Turbocharge Your Data Monitoring Whilst Slashing Costs

By Roman Khavronenko, Co-Founder of VictoriaMetrics.

  • Wednesday, 5th July 2023 Posted 1 year ago in by Phil Alsop

In the realm of data-heavy businesses, the significance of data monitoring and measurement is steadily increasing. Organisations seeking valuable insights from the influx of information within their ecosystem have likely contemplated the latest monitoring tools available in the market.

Data is omnipresent, and as technology continues to advance, the volume of data continues to increase exponentially. In fact, it has been predicted that by 2025, the world will hold over 200 zettabytes of data, meaning any monitoring solution needs to be operational on an immense scale.

Huge amounts of data have always existed, we just now have increased capabilities to monitor and observe it. New technologies mean we can produce more observations from data, simultaneously contributing to the exponential growth of data, which we couldn’t previously do. However, despite this, many traditional data management approaches often fall short of addressing these challenges, leading to scalability issues and increased costs.

Data monitoring is well-suited for open source solutions, as the innovative and adaptable nature of the community propels progress. The primary challenge in the field of monitoring lies in keeping up with the growing volume of data within. Consequently, any monitoring solution which cannot accommodate this growth will swiftly become obsolete. Hence, scalability becomes imperative. Open source monitoring solutions must strive to achieve scalability that approaches the limits of infinity in order to make a lasting impact. While the term "infinite scalability" is frequently used, it technically defies the laws of physics. Nonetheless, there are various ways in which projects can structure their monitoring systems to optimise scalability.

Vertical and horizontal scalability

The conventional approach to scaling, known as vertical scaling, involves enhancing the resources of a single server to handle increased data or load. This can be achieved by upgrading the CPU, increasing RAM, or expanding storage capacity. Vertical scaling is effective in the short term, and top-notch monitoring solutions excel in scaling across single nodes. However, to achieve scalability that is as close to "infinite" as possible, scaling vertically and horizontally proves to be the most effective method. If enterprises rely solely on vertical scaling, this can result in major costs on hardware and software, and can be a drain on data management time.

Horizontal scaling, also known as sharding, entails distributing a single dataset across multiple databases, which can then be stored on numerous machines. This enables larger datasets to be divided into smaller segments and stored across multiple data nodes, expanding the overall storage capacity of the system and ensuring scalability. Horizontal scaling allows for near-limitless scalability, enabling efficient handling of big data and demanding workloads.

This scalability enables businesses to monitor and analyse vast amounts of data without the need for expensive infrastructure upgrades. Furthermore, open-source solutions offer flexibility, allowing businesses to customise and extend the monitoring capabilities according to their specific needs, eliminating the cost of vendor lock-in.

However, it is equally important to consider the structure of these smaller clusters and the most effective way to scale out, which leads us to the concept of multilayer architecture. Sharding distributes data across smaller compute instances called shards, with each shard holding a portion of the overall database. For instance, with five shards, there would be five compute instances, each holding 1/5th of the total database.

Multilayer Architecture

Multilayer architecture empowers developers to increase the scale of data handled by each shard. This involves assigning additional smaller databases to each individual shard. Consequently, instead of having one large database split across five shards, the original database is retained while five smaller databases are added. For example, instead of having 100 shards, you have 4 shards. And each shard represents a separate database with 25 shards in it. So you logically separate data across multiple databases, while for the user it remains as one single database.

Compression Algorithms

Compression algorithms play a vital role in any scalable monitoring solution by reducing the number of bytes needed to represent data and the amount of memory required to store and transfer it. Effective compression techniques are crucial for maximising the return on investment from available disk space. Simply put, efficient compression algorithms make storing and transferring more data on the same disk easier. Compression also improves data reading and transfer times, resulting in faster overall database performance.

Downsampling

Downsampling is another technique for reducing the number of samples stored required per data series, enabling organisations to increase the amount of data stored on a disk without sacrificing insights into historical trends. Downsampling involves lowering the resolution of older records, using compression algorithms to condense them into a manageable size. This approach provides valuable insights into long-term trends and improves query times for databases.

Simplicity and Single Responsibility

When designing a scalable distributed system, using simple components facilitates future scalability. Simple components should focus on one or two functions and perform exceptionally well without relying on other components. This decoupling enables the independent scaling of individual components. Otherwise, scaling any one component can cause a ripple effect throughout the system. For instance, scaling the serving of read queries

might require scaling the caching layer, which in turn necessitates scaling another component.

Lastly, this approach helps minimise the cost of scaling. While theoretically infinite scaling is (almost) possible, cost efficiency plays a crucial role in realising this potential. The key to achieving cost efficiency lies in simplicity and transparency. The fewer hidden mechanisms and components involved, the greater the efficiency will be.

Cutting costs

Nowadays, with the plethora of data monitoring solutions out there, enterprises can fall into the trap of spending excessive amounts. For example, Coinbase previously invested $65 million on Datadog, when it could have utilised open-source technology instead, and saved copious amounts of money. Monitoring doesn’t need to break the bank. Scalable and cost-efficient solutions have existed for years, so enterprises are self-sabotaging by deploying expensive and sometimes inefficient solutions.

Open-source solutions are free by default or have lower licensing costs compared to proprietary software in the case of open source vendors who offer a commercial version of their software. This reduces upfront expenses for businesses, allowing them to allocate their resources more efficiently. Furthermore, open-source solutions can be deployed on existing hardware infrastructure, eliminating the need for additional hardware investments.

Additionally, many industries have stringent compliance and regulatory requirements, such as data retention, security, or auditing. Open-source time series data monitoring solutions often provide customizable compliance features, ensuring businesses can meet these requirements without incurring additional expenses associated with proprietary software or third-party compliance solutions.

Data Gravity

Open-source solutions allow businesses to deploy monitoring tools in close proximity to their data sources, whether it's on-premises or in the cloud. By placing monitoring agents or collectors near the data, businesses can reduce the need for data movement, thus minimising latency and network bandwidth constraints associated with data gravity.

As open-source solutions are designed to enable businesses to deploy monitoring components across multiple locations or nodes, this distributed approach aligns with the idea of data gravity by allowing businesses to monitor their data where it resides, rather than consolidating it in a central location. This not only reduces data movement but also provides a holistic view of the entire data landscape, including diverse data sources and distributed systems.

In conclusion, leveraging open-source software for data monitoring offers significant advantages in terms of scalability, cost-efficiency, and addressing the challenges of data gravity. By leveraging open-source solutions, businesses can optimise performance, reduce costs associated with data movement, and gain a comprehensive view of their entire data landscape. Moreover, open-source solutions often come with lower licensing costs and can

be deployed on existing infrastructure, enabling businesses to allocate resources efficiently and avoid additional hardware investments.