#scaling

Over the last couple of years, the popularity of the “cloud computing” has grown dramatically and along with it so has the dominance of Amazon Web Services (AWS) in the market. Unfortunately, AWS doesn’t do a great job of explaining exactly what AWS is, how its pieces work together, or what typical use cases for its components may be. This post is an effort to address this by providing a whip around overview of the key AWS components and how they can be effectively used.

Great, so what is AWS? Generally speaking, Amazon Web Services is a loosely coupled collection of “cloud” infrastructure services that allows customers to “rent” computing resources. What this means is that using AWS, you as the client are able to flexibly provision various computing resources on a “pay as you go” pricing model. Expecting a huge traffic spike? AWS has you covered. Need to flexibly store between 1 GB or 100 GB of photos? AWS has you covered. Additionally, each of the components that makes up AWS is generally loosely coupled meaning that they can work independently or in concert with other AWS resources.

Since AWS components are loosely coupled, you’d be able to mix and match only what you need but here is an overview of the key services.

Route53

What is it? Route53 is a highly available, scalable, and feature rich domain name service (DNS) web service. What a DNS service does is translate a domain name like “setfive.com” into an IP address like 64.22.80.79 which allows a client’s computer to “find” the correct server for a given domain name. In addition, Route53 also has several advanced features normally only available in pricey enterprise DNS solutions. Route53 would typically replace the DNS service provided by your registrar like GoDaddy or Register.com.

Should you use it? Definitely. Allow it isn’t free, after last year’s prolonged GoDaddy outage it’s clear that DNS is a critical component and using a company that treats it as such is important.

Simple Email Service

What is it? Simple Email Service (SES) is a hosted transactional email service. It allows you to easily send highly deliverable emails using a RESTful API call or via regular SMTP without running your own email infrastructure.

Should you use it? Maybe. SES is comparable to services like SendGrid in that it offers a highly deliverable email service. Although it is missing some of the features that you’ll find on SendGrid, its pricing is attractive and the integration is straightforward. We normally use SES for application emails (think “Forgot your password”) but then use MailChimp or SendGrid for marketing blasts and that seems to work pretty well.

Identity and Access Management

What is it? Identity and access management (IAM) provides enhanced security and identity management for your AWS account. In addition, it allows you to enable “multi factor” authentication to enhance the security of your AWS account.

Should you use it? Definitely. If you have more than 1 person accessing your AWS account using IAM will allow everyone to get a separate account with fine grained permissions. Multi factor authentication is also critically important since a compromise at the infrastructure level would be catastrophic for most businesses. Read more about IAM here.

Simple Storage Service

What is it? Simple storage service (S3) is a flexible, scalable, and highly available storage web service. Think of S3 like having an infinitely large hard drive where you can store files which are then accessible via a unique URL. S3 also supports access control, expiration times, and several other useful features. Additionally, the payment model for S3 is “pay as you go” so you’ll only be billed for the amount of data you store and how much bandwidth you use to transfer it in and out.

Should you use it? Definitely. S3 is probably the most widely used AWS service because of its attractive pricing and ease of use. If you’re running a site with lots of static assets (images, CSS assets, etc.), you’ll probably get a “free” performance boost by hosting those assets on S3. Additionally, S3 is an ideal solution for incremental backups, both data and code. We use S3 extensively, usually for hosting static files, frequently backing up MySQL databases, and backing up git repositories. The new AWS S3 Console also makes administering S3 and using it non-programmatically much easier.

Elastic Compute Cloud

What is it? Elastic Compute Cloud (EC2) is the central piece of the AWS ecosystem. EC2 provides flexible, on-demand computing resources with a “pay as you go” pricing model. Concretely, what this means is that you can “rent” computing resources for as long as you need them and process any workload on the machines you’ve provisioned. Because of its flexibility, EC2 is an attractive alternative to buying traditional servers for unpredictable workloads.

Should you use it? Maybe. Whether or not to use EC2 is always a controversial discussion because the complexity it introduces doesn’t always justify its benefits. As a rule of thumb, if you have unpredictable workloads like sporadic traffic using EC2 to run your infrastructure is probably a worthwhile investment. However, if you’re confident that you can predict the resources you’ll need you might be better served by a “normal” VPS solution like Linode.

Elastic Block Store

What is it? Elastic block store (EBS) provides persist storage volumes that attach to EC2 instances to allow you to persist data past the lifespan of a single EC2. Due to the architecture of elastic compute cloud, all the storage systems on an instance are ephemeral. This means that when an instance is terminated all the data stored on that instance is lost. EBS addresses this issue by providing persistent storage that appears on instances as a regular hard drive.

Should you use it? Maybe. If you’re using EC2, you’ll have to weigh the choice between using only ephemeral instance storage or using EBS to persist data. Beyond that, EBS has well documented performance issues so you’ll have to be cognizant of that while designing your infrastructure.

CloudWatch

What is it? CloudWatch provides monitoring for AWS resources including EC2 and EBS. CloudWatch enables administrators to view and collect key metrics and also set a series of alarms to be notified in case of trouble. In addition, CloudWatch can aggregate metrics across EC2 instances which provides useful insight into how your entire stack is operating.

Should you use it? Probably. CloudWatch is significantly easier to setup and use than tools like Nagios but its also less feature rich. We’ve had some success coupling CloudWatch with PagerDuty to provide alerts in case of critical service interruptions. You’ll probably need additional monitoring on top of CloudWatch but its certainly a good baseline to start with.

Anyway, the AWS ecosystem includes several additional services but these are the ones that I felt are key to getting started on AWS. We haven’t had a chance to use it yet but Redshift looks like it’s an exciting addition which will probably make this list soon. As always, comments and feedback welcome.

Posted In: Amazon AWS

Tags: , ,

Over the past month we’ve been working with Her Campus (HerCampus.com) to help them with issues they were having. When we started talking with Her Campus, we learned that they had different types of issues ranging from some Drupal based ones to actual server level problems. They were having some trouble keeping up to traffic demands, and if a traffic spike occurred their current infrastructure wasn’t sufficient.

After looking at their setup, we noted quite a few areas in which we could improve performance. The old setup was a fairly standard setup, a frontend server using Apache to handle HTTP requests and then a second server which was their MySQL database server. The servers were a 4 gig and 8 gig server respectively.

It was clear Apache was adding unneeded overhead, and wasn’t the best solution for them. We revamped their setup significantly. We switched them to Nginx + PHP-FPM. We immediately saw great improvement on from this change alone. However we wanted to get them to a single server, and to be able to handle traffic spikes with a single server.

We ended up doing the following:

  • Switch from Apache -> Nginx+PHP-FPM
  • Update all MyISAM tables to INNODB tables, and upgrade their MySQL to 5.1
  • Tune MySQL settings to fit their requirements
  • Update several tables adding indexes, reducing query time from in one case 34 seconds to 0.02 seconds.
  • Add the Boost Module to their setup
  • Update several of the view queries to be better written, added caching to each query.
  • Use ImageCache and sub-domains to load assets

After these updates we were able to move them from their two servers (8 gig and 4 gig) to a single server(4 gig). We have also reduced load times significantly. Their server loads dropped from 4-5 on average to 0.25. Recently they had an article on the Huffington Post and didn’t have any problems handling the 4x traffic spike they saw. At points we were seeing according to ChartBeat over 600 people on the site at once. The single server handled this without problems.

The updated infrastructure will give them a savings of about 75% from their previous setup. It also gives the users on the site a much faster and reliable experience.

We look forward to helping Her Campus with their continuing expansion of their site and user base!

Posted In: General

Tags: , , , , ,

For a client we’ve been working with recently it came to our attention that they needed more frontend servers to keep up with the traffic for their surveys. They use LimeSurvey which is powerful open source survey platform. We set the client up in the cloud to scale as necessary with a load balancer in front. This is when we noticed the problem that LimeSurvey doesn’t work well when a user is bouncing between different frontend servers. LimeSurvey keeps all the user’s session attributes on the local server and not in the database. After googling around for a while, we found other people also had this problem before, and no one had really solved it. We figured we would.

We didn’t feel like doing a ton of extra work to reinvent the wheel in terms of storing the session in the database. We snagged most of the code straight from the Symfony “storage” module which handles it’s session management if you want to store the user sessions in a database. After a quick few modifications, we got it up:

This requires you create a table in your MySQL database called session. Here is a dump of the create statement for the table:

CREATE TABLE IF NOT EXISTS `session` (
  `id` int(11) NOT NULL auto_increment,
  `sess_id` varchar(255) NOT NULL,
  `sess_data` text NOT NULL,
  `sess_time` datetime NOT NULL,
  PRIMARY KEY  (`id`)
) ENGINE=MyISAM  DEFAULT CHARSET=latin1 AUTO_INCREMENT=1;

Basically this uses PHP’s session_set_handler function to manipulate how the PHP retrieves, updates, and stores the user’s session. The final touches were to include this class where the user sessions are started in LimeSurvey. We found them in the index.php and sessioncontrol.php files. Include our file from above just before the session_start(); in the code in those two files. In admin/scripts/fckeditor.265/editor/filemanager/connectors/php/upload.php include the file before the first include. Lastly we need to update a couple locations where it does session_regenerate_id and replace it with $sfSessionHandler->regenerate(). You can find these edits in the following three files: admin/usercontrol.php on line 128, admin/login_check.php line 65, and index.php at lines 207 and 215. You should be up and running now, let us know if you have any problems.

Posted In: Launch, PHP

Tags: , , , , ,