Requests and limits

This section provides a brief overview of the service performance and recommendations about CPU and RAM requests and limits in order to reach optimal performances on common scenarios.

Results have been obtained through specific tests implementing a series of requests to the Appointment Manager Service.

Specifically, the following APIs have been tested to measure CPU and RAM usage when dynamically computing slots:

• GET /calendar/count: return the number of events for all resources in a given period;
• GET /calendar/: return the events (availabilities with slots, exceptions and appointments) for all resources in a given period;
• GET /slots/: return the slots for a resource in a given period.

We designed the tests so that, following our suggested requests and limits, you can serve more users by simply scaling the service horizontally, i.e. increasing the number of replicas proportionally to the expected traffic.

Scenarios

The performance of the Appointment Manager, and the aforementioned endpoints in particular, depend on the following variables:

• the period: we simulated a client fetching calendar events for a single day, week or month;
• the number of resources: we simulated a client fetching events for 5 or 25 resources;
• the availability frequency and duration: we simulated a daily availability from 7:00 to midnight for each resource;
• the slot duration: we simulated slots having a fixed duration of 15 minutes in the first test suite and flexible duration (15, 30 and 60 minutes) in the second.

Variable resources

In the first test suite we focus on how the number of resources affects the performance.

The following table summarizes our findings:

Users	Resources	Daily slots	Weekly slots	Monthly slots	Requests Per Second	CPU Requests	CPU Limits	Memory Requests	Memory Limits
1	5	340	2380	10540	1	250	1000	100	250
5	5	340	2380	10540	3	250	1000	100	250
5	5	340	2380	10540	3	500	1500	100	250
10	5	340	2380	10540	6	500	1500	100	250
20	5	340	2380	10540	8	500	1500	100	250
1	25	1700	11900	52700	1	500	1500	100	250
5	25	1700	11900	52700	<1	500	1500	100	250

We recommend starting with a single replica configured as follows:

• CPU request: 500 Mi
• CPU limit: 1500 Mi
• RAM request: 100 m
• RAM limit: 250 m

and adding more replicas as needed according to the number of users and resources.

Variable slots

In the next test suites we focus on how the number of slots affects the performance, averaging the number of resources to ten in all test cases.

15 minutes

Users	Resources	Daily slots	Weekly slots	Monthly slots	Requests Per Second	CPU Requests	CPU Limits	Memory Requests	Memory Limits
1	10	680	4760	21080	1	250	1000	100	200
5	10	680	4760	21080	4	250	1000	100	200
10	10	680	4760	21080	6	250	1000	100	200
20	10	680	4760	21080	6	250	1000	100	200
50	10	680	4760	21080	6	250	1000	100	200

30 minutes

Users	Resources	Daily slots	Weekly slots	Monthly slots	Requests Per Second	CPU Requests	CPU Limits	Memory Requests	Memory Limits
1	10	340	2380	10540	1	250	1000	100	200
5	10	340	2380	10540	4	250	1000	100	200
10	10	340	2380	10540	6	250	1000	100	200
20	10	340	2380	10540	8	250	1000	100	200
50	10	340	2380	10540	5	250	1000	100	200

60 minutes

Users	Resources	Daily slots	Weekly slots	Monthly slots	Requests Per Second	CPU Requests	CPU Limits	Memory Requests	Memory Limits
1	10	170	1190	5270	1	250	1000	100	200
5	10	170	1190	5270	4	250	1000	100	200
10	10	170	1190	5270	6	250	1000	100	200
20	10	170	1190	5270	8	250	1000	100	200
50	10	170	1190	5270	8	250	1000	100	200

Variable CPU requests and limits

For the third test suite, we scaled up and down the microservice CPU requests and limits to assess the minimum CPU requirements.

Users	Resources	Daily slots	Weekly slots	Monthly slots	Requests Per Second	Average response time (ms)	90%ile (ms)	CPU Requests	CPU Limits	Memory Requests	Memory Limits	Warnings
5	10	680	4760	21080	3	330	970	500	1000	50	150	None
5	10	680	4760	21080	3	721	1900	250	500	50	150	None
5	10	680	4760	21080	2	1431	3700	200	400	50	150	CPU throttling, <1% failures
5	10	680	4760	21080	1	1635	3600	150	300	50	150	Pod restart, CPU throttling, 1% failures

Minimum and recommended configurations

The following table provides some recommendations, based on the results of the load tests. We suggest you start with the best matching configuration, according to the number of expected simultaneous users. The single user configuration represents the minimum requirements, according to our internal tests. You can easily start from any of the following configurations and then fine tune it according to your specific needs.

Number of concurrent users	CPU Requests	CPU Limits	Memory Requests	Memory Limits
Single user	50	100	50	100
2-5 users	250	500	50	100
5-15 users	500	1000	100	150
20+ users	1000	1500	100	150