Perfomance tips

Product comparation perfomance tests

"The nature of productivity" is similar to the Great Store Columns of the Colony Family: Cassandra. Testing the production system, we focused on distributed (decentralized) systems.

AcapellaDB_v1.0 is more like a Key-Value with transactions, and not on Cassandra.

Cluster Transport : aeron vs grpc perfomance comparation

AcapellaDB use aeron as network and ipc transport. It is Important - perfomance the one of base target.

First compare perfomance of transport subsystem. for example fast grpc based onclassic tcp and aeron (realible udp)

TRANSPORT PERFOMANCE TEST: compare grpc tcp vs Aeron transport andcustom rpc

Jepsen test

Jepsen is an effort to improve the safety of distributed databases, queues, consensus systems, etc.

Jepsen tests is a de facto standart for testing distributed systems.

We write adapter to our AcapellaDB key-value layer and run jepsen tests in several docker nodes.

Look console gif image in "New Tab"

Lathency

todo from old site

The approximate order of the number of requests per second for one client on operations with 3 replicas (N = 3):

  • non transactional write - 6_035 ops/sec
  • non transactional read - 40_972 ops/sec
  • transactional swap's two keys - 1_180 ops/sec

Measured for examples: https://srv.nppsatek.ru:42917/hiload/AcapellaDBClient-py/blob/develop/src/examples/

Technology tests

our test way

A long article with an analysis of our approach Read blog article (RUS)

Tests on github

For tests the repository on github is made, there we will lay out our tests and examplesgithub icon .