% training and 30% testing. Amazon Lookout for Metrics allows a maximum of 3000 intervals of data, therefore the maximum number of data points you will see for backtesting is 900. The number of days of data Lookout for Metrics considers for backtesting depends on the interval of your data collection as follows:
-
5 minute interval – 3.125 days, with 12 data points per hour and 288 data points per day.
-
10 minute interval – 6.25 days, with 6 data points per hour and 144 data points per day.
-
1 hour interval – 37.5 days, with 1 data point per hour and 24 points per day.
-
1 day interval – 900 days, with 1 data point per day.
The minimum amount of data required for backtesting depends on the interval of your data collection as follows:
-
5 minute interval – ~1 day of data.
-
10 minute interval – 1.9 days of data.
-
1 hour interval – 11.8 days of data.
-
1 day interval – 285 days of data.
Record field key-value pair character limits
Character limits for record field key value pairs are as follows:
-
Dimensions key – 63 characters
-
Dimension value – 40 characters
-
Timestamp key – 63 characters
-
Measure value – 10^15 with precision of 4 after decimal point
