Skip to content

Latest commit

 

History

History
243 lines (172 loc) · 4.86 KB

performance.md

File metadata and controls

243 lines (172 loc) · 4.86 KB
Raw

Performance notes

All tests until Feb 17, 2017 ran on the old computer.

New computer and settings

  • CPU: i5 4.0 GHz (4 cores)
  • Storage: SSD
  • Logging level: WARN
  • Settings: 4 threads, no message logging

Old computer and settings

  • CPU: i3 2.40 GHz (2 cores)
  • Storage: SSD
  • Logging level: INFO
  • Settings: 4 threads, no message logging

Testing server performance

Tests for AV server

AMQP server run command: ./gradlew runAmqpServer

AMQP load test run command: ./gradlew runAmqpLoadTest

JMS server run command: ./gradlew runJmsServer

JMS load test run command: ./gradlew runJmsLoadTest

Old load test run command: ./gradlew loadTest

Message logging: no

Other scenarios

Results on the old computer

  • 1 million messages with AMQP server. It didn't show any memory leaks.
  • Result: 557 msgs/s
  • 1 million messages with JMS server. It didn't show any memory leaks.
  • Result: 643 msgs/s

Default scenarios

First run is for a warm-up and is not counted.

AMQP

May 14, 2017

With a new BufferedPerformanceTester class.

run 1. 2. 3. 4. 5.
msg/s 1628 1545 1663 1607 1552

May 2, 2017

With the new computer.

run 1. 2. 3. 4. 5.
msg/s 926 907 897 888 913

Socket pooling is default.

Results on the old computer

Feb 17, 2017

With an experimental ClamAV socket pooling.

run 1. 2. 3. 4. 5.
msg/s 563 559 560 556 553

Without the socket pooling.

run 1. 2. 3. 4. 5.
msg/s 528 532 532 517 527

Jan 16, 2017

Version 0.5-rc1.

run 1. 2. 3. 4. 5.
msg/s 524 515 515 492 516

Dec 23, 2016

Spring 5 and small tuning.

run 1. 2. 3. 4. 5.
msg/s 516 512 506 507 514

Dec 14, 2016

After message processor redesign.

run 1. 2. 3. 4. 5.
msg/s 501 484 507 492 513

Nov 17, 2016

run 1. 2. 3. 4. 5.
msg/s 471 470 480 480 481

JMS

May 2, 2017

run 1. 2. 3. 4. 5.
msg/s 949 988 951 1009 978

Results on the old computer

Jan 16, 2017

Version 0.5-rc1 with JMS async sending setting.

run 1. 2. 3. 4. 5.
msg/s 565 567 538 561 552

Dec 14, 2016

After message processor redesign.

run 1. 2. 3. 4. 5.
msg/s 91 92 92 91 91

Nov 17, 2016

run 1. 2. 3. 4. 5.
msg/s 91 91 92 91 90

AMQP results for the old load test

Nov 13, 2016

run messages/second
1. 533
2. 549
3. 561
4. 546
5. 551

AMQP load test results for the old service written in Python (retested February 20, 2016):

Message broker, server and anti-virus program were on the same machine. Sending file is EICAR.

Client is on a different machine.

  • Receiver: AVReceiver
  • Sender: AVSender
Load tests
Load test start for 1000 messages...
Load test end
Duration: 20 s
Messages: 50/s

Load test start for 10 000 messages...
Load test end
Duration: 178 s
Messages: 56/s

Load test start for 100 000 messages...
Load test end
Duration: 1800 s
Messages: 56/s

Solr

As a storage for data was used USB 3 2.5" HDD.

Documents are Message info documents.

Inserting documents

1. million

  • In collection: 10 602 ms
  • In Solr: 59 769 ms
  • After commit: 65 834 ms

201. million

  • In collection: 10 410 ms
  • In Solr: 55 889 ms
  • After commit: 273 645 ms

251. million

  • In collection: 10 375 ms
  • In Solr: 49 748 ms
  • After commit: 346 980 ms

Sorting queries

first hits

created field - ASC and then DESC

x 100 M 200 M 250 M
ASC 2348 ms 5797 ms 6795 ms
DESC 1643 ms 3549 ms 4354 ms

Commits

Inserts one document and commit after. Runs in a loop.

With empty collection.

x 10 100 1000 10 000
no 67 ms 396 ms 3 s 21.4 s
soft 149 ms 879 ms 7.9 s 1 m 6 s
hard 4.2 s 29.8 s 4 m 46 s x

With 100 million documents (25 GB).

x 10 100 1000 10 000
no 61 ms 547 ms 3.7 s 21.9 s
soft x 1.6 s 11.3 s 11 m 10 s
hard 8.5 s 58.4 s 9 m 42 s x

With 200 million documents (46 GB).

x 10 100 1000 10 000
no 76 ms 532 ms 3.7 s 26.3 s
soft 4.9 s 31.6 s 3 m 38 s 13 m 12 s
hard 9.2 s 1 m 4 s 10 m 43 s x

With 250 million documents (58 GB).

x 10 100 1000 10 000
no 48 ms 312 ms 2.4 s 19 s
soft 4.3 s 29.5 s 3 m 35 s 26 m 2 s
hard 7.8 s 1 m 17 s x x