Menu
— I built a simple app that pulls in data (50 items) from a Redis DB and throws it up at localhost. I did an ApacheBench (c = 100, n = 50000) and I'm getting a semi-decent 150 requests/sec on a dual-core T2080 @ 1.73GHz (my 6 y.o laptop), but the proc usage is very disappointing as shown: 
Only one core is used, which is as per design in Node, but I think I can nearly double my requests/sec to ~300, maybe even more, if I can use Node.js clusters. I fiddled around quite a bit but I haven't been able to figure out how to put the code given here for use with my app which is listed below:
var express = require( 'express' ), app = express.createServer(), redis = require( 'redis' ).createClient(); app.configure( function() { app.set( 'view options', { layout: false } ); app.set( 'view engine', 'jade' ); app.set( 'views', __dirname + '/views' ); app.use( express.bodyParser() ); } ); function log( what ) { console.log( what ); } app.get( '/', function( req, res ) { redis.lrange( 'items', 0, 50, function( err, items ) { if( err ) { log( err ); } else { res.render( 'index', { items: items } ); } }); }); app.listen( 8080 ); I also want to emphasize that the app is I/O intensive (not CPU-intensive, which would've made something like threads-a-gogo a better choice than clusters).
Would love some help in figuring this out.
464
Node. js runs single threaded programming, which is very memory efficient, but to take advantage of computers multi-core systems, the Cluster module allows you to easily create child processes that each runs on their own single thread, to handle the load.
The Start-ClusterNode cmdlet starts the Cluster service on a node in a failover cluster. If this is the first node started, then it will wait for other nodes to join. The cluster will begin to run when a quorum has formed.
A Look at ClusteringThe Node. js Cluster module enables the creation of child processes (workers) that run simultaneously and share the same server port. Each spawned child has its own event loop, memory, and V8 instance. The child processes use IPC (Inter-process communication) to communicate with the parent Node.
In a single thread, the individual instance of node. js runs specifically and to take advantage of various ecosystems, a cluster of node. js is launched, to distribute the load. With the help of a cluster module, child processes can be created very easily sharing the server ports.
Actually, your workload is not really I/O bound: it is CPU bound due to the cost of jade-based dynamic page generation. I cannot guess the complexity of your jade template, but even with simple templates, generating HTML pages is expensive.
For my tests I used this template:
html(lang="en") head title Example body h1 Jade - node template engine #container ul#users each user in items li User:#{user} I added 100 dummy strings to the items key in Redis.
On my box, I get 475 req/s with node.js CPU at 100% (which means 50% CPU consumption on this dual core box). Let's replace:
res.render( 'index', { items: items } ); by:
res.send( '<html lang="en"><head><title>Example</title></head><body><h1>Jade - node template engine</h1><div id="container"><ul id="users"><li>User:NOTHING</li><li>User:NOTHING</li><li>User:NOTHING</li><li>User:NOTHING</li><li>User:NOTHING</li><li>User:NOTHING</li><li>User:NOTHING</li><li>User:NOTHING</li><li>User:NOTHING</li><li>User:NOTHING</li><li>User:NOTHING</li><li>User:NOTHING</li><li>User:NOTHING</li><li>User:NOTHING</li><li>User:NOTHING</li><li>User:NOTHING</li><li>User:NOTHING</li><li>User:NOTHING</li><li>User:NOTHING</li><li>User:NOTHING</li><li>User:NOTHING</li><li>User:NOTHING</li><li>User:NOTHING</li><li>User:NOTHING</li><li>User:NOTHING</li><li>User:NOTHING</li><li>User:NOTHING</li><li>User:NOTHING</li><li>User:NOTHING</li><li>User:NOTHING</li><li>User:NOTHING</li><li>User:NOTHING</li><li>User:NOTHING</li><li>User:NOTHING</li><li>User:NOTHING</li><li>User:NOTHING</li><li>User:NOTHING</li><li>User:NOTHING</li><li>User:NOTHING</li><li>User:NOTHING</li><li>User:NOTHING</li><li>User:NOTHING</li><li>User:NOTHING</li><li>User:NOTHING</li><li>User:NOTHING</li><li>User:NOTHING</li><li>User:NOTHING</li><li>User:NOTHING</li><li>User:NOTHING</li><li>User:NOTHING</li><li>User:NOTHING</li></ul></div></body></html>' ); Now, the result of the benchmark is close to 2700 req/s. So the bottleneck is clearly due to the formatting of the HTML page.
Using the cluster package in this situation is a good idea, and it is straightforward. The code can be modified as follows:
var cluster = require('cluster') if ( cluster.isMaster ) { for ( var i=0; i<2; ++i ) cluster.fork(); } else { var express = require( 'express' ), app = express.createServer(), redis = require( 'redis' ).createClient(); app.configure( function() { app.set( 'view options', { layout: false } ); app.set( 'view engine', 'jade' ); app.set( 'views', __dirname + '/views' ); app.use( express.bodyParser() ); }); function log( what ) { console.log( what ); } app.get( '/', function( req, res ) { redis.lrange( 'items', 0, 50, function( err, items ) { if( err ) { log( err ); } else { res.render( 'index', { items: items } ); } }); }); app.listen( 8080 ); } Now the result of the benchmark is close to 750 req/s with 100 % CPU consumption (to be compared with the initial 475 req/s).
195
If you love us? You can donate to us via Paypal or buy me a coffee so we can maintain and grow! Thank you!
Donate Us With
