In the field of Information Technology, systems design involves the process of determining the architecture, interfaces, modules, and data for a system to fulfill certain requirements. It’s a crucial process for improving product/service development efficiencies and enabling a great user experience.
Let’s say you have an application that serves millions of users every day. On the server-side, you need to have excellent engineering to handle such a large volume of requests. Server requests must never fail even if there is an issue with the database or a hardware failure.
However, if the backend or frontend is not designed correctly, a simple mistake could shut down the entire application — everything from the server to the computer-wide network. One of the major sources of such errors is the Single Point Of Failure (SPOF).
In this overview article, we have explained why SPOF is the worst problem for IT professionals, its common source, and how to mitigate such failures. Let’s start with a basic question.
Whats Is Single Point Of Failure?
Definition: A single point of failure is any part of a system that causes the entire system to stop operating if it fails. In simple terms, if one thing breaks, everything goes down.
SPOFs arise due to faulty designs and poor implementation techniques. They are undesirable in any system, be it a software application, hardware module, manufacturing system, or business practice.
For example, if you are running a website that is hosted on only one server on a particular location, then that server would be a single point of failure. If the server fails, visitors won’t be able to access the website. This single point can bring every activity related to your website to a grinding halt. If such a risk exists in your business, you need to take steps to mitigate that risk.
What’s The Solution?
The most effective way to reduce the potential risk of SPOFs is to add redundancy. This involves installing redundant hardware components and software applications.
For example, one could use Redundant Arrays of Independent Disks (RAID) to store Directory Server databases, or deploy multiple, duplicate instances of Directory Servers on different hosts.
Redundancy can be added at different levels. Let’s take the example of an independent taxi driver. At a low level, he may have tools and spare parts to repair the vehicle, in case it breaks down. At the medium level, he may borrow his friend’s taxi to do the job. At the highest level, he may have another car and enough components to completely replace faulty parts in the case of multiple failures.
Now, let’s see the simple example of redundancy in computing.
1.) In a simple setup, there could be several possible single points of failure.
2.) Some single points of failure can be avoided by adding duplicate hardware components.
3). Zero SPOF can be achieved by building a fully redundant system, though this setup is way more expensive than the simple setup shown in figure 1.
Figure 3 | Zero SPOF
At the system level, a load balancer can be deployed to ensure high availability for a cluster of servers. In this case, each server can have multiple hard drives, power supplies, and other modules. A higher level of redundancy can be achieved by adding extra servers that could take on a load of active servers if they fail.
The data center itself supports many operations like business logic. Therefore, it is in itself a potential SPOF for the business, if its features cannot be replicated elsewhere.
At the site (highest) level, the entire data center can be replicated in different settings and accessed when the primary server becomes unresponsive. This type of redundancy is usually the focus of an IT disaster resiliency or recovery program.
In order to have no SPOF, large networks –including the Internet and ARPANET — use packet switching, a technique of routing and transferring data over a digital network into packets. It utilizes the multiple paths between two hosts on the networks and optimally uses the channel capacity.
When there is a fault in any node between the two hosts, the data is transferred through an alternative node. Packet switching also minimizes transmission latency and increases the robustness of communication.
There are three network protocols that are widely used to prevent the single point of failure:
- Intermediate System to Intermediate System moves information efficiently within a computer network by determining the best route for data.
- Open Shortest Path First distributes routing information between routers belonging to a single autonomous system. It uses the shortest path first algorithm (Dijkstra’s algorithm) to transmit data.
- Shortest Path Bridging simplifies the development and configuration of a network while enabling multipath routing.
The three common places SPOF tend to show up are hardware, software, and third-party services/providers. Humans are also a single point of failure in most organizations, but they are often overlooked. People in a business can be SPOFs for several reasons, such as mistakes, fraud, dishonesty, lack of knowledge, and limited experience.
Once you detect SPOF, the next step is to classify it in terms of how difficult it is to fix it. There could be three categories:
- Easy: Can be fixed within a reasonable time and cost.
- Moderate: Cannot be remediated directly; however, a reliable workaround could be developed.
- Difficult: The fault is tricky and very expensive to remediate.
In addition to this, SPOFs can be classified in terms of probability of occurrence (low, medium, and high risk) and their impact on the business (low, medium, and high impact).
Preventing Single Point of Failures
Since so many mission-critical processes depend on network connectivity, data center outages cannot be simply tolerated. Still, over 30% of all data centers experience an outage annually. About 34% of businesses say that one hour of downtime costs more than $1 million.
Regardless of losses that occurred due to data center downtime, the reality is as much as 80% of outages are preventable. While any tool in the network can be a SPOF hazard, most of the outages are caused by malware and other cyber threats.
Modern threat protection tools, including load balancers, Intrusion Prevention Systems, web application firewalls, and Advanced Threat Protection solutions are always at risk during power failures or Network Interface Controller failure, or when they filter internet traffic.
These tools are vulnerable to both ordinary threats like brute force attacks and complex threats such as implementing XML external entities or cross-site request forgery. Since they can’t protect the network all the time, it is necessary to take redundant security measures.
There are several methods to implement web application firewall architecture that minimizes the effectiveness and frequency of a broad range of attacks. Multi-tier web app firewalls, for instance, separates different application modules based on their operations into multiple tiers.
Since each tier runs on an individual system, there is no SPOF. In the same way, properly implementing multiple load balancers can reduce the single point of failure within a network.
Do Not Put All Of Your Eggs In One Basket
Although many companies provide their own version of cloud backups, it is not advisable to rely solely on one backup to protect your business’s data. Even cloud services from tech giants like Amazon, Microsoft, and Google fail multiple times annually.
If you are running a company, you need to take all probable scenarios into account when building redundancy. Do not assume anyone can provide a 100% uptime and always be ready with Plan B if things go wrong.
SPOF Within Systems-of-Systems
Researchers at the Liverpool John Moores University, UK, highlight major challenges faced when integrating individual systems to form large complex heterogeneous Systems-of-Systems.
Many modern approaches tend to concentrate on one small specific vulnerable area. Some of them are highly theoretical or unscalable due to a large number of collaborating components. The study describes how a single point of failure can heavily impact collaborating systems and cause businesses significant financial loss.
Eliminating SPOF in Software-Based Redundancy
Software-based redundancy is usually considered an effective and inexpensive way to enhance reliability. Redundant execution in terms of triple modular redundancy (TMR) is quite popular, but it leaves unprotected SPOFs.
This research presents a holistic approach, named Combined Redundancy, that hardens safety-critical components of a system against soft efforts, while eliminating the vulnerability caused by SPOFs. It leverages redundant execution along with encoded processing, and can be easily integrated into existing projects.
Minimizing SPOFs In Tree Routing
Tree Routing (TR) utilizes parent-child links to transmit packets. These links require more hops when the source and destination nodes belong to different tree branches. Nodes closer to the coordinator may transmit more packets, leading to higher energy consumption and more congestion. This could create a SPOF problem.
A team of researchers at the National Kaohsiung University of Applied Sciences, Taiwan, has developed a Relieving SPOF Tree Routing algorithm to transmit packets via the shortest path and avoid congestion. The algorithm decreases average hop count, minimizes end-to-end delay, increases throughput, and prolongs the lifetime of tree nodes.