Latency benchmark and other useful tips

This blog post summarizes some of the tips you should follow to optimize your application for best latency.

The following benchmark measures the latency between one remote space or two replicated spaces with varying number of fields in each entry. This benchmark provides indicative results for optimum latency. The benchmark has been conducted over 1G network with 0.208ms round trip ping latency for 1K object.

Latency vs Rate - Sync Replication/No Replication

Rate

Results

• Remote call latency: empty object 0.27ms and 0.29ms for 8 fields object.
• Remote with synchronous replication latency: empty object write latency is 0.63ms and with 8 fields object is 0.69ms

Benchmark Summary

The results shows that the overhead of the space operation on the network latency is close to ~0.1ms or ~0.2 msec with replication. It also shows that the number of fields in the object impact on latency is relatively low compared with the network overhead. Synchronous replication doubles the latency as could be expected.

Other Benchmarks from GigaSpaces

• Spaces Based Architecture (SBA) vs Tier Based Implementation
• Web Application Scalability
• Scale-up on Multi-Core Benchmark
• Raw Performance of Space Operations in Java and .Net