APM tools are your window into your application's performance — its capacity and levels of service. These tools help admins conduct regular health checks on the app so they can tell the state of the app without any ambiguity.
Any application is made up of its layers and its subsystems — the servers, the virtualization layers, the dependencies, and its components. The purpose of such tools has traditionally been to monitor the performance of all the subsystems.
A traditional approach to APM involved the use of arbitrary sampling strategies, algorithm-based data completion, and a fair bit of prediction to analyze the root cause. So, the agents had to come up with a hypothesis of why things were wrong and devise a sampling strategy to test that theory. Any data gaps were predictively filled by algorithms.
Automation is one of the many ways that founders can scale their business. As organizations grow, their automated processes will only generate more data, not less. As automation seeps into every facet of the digital enterprise, the applications interfacing organizations with their audience generate large swathes of raw, unsampled data.
Traditional APM tools are now struggling due to the mismatch between their specifications and expectations.
Modern application architectures are multi-faceted; they contain hybrid components across a variety of on-premise and cloud applications. Modern enterprises often generate data in silos with each outflow having its own data structure. This data comes from several tools over different periods of time.
Such diversity in sources, structure, and formats present unique challenges for traditional enterprise tools.
1. Inability to handle massive, multi-dimensional data
As discussed before, modern applications are not atomic; they are constituent of several components and subsystems all of which contribute to its overall performance.
Each subsystem can produce several terabytes of data. Such scale of data brings forth at least a few problems with the earlier-generation APM tools:
■ The efficient storage of and access to this data is a peculiar challenge.
■ Real-time analysis of this data on the mammoth-scale is an even bigger challenge for traditional APM tools.
■ Often the data may be multiple types of data sources — in flat files, structured query-based databases, or even complete systems of their own with API-based access.
2. Propagation of fragmentation into APM tools
Often, we see new tools for each functional area even within the same data center. This fuels silo creation as segregated teams support individual tools for managing the server, network, storage, and virtual layers.
A count of anywhere between 6 to 10 tools would not be uncommon. Each of these proprietary tools may come with vendor lock-in, forcing companies to continue using them with restrictions or pay more when the usage increases.
This is not ideal for enterprises as most modern applications are dynamic and interdependent in nature. For example, as user-base increases, a single business request to increase capacity will mean synchronous updating and coordination among silos for databases, servers, networks, and virtual layers.
At the intersection of these functional areas, agents do the job of coordinating the data and passing on the configurations. Without a cohesive plan to manage these agents (automated or manual), it becomes difficult to collectively address issues to optimize their efficiency.
Due to the fragmentation in tools, other issues like long-term licensing come to surface and companies have to keep paying for these tools over the long term. One possible solution is to outsource product development. This way companies can target multiple functionalities with a single custom-developed app and finite vendor contracts.
3. Security risks during seasonal spikes
To proactively identify problems, these tools rely on detecting anomalies in data sources that are infrastructure-centric. This would typically include log files, memory metrics, CPU usage, and so on.
If there are seasonal spikes, such as massive holiday sales like Black Friday, the admins would be flooded with spikes across the board. Hiding an attack in between these spikes becomes easier as most traditional APM tools can't differentiate between these spikes from distributed denial of service (DDoS) attacks.
4. Difficulty in root-cause analysis
Agents can stitch together data from various systems to identify root cause of major problems. To detect anomalies, agents identify patterns and then use queries to confirm their assumptions of a diagnosis.
Because of human involvement in the diagnosis process, there is a strong possibility of selection/sampling bias being introduced in the process.
Also, these analyses are estimates at best as they rely on testing a hypothesis.
An accurate, tools-agnostic analysis of the root cause requires not only identifying anomalies but patterns of these aberrations over time. This is where traditional APM tools fall short and predictive analysis tools truly shine.
Final Words
Traditional APM tools lack the capacity to handle the scale of data being generated by modern applications. Also, these applications generally occupy status of legacy apps in enterprises, which makes replacing them even more difficult.
So, while management is likely to see them as roadblocks, removing these legacy apps completely from the enterprise would mean ripping the band-aid off. It is a hard decision to make and one that requires a fair bit of convincing and strategy.
This might look like hard work, but it is better than letting these roadblocks continue to slow your processes down. It is important to take action before the damage becomes critical.