Baseline performance tests results

Results presented here show the maximum throughput of Kea using:

• A very basic Kea configuration
• The simplest traffic characteristics
• Results presented here showing top performance of single Kea setup (one server) and High Availability setup (two connected servers in the network)
• All tests here are type A - which are calculating top performance. Calculated rate will be used in tests type B.

Kea configuration

Tests here are using simplest configuration possible with default configuration.

• one subnet with one pool (/8 for v4 and /64 for v6)
• no options
• no host reservations
• no client classification
• lease lifetimes are longer than test duration (to prevent renewals)
• default values are used everywhere possible (e.g. congestion control disabled, LFC enabled for memfile)

Traffic characteristics

ISC's perfdhcp generates simple DORA/SARR exchanges without any additional options or option requests. Normally we configure perfdhcp to simulate up to 500 million different clients (to avoid repeating any client IDs in the course of a single test) unless it's stated differently in the test description.

Different Thread settings

These tests use the basic Kea configuration described in the baseline test report with the simplest traffic characteristics. Other than thread count, the configuration is exactly the same as that used in the "Baseline results" test.

1. The first of these tests (results in the bar charts below) measures how the different Kea backends respond to variation in the number of threads used. The results of this test show that different Kea configurations require different settings for `thread-pool-size` to achieve optimum performance, measured in leases per second.

2. The second test (results in the line charts below) are using the value of `thread-pool-size` which had the highest results for that scenario from the first test. We vary the `packet-queue-size` value and measure the leases per second . This test shows that the optimum queue size depends on the Kea configuration.

We use the thread count settings ("thread-pool-size" and "packet-queue-size") that provide the optimum results from this test to establish the thread count configuration for the other tests included in this report.

How various ways of host reservations configurations affect Kea performance

Each test has an additional description. Set of bar charts display results comparison of single runs presented in second part of this page.

Scenario: 1. Test key words: reservations default global 30 memfile v4

How 3000 (30% of all clients) global reservations kept in memfile decreases performance (leases in memfile)
Reservation optimization: None, default settings
MT settings: 8 threads, queue size 4 per thread

Scenario: 2. Test key words: reservations default global 30 memfile v6

How 3000 (30% of all clients) global reservations kept in memfile decreases performance (leases in memfile)
Reservation optimization: None, default settings
MT settings: 4 threads, queue size 4 per thread

Scenario: 3. Test key words: reservations default global 30 mysql v4

How 3000 (30% of all clients) global reservations kept in mysql decreases performance (leases in memfile)
Reservation optimization: None, default settings
MT settings: 14 threads, queue size 160 per thread

Scenario: 4. Test key words: reservations default global 30 mysql v6

How 3000 (30% of all clients) global reservations kept in mysql decreases performance (leases in memfile)
Reservation optimization: None, default settings
MT settings: 14 threads, queue size 160 per thread

Scenario: 5. Test key words: reservations default global 30 postgresql v4

How 3000 (30% of all clients) global reservations kept in postgresql decreases performance (leases in memfile)
Reservation optimization: None, default settings
MT settings: 8 threads, queue size 20 per thread

Scenario: 6. Test key words: reservations default global 30 postgresql v6

How 3000 (30% of all clients) global reservations kept in postgresql decreases performance (leases in memfile)
Reservation optimization: None, default settings
MT settings: 6 threads, queue size 20 per thread

Scenario: 7. Test key words: reservations default global 100 memfile v4

How 10000 (100% of all clients) global reservations kept in memfile decreases performance (leases in memfile)
Reservation optimization: None, default settings
MT settings: 8 threads, queue size 4 per thread

Scenario: 8. Test key words: reservations default global 100 memfile v6

How 10000 (100% of all clients) global reservations kept in memfile decreases performance (leases in memfile)
Reservation optimization: None, default settings
MT settings: 4 threads, queue size 4 per thread

Scenario: 9. Test key words: reservations default global 100 mysql v4

How 10000 (100% of all clients) global reservations kept in mysql decreases performance (leases in memfile)
Reservation optimization: None, default settings
MT settings: 14 threads, queue size 160 per thread

Scenario: 10. Test key words: reservations default global 100 mysql v6

How 10000 (100% of all clients) global reservations kept in mysql decreases performance (leases in memfile)
Reservation optimization: None, default settings
MT settings: 14 threads, queue size 160 per thread

Scenario: 11. Test key words: reservations default global 100 postgresql v4

How 10000 (100% of all clients) global reservations kept in postgresql decreases performance (leases in memfile)
Reservation optimization: None, default settings
MT settings: 8 threads, queue size 20 per thread

Scenario: 12. Test key words: reservations default global 100 postgresql v6

How 10000 (100% of all clients) global reservations kept in postgresql decreases performance (leases in memfile)
Reservation optimization: None, default settings
MT settings: 6 threads, queue size 20 per thread

Scenario: 13. Test key words: reservations default global 150 memfile v4

How 15000 (150% of all clients) global reservations kept in memfile decreases performance (leases in memfile)
Reservation optimization: None, default settings
MT settings: 8 threads, queue size 4 per thread

Scenario: 14. Test key words: reservations default global 150 memfile v6

How 15000 (150% of all clients) global reservations kept in memfile decreases performance (leases in memfile)
Reservation optimization: None, default settings
MT settings: 4 threads, queue size 4 per thread

Scenario: 15. Test key words: reservations default global 150 mysql v4

How 15000 (150% of all clients) global reservations kept in mysql decreases performance (leases in memfile)
Reservation optimization: None, default settings
MT settings: 14 threads, queue size 160 per thread

Scenario: 16. Test key words: reservations default global 150 mysql v6

How 15000 (150% of all clients) global reservations kept in mysql decreases performance (leases in memfile)
Reservation optimization: None, default settings
MT settings: 14 threads, queue size 160 per thread

Scenario: 17. Test key words: reservations default global 150 postgresql v4

How 15000 (150% of all clients) global reservations kept in postgresql decreases performance (leases in memfile)
Reservation optimization: None, default settings
MT settings: 8 threads, queue size 20 per thread

Scenario: 18. Test key words: reservations default global 150 postgresql v6

How 15000 (150% of all clients) global reservations kept in postgresql decreases performance (leases in memfile)
Reservation optimization: None, default settings
MT settings: 6 threads, queue size 20 per thread

Scenario: 19. Test key words: reservations default subnet 30 memfile v4

How 3000 (30% of all clients) subnet reservations kept in memfile decreases performance (leases in memfile)
Reservation optimization: None, default settings
MT settings: 8 threads, queue size 4 per thread

Scenario: 20. Test key words: reservations default subnet 30 memfile v6

How 3000 (30% of all clients) subnet reservations kept in memfile decreases performance (leases in memfile)
Reservation optimization: None, default settings
MT settings: 4 threads, queue size 4 per thread

Scenario: 21. Test key words: reservations default subnet 30 mysql v4

How 3000 (30% of all clients) subnet reservations kept in mysql decreases performance (leases in memfile)
Reservation optimization: None, default settings
MT settings: 14 threads, queue size 160 per thread

Scenario: 22. Test key words: reservations default subnet 30 mysql v6

How 3000 (30% of all clients) subnet reservations kept in mysql decreases performance (leases in memfile)
Reservation optimization: None, default settings
MT settings: 14 threads, queue size 160 per thread

Scenario: 23. Test key words: reservations default subnet 30 postgresql v4

How 3000 (30% of all clients) subnet reservations kept in postgresql decreases performance (leases in memfile)
Reservation optimization: None, default settings
MT settings: 8 threads, queue size 20 per thread

Scenario: 24. Test key words: reservations default subnet 30 postgresql v6

How 3000 (30% of all clients) subnet reservations kept in postgresql decreases performance (leases in memfile)
Reservation optimization: None, default settings
MT settings: 6 threads, queue size 20 per thread

Scenario: 25. Test key words: reservations default subnet 100 memfile v4

How 10000 (100% of all clients) subnet reservations kept in memfile decreases performance (leases in memfile)
Reservation optimization: None, default settings
MT settings: 8 threads, queue size 4 per thread

Scenario: 26. Test key words: reservations default subnet 100 memfile v6

How 10000 (100% of all clients) subnet reservations kept in memfile decreases performance (leases in memfile)
Reservation optimization: None, default settings
MT settings: 4 threads, queue size 4 per thread

Scenario: 27. Test key words: reservations default subnet 100 mysql v4

How 10000 (100% of all clients) subnet reservations kept in mysql decreases performance (leases in memfile)
Reservation optimization: None, default settings
MT settings: 14 threads, queue size 160 per thread

Scenario: 28. Test key words: reservations default subnet 100 mysql v6

How 10000 (100% of all clients) subnet reservations kept in mysql decreases performance (leases in memfile)
Reservation optimization: None, default settings
MT settings: 14 threads, queue size 160 per thread

Scenario: 29. Test key words: reservations default subnet 100 postgresql v4

How 10000 (100% of all clients) subnet reservations kept in postgresql decreases performance (leases in memfile)
Reservation optimization: None, default settings
MT settings: 8 threads, queue size 20 per thread

Scenario: 30. Test key words: reservations default subnet 100 postgresql v6

How 10000 (100% of all clients) subnet reservations kept in postgresql decreases performance (leases in memfile)
Reservation optimization: None, default settings
MT settings: 6 threads, queue size 20 per thread

Scenario: 31. Test key words: reservations default subnet 150 memfile v4

How 15000 (150% of all clients) subnet reservations kept in memfile decreases performance (leases in memfile)
Reservation optimization: None, default settings
MT settings: 8 threads, queue size 4 per thread

Scenario: 32. Test key words: reservations default subnet 150 memfile v6

How 15000 (150% of all clients) subnet reservations kept in memfile decreases performance (leases in memfile)
Reservation optimization: None, default settings
MT settings: 4 threads, queue size 4 per thread

Scenario: 33. Test key words: reservations default subnet 150 mysql v4

How 15000 (150% of all clients) subnet reservations kept in mysql decreases performance (leases in memfile)
Reservation optimization: None, default settings
MT settings: 14 threads, queue size 160 per thread

Scenario: 34. Test key words: reservations default subnet 150 mysql v6

How 15000 (150% of all clients) subnet reservations kept in mysql decreases performance (leases in memfile)
Reservation optimization: None, default settings
MT settings: 14 threads, queue size 160 per thread

Scenario: 35. Test key words: reservations default subnet 150 postgresql v4

How 15000 (150% of all clients) subnet reservations kept in postgresql decreases performance (leases in memfile)
Reservation optimization: None, default settings
MT settings: 8 threads, queue size 20 per thread

Scenario: 36. Test key words: reservations default subnet 150 postgresql v6

How 15000 (150% of all clients) subnet reservations kept in postgresql decreases performance (leases in memfile)
Reservation optimization: None, default settings
MT settings: 6 threads, queue size 20 per thread

Scenario: 37. Test key words: reservations optimized global 30 memfile v4

How 3000 (30% of all clients) global reservations kept in memfile decreases performance (leases in memfile)
Reservation optimization: host-reservation-identifiers set to hw-address, and reservation out-of-pool
MT settings: 8 threads, queue size 4 per thread

Scenario: 38. Test key words: reservations optimized global 30 memfile v6

How 3000 (30% of all clients) global reservations kept in memfile decreases performance (leases in memfile)
Reservation optimization: host-reservation-identifiers set to hw-address, and reservation out-of-pool
MT settings: 4 threads, queue size 4 per thread

Scenario: 39. Test key words: reservations optimized global 30 mysql v4

How 3000 (30% of all clients) global reservations kept in mysql decreases performance (leases in memfile)
Reservation optimization: host-reservation-identifiers set to hw-address, and reservation out-of-pool
MT settings: 14 threads, queue size 160 per thread

Scenario: 40. Test key words: reservations optimized global 30 mysql v6

How 3000 (30% of all clients) global reservations kept in mysql decreases performance (leases in memfile)
Reservation optimization: host-reservation-identifiers set to hw-address, and reservation out-of-pool
MT settings: 14 threads, queue size 160 per thread

Scenario: 41. Test key words: reservations optimized global 30 postgresql v4

How 3000 (30% of all clients) global reservations kept in postgresql decreases performance (leases in memfile)
Reservation optimization: host-reservation-identifiers set to hw-address, and reservation out-of-pool
MT settings: 8 threads, queue size 20 per thread

Scenario: 42. Test key words: reservations optimized global 30 postgresql v6

How 3000 (30% of all clients) global reservations kept in postgresql decreases performance (leases in memfile)
Reservation optimization: host-reservation-identifiers set to hw-address, and reservation out-of-pool
MT settings: 6 threads, queue size 20 per thread

Scenario: 43. Test key words: reservations optimized global 100 memfile v4

How 10000 (100% of all clients) global reservations kept in memfile decreases performance (leases in memfile)
Reservation optimization: host-reservation-identifiers set to hw-address, and reservation out-of-pool
MT settings: 8 threads, queue size 4 per thread

Scenario: 44. Test key words: reservations optimized global 100 memfile v6

How 10000 (100% of all clients) global reservations kept in memfile decreases performance (leases in memfile)
Reservation optimization: host-reservation-identifiers set to hw-address, and reservation out-of-pool
MT settings: 4 threads, queue size 4 per thread

Scenario: 45. Test key words: reservations optimized global 100 mysql v4

How 10000 (100% of all clients) global reservations kept in mysql decreases performance (leases in memfile)
Reservation optimization: host-reservation-identifiers set to hw-address, and reservation out-of-pool
MT settings: 14 threads, queue size 160 per thread

Scenario: 46. Test key words: reservations optimized global 100 mysql v6

How 10000 (100% of all clients) global reservations kept in mysql decreases performance (leases in memfile)
Reservation optimization: host-reservation-identifiers set to hw-address, and reservation out-of-pool
MT settings: 14 threads, queue size 160 per thread

Scenario: 47. Test key words: reservations optimized global 100 postgresql v4

How 10000 (100% of all clients) global reservations kept in postgresql decreases performance (leases in memfile)
Reservation optimization: host-reservation-identifiers set to hw-address, and reservation out-of-pool
MT settings: 8 threads, queue size 20 per thread

Scenario: 48. Test key words: reservations optimized global 100 postgresql v6

How 10000 (100% of all clients) global reservations kept in postgresql decreases performance (leases in memfile)
Reservation optimization: host-reservation-identifiers set to hw-address, and reservation out-of-pool
MT settings: 6 threads, queue size 20 per thread

Scenario: 49. Test key words: reservations optimized global 150 memfile v4

How 15000 (150% of all clients) global reservations kept in memfile decreases performance (leases in memfile)
Reservation optimization: host-reservation-identifiers set to hw-address, and reservation out-of-pool
MT settings: 8 threads, queue size 4 per thread

Scenario: 50. Test key words: reservations optimized global 150 memfile v6

How 15000 (150% of all clients) global reservations kept in memfile decreases performance (leases in memfile)
Reservation optimization: host-reservation-identifiers set to hw-address, and reservation out-of-pool
MT settings: 4 threads, queue size 4 per thread

Scenario: 51. Test key words: reservations optimized global 150 mysql v4

How 15000 (150% of all clients) global reservations kept in mysql decreases performance (leases in memfile)
Reservation optimization: host-reservation-identifiers set to hw-address, and reservation out-of-pool
MT settings: 14 threads, queue size 160 per thread

Scenario: 52. Test key words: reservations optimized global 150 mysql v6

How 15000 (150% of all clients) global reservations kept in mysql decreases performance (leases in memfile)
Reservation optimization: host-reservation-identifiers set to hw-address, and reservation out-of-pool
MT settings: 14 threads, queue size 160 per thread

Scenario: 53. Test key words: reservations optimized global 150 postgresql v4

How 15000 (150% of all clients) global reservations kept in postgresql decreases performance (leases in memfile)
Reservation optimization: host-reservation-identifiers set to hw-address, and reservation out-of-pool
MT settings: 8 threads, queue size 20 per thread

Scenario: 54. Test key words: reservations optimized global 150 postgresql v6

How 15000 (150% of all clients) global reservations kept in postgresql decreases performance (leases in memfile)
Reservation optimization: host-reservation-identifiers set to hw-address, and reservation out-of-pool
MT settings: 6 threads, queue size 20 per thread

Scenario: 55. Test key words: reservations optimized subnet 30 memfile v4

How 3000 (30% of all clients) subnet reservations kept in memfile decreases performance (leases in memfile)
Reservation optimization: host-reservation-identifiers set to hw-address, and reservation out-of-pool
MT settings: 8 threads, queue size 4 per thread

Scenario: 56. Test key words: reservations optimized subnet 30 memfile v6

How 3000 (30% of all clients) subnet reservations kept in memfile decreases performance (leases in memfile)
Reservation optimization: host-reservation-identifiers set to hw-address, and reservation out-of-pool
MT settings: 4 threads, queue size 4 per thread

Scenario: 57. Test key words: reservations optimized subnet 30 mysql v4

How 3000 (30% of all clients) subnet reservations kept in mysql decreases performance (leases in memfile)
Reservation optimization: host-reservation-identifiers set to hw-address, and reservation out-of-pool
MT settings: 14 threads, queue size 160 per thread

Scenario: 58. Test key words: reservations optimized subnet 30 mysql v6

How 3000 (30% of all clients) subnet reservations kept in mysql decreases performance (leases in memfile)
Reservation optimization: host-reservation-identifiers set to hw-address, and reservation out-of-pool
MT settings: 14 threads, queue size 160 per thread

Scenario: 59. Test key words: reservations optimized subnet 30 postgresql v4

How 3000 (30% of all clients) subnet reservations kept in postgresql decreases performance (leases in memfile)
Reservation optimization: host-reservation-identifiers set to hw-address, and reservation out-of-pool
MT settings: 8 threads, queue size 20 per thread

Scenario: 60. Test key words: reservations optimized subnet 30 postgresql v6

How 3000 (30% of all clients) subnet reservations kept in postgresql decreases performance (leases in memfile)
Reservation optimization: host-reservation-identifiers set to hw-address, and reservation out-of-pool
MT settings: 6 threads, queue size 20 per thread

Scenario: 61. Test key words: reservations optimized subnet 100 memfile v4

How 10000 (100% of all clients) subnet reservations kept in memfile decreases performance (leases in memfile)
Reservation optimization: host-reservation-identifiers set to hw-address, and reservation out-of-pool
MT settings: 8 threads, queue size 4 per thread

Scenario: 62. Test key words: reservations optimized subnet 100 memfile v6

How 10000 (100% of all clients) subnet reservations kept in memfile decreases performance (leases in memfile)
Reservation optimization: host-reservation-identifiers set to hw-address, and reservation out-of-pool
MT settings: 4 threads, queue size 4 per thread

Scenario: 63. Test key words: reservations optimized subnet 100 mysql v4

How 10000 (100% of all clients) subnet reservations kept in mysql decreases performance (leases in memfile)
Reservation optimization: host-reservation-identifiers set to hw-address, and reservation out-of-pool
MT settings: 14 threads, queue size 160 per thread

Scenario: 64. Test key words: reservations optimized subnet 100 mysql v6

How 10000 (100% of all clients) subnet reservations kept in mysql decreases performance (leases in memfile)
Reservation optimization: host-reservation-identifiers set to hw-address, and reservation out-of-pool
MT settings: 14 threads, queue size 160 per thread

Scenario: 65. Test key words: reservations optimized subnet 100 postgresql v4

How 10000 (100% of all clients) subnet reservations kept in postgresql decreases performance (leases in memfile)
Reservation optimization: host-reservation-identifiers set to hw-address, and reservation out-of-pool
MT settings: 8 threads, queue size 20 per thread

Scenario: 66. Test key words: reservations optimized subnet 100 postgresql v6

How 10000 (100% of all clients) subnet reservations kept in postgresql decreases performance (leases in memfile)
Reservation optimization: host-reservation-identifiers set to hw-address, and reservation out-of-pool
MT settings: 6 threads, queue size 20 per thread

Scenario: 67. Test key words: reservations optimized subnet 150 memfile v4

How 15000 (150% of all clients) subnet reservations kept in memfile decreases performance (leases in memfile)
Reservation optimization: host-reservation-identifiers set to hw-address, and reservation out-of-pool
MT settings: 8 threads, queue size 4 per thread

Scenario: 68. Test key words: reservations optimized subnet 150 memfile v6

How 15000 (150% of all clients) subnet reservations kept in memfile decreases performance (leases in memfile)
Reservation optimization: host-reservation-identifiers set to hw-address, and reservation out-of-pool
MT settings: 4 threads, queue size 4 per thread

Scenario: 69. Test key words: reservations optimized subnet 150 mysql v4

How 15000 (150% of all clients) subnet reservations kept in mysql decreases performance (leases in memfile)
Reservation optimization: host-reservation-identifiers set to hw-address, and reservation out-of-pool
MT settings: 14 threads, queue size 160 per thread

Scenario: 70. Test key words: reservations optimized subnet 150 mysql v6

How 15000 (150% of all clients) subnet reservations kept in mysql decreases performance (leases in memfile)
Reservation optimization: host-reservation-identifiers set to hw-address, and reservation out-of-pool
MT settings: 14 threads, queue size 160 per thread

Scenario: 71. Test key words: reservations optimized subnet 150 postgresql v4

How 15000 (150% of all clients) subnet reservations kept in postgresql decreases performance (leases in memfile)
Reservation optimization: host-reservation-identifiers set to hw-address, and reservation out-of-pool
MT settings: 8 threads, queue size 20 per thread

Scenario: 72. Test key words: reservations optimized subnet 150 postgresql v6

How 15000 (150% of all clients) subnet reservations kept in postgresql decreases performance (leases in memfile)
Reservation optimization: host-reservation-identifiers set to hw-address, and reservation out-of-pool
MT settings: 6 threads, queue size 20 per thread

Resource consumption of single Kea running with different backends

Each test has an additional description.

Scenario: 1. Test key words: resources consumption memfile v6 multi

Checking if memory usage still grows when there are no new clients, just returning. 10mln unique clients. 4 threads, queue size 4 per thread Memory usage should stop growing after ~333.Traffic is generated for 1000 but Kea is monitored for additional 100 seconds.

Scenario: 2. Test key words: resources consumption memfile v4 multi

Checking if memory usage still grows when there are no new clients, just returning. 10mln unique clients. 8 threads, queue size 4 per thread Memory usage should stop growing after ~466.Traffic is generated for 1400 but Kea is monitored for additional 100 seconds.

Scenario: 3. Test key words: resources consumption mysql v6 multi

Checking if memory usage still grows when there are no new clients, just returning. 10mln unique clients. 14 threads, queue size 160 per thread Memory usage should stop growing after ~333.Traffic is generated for 1000 but Kea is monitored for additional 100 seconds.

Scenario: 4. Test key words: resources consumption mysql v4 multi

Checking if memory usage still grows when there are no new clients, just returning. 10mln unique clients. 14 threads, queue size 160 per thread Memory usage should stop growing after ~466.Traffic is generated for 1400 but Kea is monitored for additional 100 seconds.

Scenario: 5. Test key words: resources consumption postgresql v6 multi

Checking if memory usage still grows when there are no new clients, just returning. 10mln unique clients. 6 threads, queue size 20 per thread Memory usage should stop growing after ~333.Traffic is generated for 1000 but Kea is monitored for additional 100 seconds.

Scenario: 6. Test key words: resources consumption postgresql v4 multi

Checking if memory usage still grows when there are no new clients, just returning. 10mln unique clients. 8 threads, queue size 20 per thread Memory usage should stop growing after ~466.Traffic is generated for 1400 but Kea is monitored for additional 100 seconds.

How different subnet/network configuration affect Kea performance

Each test has an additional description

Scenario: 1. Test key words: multiple subnets memfile 100 v6

How adding 100 subnets decreases performance.
4 threads, queue size 4 per thread

Scenario: 2. Test key words: multiple subnets memfile 100 v4

How adding 100 subnets decreases performance.
8 threads, queue size 4 per thread

Scenario: 3. Test key words: multiple subnets memfile 200 v6

How adding 200 subnets decreases performance.
4 threads, queue size 4 per thread

Scenario: 4. Test key words: multiple subnets memfile 200 v4

How adding 200 subnets decreases performance.
8 threads, queue size 4 per thread

Scenario: 5. Test key words: multiple subnets mysql 100 v6

How adding 100 subnets decreases performance.
14 threads, queue size 160 per thread

Scenario: 6. Test key words: multiple subnets mysql 100 v4

How adding 100 subnets decreases performance.
14 threads, queue size 160 per thread

Scenario: 7. Test key words: multiple subnets mysql 200 v6

How adding 200 subnets decreases performance.
14 threads, queue size 160 per thread

Scenario: 8. Test key words: multiple subnets mysql 200 v4

How adding 200 subnets decreases performance.
14 threads, queue size 160 per thread

Scenario: 9. Test key words: multiple subnets postgresql 100 v6

How adding 100 subnets decreases performance.
6 threads, queue size 20 per thread

Scenario: 10. Test key words: multiple subnets postgresql 100 v4

How adding 100 subnets decreases performance.
8 threads, queue size 20 per thread

Scenario: 11. Test key words: multiple subnets postgresql 200 v6

How adding 200 subnets decreases performance.
6 threads, queue size 20 per thread

Scenario: 12. Test key words: multiple subnets postgresql 200 v4

How adding 200 subnets decreases performance.
8 threads, queue size 20 per thread

Stability of HA setups

Each test has an additional description

Scenario: 1. Test key words: resources consumption ha multi 256 hotstandby memfile v4

Lets run v4 HA hotstandby setup with memfile lease backend.
Multi threading settings:
8 threads, queue size 4 per thread
"parked-packet-limit" is set to 256Traffic is generated 800 and systems are monitored for another 100 seconds

Scenario: 2. Test key words: resources consumption ha multi 256 hotstandby memfile v6

Lets run v6 HA hotstandby setup with memfile lease backend.
Multi threading settings:
4 threads, queue size 4 per thread
"parked-packet-limit" is set to 256Traffic is generated 800 and systems are monitored for another 100 seconds

Scenario: 3. Test key words: resources consumption ha multi 256 hotstandby mysql v4

Lets run v4 HA hotstandby setup with mysql lease backend.
Multi threading settings:
14 threads, queue size 160 per thread
"parked-packet-limit" is set to 256Traffic is generated 800 and systems are monitored for another 100 seconds

Scenario: 4. Test key words: resources consumption ha multi 256 hotstandby mysql v6

Lets run v6 HA hotstandby setup with mysql lease backend.
Multi threading settings:
14 threads, queue size 160 per thread
"parked-packet-limit" is set to 256Traffic is generated 800 and systems are monitored for another 100 seconds

Scenario: 5. Test key words: resources consumption ha multi 256 hotstandby postgresql v4

Lets run v4 HA hotstandby setup with postgresql lease backend.
Multi threading settings:
8 threads, queue size 20 per thread
"parked-packet-limit" is set to 256Traffic is generated 800 and systems are monitored for another 100 seconds

Scenario: 6. Test key words: resources consumption ha multi 256 hotstandby postgresql v6

Lets run v6 HA hotstandby setup with postgresql lease backend.
Multi threading settings:
6 threads, queue size 20 per thread
"parked-packet-limit" is set to 256Traffic is generated 800 and systems are monitored for another 100 seconds

Scenario: 7. Test key words: resources consumption ha multi 256 loadbalancing memfile v4

Lets run v4 HA loadbalancing setup with memfile lease backend.
Multi threading settings:
8 threads, queue size 4 per thread
"parked-packet-limit" is set to 256Traffic is generated 800 and systems are monitored for another 100 seconds

Scenario: 8. Test key words: resources consumption ha multi 256 loadbalancing memfile v6

Lets run v6 HA loadbalancing setup with memfile lease backend.
Multi threading settings:
4 threads, queue size 4 per thread
"parked-packet-limit" is set to 256Traffic is generated 800 and systems are monitored for another 100 seconds

Scenario: 9. Test key words: resources consumption ha multi 256 loadbalancing mysql v4

Lets run v4 HA loadbalancing setup with mysql lease backend.
Multi threading settings:
14 threads, queue size 160 per thread
"parked-packet-limit" is set to 256Traffic is generated 800 and systems are monitored for another 100 seconds

Scenario: 10. Test key words: resources consumption ha multi 256 loadbalancing mysql v6

Lets run v6 HA loadbalancing setup with mysql lease backend.
Multi threading settings:
14 threads, queue size 160 per thread
"parked-packet-limit" is set to 256Traffic is generated 800 and systems are monitored for another 100 seconds

Scenario: 11. Test key words: resources consumption ha multi 256 loadbalancing postgresql v4

Lets run v4 HA loadbalancing setup with postgresql lease backend.
Multi threading settings:
8 threads, queue size 20 per thread
"parked-packet-limit" is set to 256Traffic is generated 800 and systems are monitored for another 100 seconds

Scenario: 12. Test key words: resources consumption ha multi 256 loadbalancing postgresql v6

Lets run v6 HA loadbalancing setup with postgresql lease backend.
Multi threading settings:
6 threads, queue size 20 per thread
"parked-packet-limit" is set to 256Traffic is generated 800 and systems are monitored for another 100 seconds

Scenario: 13. Test key words: resources consumption ha multi 256 passive backup memfile v4

Lets run v4 HA passive-backup setup with memfile lease backend.
Multi threading settings:
8 threads, queue size 4 per thread
"parked-packet-limit" is set to 256Traffic is generated 800 and systems are monitored for another 100 seconds

Scenario: 14. Test key words: resources consumption ha multi 256 passive backup memfile v6

Lets run v6 HA passive-backup setup with memfile lease backend.
Multi threading settings:
4 threads, queue size 4 per thread
"parked-packet-limit" is set to 256Traffic is generated 800 and systems are monitored for another 100 seconds

Scenario: 15. Test key words: resources consumption ha multi 256 passive backup mysql v4

Lets run v4 HA passive-backup setup with mysql lease backend.
Multi threading settings:
14 threads, queue size 160 per thread
"parked-packet-limit" is set to 256Traffic is generated 800 and systems are monitored for another 100 seconds

Scenario: 16. Test key words: resources consumption ha multi 256 passive backup mysql v6

Lets run v6 HA passive-backup setup with mysql lease backend.
Multi threading settings:
14 threads, queue size 160 per thread
"parked-packet-limit" is set to 256Traffic is generated 800 and systems are monitored for another 100 seconds

Scenario: 17. Test key words: resources consumption ha multi 256 passive backup postgresql v4

Lets run v4 HA passive-backup setup with postgresql lease backend.
Multi threading settings:
8 threads, queue size 20 per thread
"parked-packet-limit" is set to 256Traffic is generated 800 and systems are monitored for another 100 seconds

Scenario: 18. Test key words: resources consumption ha multi 256 passive backup postgresql v6

Lets run v6 HA passive-backup setup with postgresql lease backend.
Multi threading settings:
6 threads, queue size 20 per thread
"parked-packet-limit" is set to 256Traffic is generated 800 and systems are monitored for another 100 seconds

Scenario: 19. Test key words: resources consumption ha backup wait for response multi passive backup mysql v6

Lets run v6 HA passive-backup setup with mysql lease backend.
Multi threading settings:
14 threads, queue size 160 per thread
"parked-packet-limit" is set to 256
"wait-backup-ack" value is set to True!

Scenario: 20. Test key words: resources consumption ha backup wait for response multi passive backup mysql v4

Lets run v4 HA passive-backup setup with mysql lease backend.
Multi threading settings:
14 threads, queue size 160 per thread
"parked-packet-limit" is set to 256
"wait-backup-ack" value is set to True!

Scenario: 21. Test key words: resources consumption ha backup wait for response multi passive backup memfile v6

Lets run v6 HA passive-backup setup with memfile lease backend.
Multi threading settings:
4 threads, queue size 4 per thread
"parked-packet-limit" is set to 256
"wait-backup-ack" value is set to True!

Scenario: 22. Test key words: resources consumption ha backup wait for response multi passive backup memfile v4

Lets run v4 HA passive-backup setup with memfile lease backend.
Multi threading settings:
8 threads, queue size 4 per thread
"parked-packet-limit" is set to 256
"wait-backup-ack" value is set to True!

Scenario: 23. Test key words: resources consumption ha backup wait for response multi passive backup postgresql v6

Lets run v6 HA passive-backup setup with postgresql lease backend.
Multi threading settings:
6 threads, queue size 20 per thread
"parked-packet-limit" is set to 256
"wait-backup-ack" value is set to True!

Scenario: 24. Test key words: resources consumption ha backup wait for response multi passive backup postgresql v4

Lets run v4 HA passive-backup setup with postgresql lease backend.
Multi threading settings:
8 threads, queue size 20 per thread
"parked-packet-limit" is set to 256
"wait-backup-ack" value is set to True!

Real life scenarios - This set of tests measure how using various features of Kea affect performance.

In Scenarios 1 & 2 we show the impact of using Forensic Logging (using the default logging template). Scenarios 3 & 4 show the benefit of using the Lease Caching feature, both for DHCPv4 and DHCPv6. Please view the mouseover text for additional test scenario information.

Scenario: 1. Test key words: legal logging hook feature basic memfile v4 multi

How enabling default legal logging will impact performance
8 threads, queue size 4 per thread

Scenario: 2. Test key words: legal logging hook feature basic memfile v6 multi

How enabling default legal logging will impact performance
4 threads, queue size 4 per thread

Scenario: 3. Test key words: classification feature 100 empty memfile v6 multi

Check how different classification configurations impact performance.

Scenario: 4. Test key words: classification feature 100 empty memfile v4 multi

Check how different classification configurations impact performance.

Scenario: 5. Test key words: classification feature 100 empty mysql v6 multi

Check how different classification configurations impact performance.

Scenario: 6. Test key words: classification feature 100 empty mysql v4 multi

Check how different classification configurations impact performance.

Scenario: 7. Test key words: classification feature 100 empty postgresql v6 multi

Check how different classification configurations impact performance.

Scenario: 8. Test key words: classification feature 100 empty postgresql v4 multi

Check how different classification configurations impact performance.

Scenario: 9. Test key words: classification feature 100 a==a memfile v6 multi

Check how different classification configurations impact performance.

Scenario: 10. Test key words: classification feature 100 a==a memfile v4 multi

Check how different classification configurations impact performance.

Scenario: 11. Test key words: classification feature 100 a==a mysql v6 multi

Check how different classification configurations impact performance.

Scenario: 12. Test key words: classification feature 100 a==a mysql v4 multi

Check how different classification configurations impact performance.

Scenario: 13. Test key words: classification feature 100 a==a postgresql v6 multi

Check how different classification configurations impact performance.

Scenario: 14. Test key words: classification feature 100 a==a postgresql v4 multi

Check how different classification configurations impact performance.

Scenario: 15. Test key words: classification feature 200 empty memfile v6 multi

Check how different classification configurations impact performance.

Scenario: 16. Test key words: classification feature 200 empty memfile v4 multi

Check how different classification configurations impact performance.

Scenario: 17. Test key words: classification feature 200 empty mysql v6 multi

Check how different classification configurations impact performance.

Scenario: 18. Test key words: classification feature 200 empty mysql v4 multi

Check how different classification configurations impact performance.

Scenario: 19. Test key words: classification feature 200 empty postgresql v6 multi

Check how different classification configurations impact performance.

Scenario: 20. Test key words: classification feature 200 empty postgresql v4 multi

Check how different classification configurations impact performance.

Scenario: 21. Test key words: classification feature 200 a==a memfile v6 multi

Check how different classification configurations impact performance.

Scenario: 22. Test key words: classification feature 200 a==a memfile v4 multi

Check how different classification configurations impact performance.

Scenario: 23. Test key words: classification feature 200 a==a mysql v6 multi

Check how different classification configurations impact performance.

Scenario: 24. Test key words: classification feature 200 a==a mysql v4 multi

Check how different classification configurations impact performance.

Scenario: 25. Test key words: classification feature 200 a==a postgresql v6 multi

Check how different classification configurations impact performance.

Scenario: 26. Test key words: classification feature 200 a==a postgresql v4 multi

Check how different classification configurations impact performance.

Scenario: 27. Test key words: lease reclamation feature 10000 memfile v6 multi

Each backend has different lease reclamation process, this test checks performance impact of lease reclamation process using default configuration and 'memfile's lease database backend.In this case no leases will be expired during the test (validlifetime is 10000 which is more than test duration)

Scenario: 28. Test key words: lease reclamation feature 10000 memfile v4 multi

Each backend has different lease reclamation process, this test checks performance impact of lease reclamation process using default configuration and 'memfile's lease database backend.In this case no leases will be expired during the test (validlifetime is 10000 which is more than test duration)

Scenario: 29. Test key words: lease reclamation feature 10000 mysql v6 multi

Each backend has different lease reclamation process, this test checks performance impact of lease reclamation process using default configuration and 'mysql's lease database backend.In this case no leases will be expired during the test (validlifetime is 10000 which is more than test duration)

Scenario: 30. Test key words: lease reclamation feature 10000 mysql v4 multi

Each backend has different lease reclamation process, this test checks performance impact of lease reclamation process using default configuration and 'mysql's lease database backend.In this case no leases will be expired during the test (validlifetime is 10000 which is more than test duration)

Scenario: 31. Test key words: lease reclamation feature 10000 postgresql v6 multi

Each backend has different lease reclamation process, this test checks performance impact of lease reclamation process using default configuration and 'postgresql's lease database backend.In this case no leases will be expired during the test (validlifetime is 10000 which is more than test duration)

Scenario: 32. Test key words: lease reclamation feature 10000 postgresql v4 multi

Each backend has different lease reclamation process, this test checks performance impact of lease reclamation process using default configuration and 'postgresql's lease database backend.In this case no leases will be expired during the test (validlifetime is 10000 which is more than test duration)

Scenario: 33. Test key words: lease reclamation feature 10 memfile v6 multi

Each backend has different lease reclamation process, this test checks performance impact of lease reclamation process using default configuration and 'memfile's lease database backend.In this case there will be always more expired leases than single expiration query can handle (validlifetime is 10)

Scenario: 34. Test key words: lease reclamation feature 10 memfile v4 multi

Each backend has different lease reclamation process, this test checks performance impact of lease reclamation process using default configuration and 'memfile's lease database backend.In this case there will be always more expired leases than single expiration query can handle (validlifetime is 10)

Scenario: 35. Test key words: lease reclamation feature 10 mysql v6 multi

Each backend has different lease reclamation process, this test checks performance impact of lease reclamation process using default configuration and 'mysql's lease database backend.In this case there will be always more expired leases than single expiration query can handle (validlifetime is 10)

Scenario: 36. Test key words: lease reclamation feature 10 mysql v4 multi

Each backend has different lease reclamation process, this test checks performance impact of lease reclamation process using default configuration and 'mysql's lease database backend.In this case there will be always more expired leases than single expiration query can handle (validlifetime is 10)

Scenario: 37. Test key words: lease reclamation feature 10 postgresql v6 multi

Each backend has different lease reclamation process, this test checks performance impact of lease reclamation process using default configuration and 'postgresql's lease database backend.In this case there will be always more expired leases than single expiration query can handle (validlifetime is 10)

Scenario: 38. Test key words: lease reclamation feature 10 postgresql v4 multi

Each backend has different lease reclamation process, this test checks performance impact of lease reclamation process using default configuration and 'postgresql's lease database backend.In this case there will be always more expired leases than single expiration query can handle (validlifetime is 10)

This section provides insight into how Kea performs second by second during longer running tests. This information is primarily useful for the development team and does not provide guidance on any user-settings.
Thanks to this data visualization we can:

• Detect regularly occurring problems
• Find out how processes like LFC impact Kea performance
• See how different database backends work

Scenario: 1. Test key words: long running memfile v4 multi

Check how Kea response rates changes over longer time of maximum load.
Multi threading settings:
8 threads, queue size 4 per thread

Scenario: 2. Test key words: long running memfile v6 multi

Check how Kea response rates changes over longer time of maximum load.
Multi threading settings:
4 threads, queue size 4 per thread

Scenario: 3. Test key words: long running mysql v4 multi

Check how Kea response rates changes over longer time of maximum load.
Multi threading settings:
14 threads, queue size 160 per thread

Scenario: 4. Test key words: long running mysql v6 multi

Check how Kea response rates changes over longer time of maximum load.
Multi threading settings:
14 threads, queue size 160 per thread

Scenario: 5. Test key words: long running postgresql v4 multi

Check how Kea response rates changes over longer time of maximum load.
Multi threading settings:
8 threads, queue size 20 per thread

Scenario: 6. Test key words: long running postgresql v6 multi

Check how Kea response rates changes over longer time of maximum load.
Multi threading settings:
6 threads, queue size 20 per thread

Performance testing report

Welcome to the Kea performance testing report, this document is generated automatically after each test run.
In the section "Testing setup" we describe the common factors that apply to all tests. Specific test details and explanations can be found with the results.

Kea performance test types

All tests in this report can be divided into two groups. The first type (A) is where we are calculating the top performance of this Kea version, and the second type (B) is where we are testing Kea with different features, settings, or longer periods of time to observe the impact of various configurations and features, as compared with the maximum throughput measured in (A).

How maximum performance is calculated

Calculating the maximum performance is a multistage process, with a couple of assumptions:

• We use the simplest possible configuration, mostly default settings.
• We gradually increase the generated traffic rate until the packet drop rate reaches 2,5% (meaning 2,5% of all responses sent by the server do not reach the client within 1 second).
• The 2,5% drop rate was chosen because it represents a busy, but not overloaded, server. A 0% drop rate does not show the maximum throughput and a higher drop rate is undesirable in production.

Report Details

• In the "Single Kea and HA setup" tab we present testing results for a single Kea server and for two Kea servers running in High Availability configuration, both with and without multi-threading enabled. These are type A tests, determining the maximum performance of this version of Kea.
• In the "Multi-threading settings" tab we determine how different multi-threading settings ("thread-pool-size" and "packet-queue-size") affect performance for all three lease backends. These are type A tests, used to determine the optimum settings to achieve the maximum performance of this version of Kea. These optimal settings are used in subsequent tests.
• In the "Host Reservations" tab we present the performance impact of using host-reservations with different settings (e.g. looking at the difference between out of pool and in pool reservation, or between subnet and global reservations). These are type B tests that use the thread settings from the type A tests and compare the results.
• Most Kea deployments will use multiple subnets or shared networks. The "Subnets" tab compares Kea performance configured with hundreds of subnets and networks to the maximum baseline performance. These are type B tests using previously calculated traffic rates from A. We measure the impact of adding subnets on the memfile, MySQL, and PostgreSQL lease file scenarios. Tests 1, 3, and 5 use 100 subnets, tests 2, 4, and 6 use 200 subnets, so you can see the incremental performance difference.
• The "Resource Consumption" and "Stability" reports present traffic details, CPU, and MEM utilization over a longer time for all lease backends. Those are tests type (B) that are using previously calculated traffic rates. These tests are primarily useful for the development team.
• The "Various Features" tab contains testing results for forensic logging and lease caching. These are type B tests that are using previously calculated traffic rates.
• The "HA stability" tab contains results of longer running tests for all High Availability setup (hot-standby, load-balancing, passive-backup).
• The "History of basic results" tab tracks the results of the "Single Kea and HA setup" baseline over time.

Testing setup

Network

Testing is performed in ISC's internal network and uses 3 systems. Two are running Kea and database backends (specifications below) and one is running perfdhcp. All three are connected to one VLAN using 1 gigabit ethernet network.

Hardware specs - R340 server

• CPU Intel Xeon E-2146G 3.5GHz 6 cores/12 threads
• 64GB RAM
• 3 x SSDs 446GB each in HW RAID-0 configuration (virtual disk size 1338GB)
• Intel(R) 10GbE 2P X710 Adapter (2 ports)
• Intel(R) GbE 4P I350-t Adapter (4 ports)
• Broadcom Gigabit Ethernet BCM5720 (2 ports)

OS details, software versions

Tests were executed using:

• OS - Ubuntu 18.04.6 LTS
• Mysql - 5.7.37-0ubuntu0.18.04.1
• Postgresql - 190ubuntu0.1
• boost - 1.65.1.0ubuntu1
• openssl - 1.1.1-1ubuntu2.1~18.04.15
• Kea - 2.0.2

Test Design

• Maximum performance is measured when packet loss reaches 2.5%. This is assumed to be the maximum acceptable rate of packet loss.
• If Kea is using a database backend, the database is always located on the same system as Kea.
• A packet is considered to have been dropped when the response doesn't reach perfdhcp within 2 seconds of the initial packet.

Kea configuration

Configurations vary between tests and test types, details are described with the test results.
If not stated explicitly in the test description Kea has default configuration values.

Database configuration

• Databases are running on the same systems as Kea (connection delays between Kea server and database is negligible)
• MySQL is configured with innodb_flush_log_at_trx_commit=2 as described in the KEA ARM. Nothing else has been changed.
• The postgresql configuration uses default values.

Default clients behaviour

Unless stated differently in the test, only the basic 4 message exchange (SARR and DORA) is generated. No release/renew or rebind packets are generated.
Each client performs an exchange just once. Perfdhcp will simulate up to 500 million unique clientIDs so Kea will not recognize any client as returning.
Messages do not include any additional options, except those necessary to get an address from the DHCP server.

Traffic generator

We use the traffic generator "perfdhcp" for all tests. "perfdhcp" is developed by ISC and is available in Kea source/packages.
We encourage the user to refer to the KEA ARM for more details.

Disclaimers

Performance testing results are volatile, multiple factors have to be taken into account e.g.: hardware, OS type, network, database location (local, remote), compilation CXX flags etc.
The results shown in this report are what we were able to observe within our testing network - they are not necessarily representative of what will be observed in another network.
The Kea development team takes performance and stability very seriously - please report any irregularities you observe on your network to the kea-users mailing list.

ISC strongly recommends making yourself familiar with the Kea performance optimization article.