RelaxFactory is a Server Toolkit for implementing web protocols like HTTP, QUIC, WebSockets, and REST middleware using Java language but without a java resource cost associated with application containers. This is accomplished by carefully proceeding in whatever way avoids object allocation and object reference longevity. The C10k Reactor pattern implentation imposes no overhead on inbound sockets and in fact, knows nothing about the sockets outside of the spartan protocol dispatch routines.
The first implementation features of RelaxFactory produced a high quality GWT middle-tier for couchdb data persistence. Smaller professional engagements have brought the toolkit to bear on iterations bridging hadoop, natural language processing tools, and RelaxFactory powers development web sites on major cloud providers.
benchmarks below on my intel core-i7 4771 8-core
Developer Usecases:
- edge node high-throughput/low-resource design
- streaming java async events on couchdb _changes
- hosting GWT with long-poll/channels via couch REST _changes
- HA RequestFactory concurrency via "The Big Merge" couchdb
- basis for KouchDb platform for re-implementing known-good web based document store features
- Suitable platform to build java QUIC serverside
- new! ssl client and server options via 1xio
RelaxFactory is dependent upon 1xio (https://github.com/jnorthrup/1xio) and potentially other projects that have not yet become hosted in maven central. to use RelaxFactory without building it add
<repository>
<id>misc-repo</id>
<url>http://raw.github.com/jnorthrup/misc-repo/master/</url>
</repository>
to your pom.xml
// -D1xio.realtime.unit="MINUTES" // -D1xio.realtime.cutoff="3" // -D1xio.debug.sendjson="false"
Server Software: null Server Hostname: localhost Server Port: 8989 Document Path: http://localhost:8989/rnd?c=2048 Document Length: 2048 bytes
Concurrency Level: 50 Time taken for tests: 23.231000 seconds Complete requests: 5000000 Failed requests: 0 Write errors: 0 Kept alive: 5000000 Total transferred: 10240000000 bytes Requests per second: 215,229.65 [#/sec] (mean) Time per request: 0.232 [ms] (mean) Time per request: 0.005 [ms] (mean, across all concurrent requests) Transfer rate: 440,790.32 [Kbytes/sec] received -1 kb/s sent 440,790.32 kb/s total
Server Software: Jetty(9.3.0.M2) Server Hostname: localhost Server Port: 8989 Document Path: http://localhost:8989/rnd?c=2048 Document Length: 2048 bytes
Concurrency Level: 50 Time taken for tests: 24.320000 seconds Complete requests: 5000000 Failed requests: 0 Write errors: 0 Kept alive: 5000000 Total transferred: 10240000000 bytes Requests per second: 205,592.11 [#/sec] (mean) Time per request: 0.243 [ms] (mean) Time per request: 0.005 [ms] (mean, across all concurrent requests) Transfer rate: 421,052.63 [Kbytes/sec] received -1 kb/s sent 421,052.63 kb/s total
Server Software: HttpCore-NIO-Test/1.1 Server Hostname: localhost Server Port: 8989 Document Path: http://localhost:8989/rnd?c=2048 Document Length: 2048 bytes
Concurrency Level: 50 Time taken for tests: 24.626000 seconds Complete requests: 5000000 Failed requests: 0 Write errors: 0 Kept alive: 5000000 Total transferred: 10240000000 bytes Requests per second: 203,037.44 [#/sec] (mean) Time per request: 0.246 [ms] (mean) Time per request: 0.005 [ms] (mean, across all concurrent requests) Transfer rate: 415,820.68 [Kbytes/sec] received -1 kb/s sent 415,820.68 kb/s total
Server Software: null Server Hostname: localhost Server Port: 8989 Document Path: http://localhost:8989/rnd?c=2048 Document Length: 2048 bytes
Concurrency Level: 50 Time taken for tests: 22.801000 seconds Complete requests: 5000000 Failed requests: 0 Write errors: 0 Kept alive: 5000000 Total transferred: 10240000000 bytes Requests per second: 219,288.63 [#/sec] (mean) Time per request: 0.228 [ms] (mean) Time per request: 0.005 [ms] (mean, across all concurrent requests) Transfer rate: 449,103.11 [Kbytes/sec] received -1 kb/s sent 449,103.11 kb/s total
overall profiler shows roughly equivalent benchmark camel humps in the cpu score.
benchmarks seem to show allocation of Direct ByteBuffers allocated in force in rxf, with the other 3 implementations presumably using heap or possibly fast "Unsafe" shortcuts.
