Deployment Options
The section looks at the various deployment options that you have for Aiimi Insight Engine. These are guides and are not representative of what your deployment will look like.
Development Environments
For a development environment we recommend a single server set up. This should contain all the Aiimi Insight Engine components.
Single Server Requirement
A laptop with the following requirements is ample for a single server installation of Aiimi Insight Engine.
4 CPU Cores
16GB RAM
250GB Disk Space
You should be able to ingest a few hundred thousand documents and continue development and testing activities.
Two Server Set Up
Developers may set the whole platform up on a laptop or workstation computer using the below split.
One server should host the repository.
One server should host the agents and the gateway.
Testing and QA Environments
The key decision about your QA environment is driven do you want to do performance testing. If you do, how much should it match the production environment in terms of scale.
End-user performance testing of ‘search’ related transactions
The elastic cluster and web server should be scaled as production.
The QA system should contain the same volume of content.
If you have not scaled the QA platform to match production, you can extrapolate some estimations for performance. Please note this is not an exact science, and the only true test is to have the QA platform sized the same as production.
Strict pre-production testing environments:
If production has a separate web server and agent servers so should QA.
If production has multiple agent servers, QA should have at least 2 agent servers.
If production has separate proxy nodes on the elastic cluster so should QA.
If production uses a multi node elastic cluster than QA should.
Performance Testing
For performance testing your Aiimi contact can provide separate guidelines.
The most significant process for end user response times is search and aggregation requests.
For performance, test the Elasticsearch cluster and the web servers being scaled.
If you want to test ingestion throughput ensure the Elasticsearch cluster and Agent servers are scaled.
Production Environments
Production environments need to be designed and scaled with your specific requirements.
Below are small, medium, and large sizing guides for each type of server. This can help as a starting point when designing your platform.
For a medium or large platform you will have multiple servers for the agent and repository tiers.Scaling Agent Servers
Small – 4 CPU cores, 16GB RAM, 100GB local disk
Medium – 8 CPU cores, 32GB RAM, 100GB local disk
Large – 16 CPU cores, 64GB RAM, 100GB local disk
Scaling Repository Servers
Small – 8 CPU cores, 16GB RAM, 200GB SSD local disk
Medium – 16 CPU cores, 32GB RAM, 400GB SSD local disk
Large – 32 CPU cores, 64GB RAM, 800GB SSD local disk
Proxy servers do not need the same amount of memory and local disk since they store no data. For this 16GB RAM will be sufficient and the CPU loading will increase with transactional concurrency.
Scaling Web Servers
Small – 4 CPU cores, 16GB RAM, 100GB local disk
Medium – 8 CPU cores, 32GB RAM, 100GB local disk
Large – 16 CPU cores, 32GB RAM, 100GB local disk
Production Platform Considerations
Separate agent and web servers should be used.
The number of agent servers required depends on:
The number of source repositories
Their respective size
Pipeline complexity
Content Ingestion rate
In most scenarios only a single web server will be required.
If you use two you will need to load balance the user requests using a standard load balancer.
Load balancing should be ‘sticky’ – i.e. send the same users requests to the same web server
If OCR is heavily used, then this should have it's own agent server.
Separate proxy nodes in the repository tier.
Multiple elastic servers and limit the maximum amount of memory per-node to 64GB.
Disaster Recovery Environments
Disaster recovery environments need to consider the service needed during a ‘disaster scenario’.
The business criticality of the system can be scaled accordingly. If you need the same level of users and ingestion rates, the recovery platform should replicate the production environment.