The Harvard OpenScholar Drupal distribution provides universities, research institutes, and their affiliates with professional, knowledge-focused websites that allow them to easily show themselves to the world. In addition to Software as a Service (SaaS), they also provide end-to-end services: hosting and full product support, as well as training.
OpenScholar used a customized AWS-based web hosting stack to run Drupal’s multi-site SaaS service. The task of our colleagues was first to stabilize the existing platform and then to increase performance and implement new features and enhancements, following the changing technological and business needs of the customer.
- Insufficient in-house AWS expertise
- Development environments with varying configurations hosted outside of managed infrastructure
- Improper metrics gathering
- Insufficient alerting and notifications
- Unoptimized resource sizing
- Lack of automation to handle day-to-day tasks
- Migrated development environments to similar ones as production on AWS
- Implemented CloudWatch log streaming, metrics collection, dashboards, alerts
- Reconfigured software all around the stack (nginx, Varnish, Apache, php-fpm, etc) and optimized for performance and scalability
- Optimized resource usage and types for cost effectiveness (EC2, RDS, ElastiCache, etc.)
- Provided 24/7 support and monitoring for the infrastructure
- Implemented custom tooling to manage the production infrastructure and development resources
- Decreased infrastructure costs by 30-40% depending on traffic
- Reduced downtimes via analyzing and fixing recurring issues
- Improved service monitoring
- Significantly reduced response time by infrastructure and software optimizations
- AWS: CloudFormation, Elastic Beanstalk, Auto Scaling, Load Balancing, RDS Aurora, ElastiCache, Elastic File System, S3, CloudFront, CloudWatch
- PHP, Ruby
- Varnish, nginx, Apache httpd, php-fpm
Redesigning a product website after years of stable performance is always both a great challenge, and great opportunity.
To serve unexpected visitor peaks and ever-changing datasets, the Cheppers team created a backend solution.