disbalancer-node

README

Gladius Node (Golang version)

The full suite of binaries for running a Gladius Node

Development

Dependencies

To test and build the gladius binaries you need go, glide and the make on your machine.

• Install go
• Install glide
• Mac Users: Install xcode for make xcode-select --install
• Windows Users: Install Linux Subsystem

Install dependencies

We use glide to manage the go dependencies. To install the dependencies you need to executethe dependencies target.

# install depdencies for the project with glide
make dependencies

Build

To build all binaries for your current os and architecture simply execute make. After the build process you will find all binaries in ./build/.

Build specific binary

The Makefile can build single binaries too.

# build only the cli
make cli

# build the network daemon
make networkd

# build the control daemon (not implemented yet)
make controld

Build for a different platform

To build for a different platform specify toe GOOS and GOARCH variable.

# build for windows 64bit
GOOS=windows GOARCH=amd64 make

# build for linux 32bit
GOOS=linux GOARCH=386 make

Usage

Run the binaries as a process

Run the executable created by the above step with gladius-<executable-name>

Run networkd or controld as a service

You can also install networkd and controld as a service. Attention: The service implementation is not tested.

# install networkd as a service
gladius-networkd install

# start the networkd service
gladius-networkd start

# stop the networkd service
gladius-networkd stop

CLI

TODO

Network Daemon

Test the RPC server (Only Start and Stop work now)

$ HDR1='Content-type: application/json'
$ HDR2='Accept: application/json'

$ MSG='{"jsonrpc": "2.0", "method": "GladiusEdge.Start", "id": 1}'
$ curl -H $HDR1 -H $HDR2 -d $MSG http://localhost:5000/rpc
{"jsonrpc":"2.0","result":"Started server","id":1}

$ MSG='{"jsonrpc": "2.0", "method": "GladiusEdge.Stop", "id": 1}'
$ curl -H $HDR1 -H $HDR2 -d $MSG http://localhost:5000/rpc
{"jsonrpc":"2.0","result":"Stopped server","id":1}

$ MSG='{"jsonrpc": "2.0", "method": "GladiusEdge.Status", "id": 1}'
$ curl -H $HDR1 -H $HDR2 -d $MSG http://localhost:5000/rpc
{"jsonrpc":"2.0","result":"Not implemented","id":1}

Set up content delivery

Right now files are loaded from ~/.config/gladius/gladius-networkd/ and take the format of example.com.json. This functionality only works on linux right now, and serving is not backwards compatible with the previous release. Content can then be accessed at http://<host>:8080/content?website=example.com

---

Some benchmarks compared to the previous version

Done over a gigabit link between two machines with the same bundle file being served.

Node version (with express routing)

ab -n 5000 -c 1000 http://<remote IP>:8080/content_bundle

This is ApacheBench, Version 2.3 <$Revision: 1796539 $>
Copyright 1996 Adam Twiss, Zeus Technology Ltd, http://www.zeustech.net/
Licensed to The Apache Software Foundation, http://www.apache.org/

Benchmarking <remote IP> (be patient)
Finished 5000 requests


Server Software:        
Server Hostname:        <remote IP>
Server Port:            8080

Document Path:          /content_bundle
Document Length:        452460 bytes

Concurrency Level:      1000
Time taken for tests:   32.079 seconds
Complete requests:      5000
Failed requests:        0
Total transferred:      2263530000 bytes
HTML transferred:       2262300000 bytes
Requests per second:    155.87 [#/sec] (mean)
Time per request:       6415.760 [ms] (mean)
Time per request:       6.416 [ms] (mean, across all concurrent requests)
Transfer rate:          68907.77 [Kbytes/sec] received

Connection Times (ms)
              min  mean[+/-sd] median   max
Connect:        4  268 611.6     39    3066
Processing:  1095 3654 3755.5   2974   31027
Waiting:       18 1029 3766.7     92   30872
Total:       1112 3922 3957.5   3072   32070

Percentage of the requests served within a certain time (ms)
  50%   3072
  66%   3402
  75%   3549
  80%   3596
  90%   5293
  95%   8859
  98%  18118
  99%  31110
 100%  32070 (longest request)

Go version

ab -n 5000 -c 1000 http://<remote IP>:8080/content\?website\=test.com

This is ApacheBench, Version 2.3 <$Revision: 1796539 $>
Copyright 1996 Adam Twiss, Zeus Technology Ltd, http://www.zeustech.net/
Licensed to The Apache Software Foundation, http://www.apache.org/

Benchmarking <remote IP> (be patient)

Finished 5000 requests


Server Software:        fasthttp
Server Hostname:        <remote IP>
Server Port:            8080

Document Path:          /content?website=test.com
Document Length:        452461 bytes

Concurrency Level:      1000
Time taken for tests:   19.265 seconds
Complete requests:      5000
Failed requests:        0
Total transferred:      2263050000 bytes
HTML transferred:       2262305000 bytes
Requests per second:    259.54 [#/sec] (mean)
Time per request:       3853.006 [ms] (mean)
Time per request:       3.853 [ms] (mean, across all concurrent requests)
Transfer rate:          114716.15 [Kbytes/sec] received

Connection Times (ms)
              min  mean[+/-sd] median   max
Connect:        9  185 365.8     47    3089
Processing:    14 3534 458.6   3622    6552
Waiting:        3   76  99.7     48    2276
Total:         32 3719 637.0   3677    7604

Percentage of the requests served within a certain time (ms)
  50%   3677
  66%   3706
  75%   3749
  80%   3919
  90%   4662
  95%   4834
  98%   5144
  99%   5535
 100%   7604 (longest request)

As you can see above, the Go version handles high concurrent request loads significantly better than the Node.js version of the network daemon. The Go version also saturates the Gigabit link, which likely means it can provide even more performance.