High Availability vs Fault Tolerance

-
Reliability/Resiliency: Trait of an application/system to avoid or recover from failure with minimal manual intervention.

Availability: Is measure of reliability or the duration of time that a system is performing as expected. If there's performance degradation it implies low availability.

Redundancy (multiple machines) enables loose coupling (achieved using load balancers) which improves performance and availability.

Performance and availability are directly proportional; good availability, good performance, and vice versa. 

High Availability (HA) and Fault Tolerance (FT/continuously available) are used interchangeably when talking about keeping systems running with little or no degradation in their availability.

Though the goal of HA and FT is the same the difference lies in how they are achieved.

HA typically aims at keeping the systems available with a slight degradation of SLAs and can tolerate some downtime. HA is achieved by either provisioning fixed number of servers (typically at least 2 or more) or auto scaling. 

Businesses will define SLAs for a system during normal and impact times. 

Example:

  • During normal operation the service needs to handle 500 transactions per sec.
  • During impact, when one node is down, 100 transactions per sec is acceptable.
HA tries to keep up with the minimum acceptable SLA during an impact by either scaling out to a max number of nodes or manually failing over to a secondary region within defined downtime.

FT also uses multiple nodes or auto-scaling to provide tolerance, but it tries to maintain the SLA defined for normal operation even during an impact. That means the system adjusts to maintain 500 transactions per sec by spinning up nodes automatically or failing over to secondary region with zero downtime. This makes FT complex to implement and expensive compared to HA.

So, to summarize HA tolerates some downtime and compromises on SLA but FT does not and is more complex and expensive. Fault tolerant systems can be said to be continuously available.


Comments

Post a Comment

Popular posts from this blog

Offline Apps Patterns and Tools

-
Apps capable of running offline provide a great user experience even when there's slow/unreliable or no network/internet available. They also minimize data usage and save battery power when run on mobile devices. Design Considerations: Upfront, both front and back end must be designed to work offline.  Think about features that cannot work in offline mode. Single users using the app on multiple devices. Automatically sync data (uni/bi-directional) stored locally on the device with the central database once the app is able to connect to the internet.  If the app is allowed to be used from multiple devices, conflicting data changes made by users and any data sync conflicts need to be handled. To make UI work offline, in the case of Mobile applications, there's nothing to do, but web applications will have to use something called service workers. By implementing a service worker, you can turn a regular web application into a Progressive Web App, making it downloadable and install...
Read more

Private Cloud Platforms

-
The below services let you   take public cloud infrastructure, private data centers or mix both to build a private cloud on-prem. AWS Outposts and Azure Stack are quite similar in that they essentially take public cloud computing services and extend them into on-premises data centers.  Using these frameworks, services available in public clouds like AWS EC2 , or Azure Virtual Machines can be run on on-premises data centers. You can also use the same monitoring and management tools that you would in the public cloud. Azure Stack Hub Azure Stack Hub brings core cloud services to customers' data centers, like virtual machines, storage, networking, VPN gateway and load balancing, as well as services like functions, containers and database. Certified hardware can be purchased from  a variety of partner vendors, such as HPE, Dell EMC, Cisco, Huawei and Lenovo.   Users need to manage their infrastructure, but professional support from Microsoft is available as part of the p...
Read more

Application Delivery in the cloud

-
Traditional load balancers have lately evolved to become intelligent application delivery controllers (ADC) which implement various techniques to improve performance along with balancing load across a group of backend servers. Some of the techniques include caching content, compression of assets (js files, images etc.), taking over SSL related process from web servers, providing security from DDoS, SQL injection, XSS etc.. ADCs are like one-stop-shop for providing performance and security.  Load balancers for non-HTTP(s) workloads: Azure Traffic Manager: Is a global non-HTTP(s)/DNS load balancer. DNS load balancers provide IP address of an healthy endpoint based on the configured rules. The endpoint can be an application gateway or a load balancer depending on whether SSL offloading or application layer processing is required or not. Azure Load Balancer: is a non-HTTP(s) global (cross-region)  load balancer (aka network load balancer) operates at layer 4 (transport laye...
Read more