Why Buying More Storage Doesn't Always Solve a Storage Problem
Storage problems have a habit of being misunderstood. A server becomes slow, someone checks the available disk space, and the discussion quickly turns towards buying additional capacity or moving to faster drives. Sometimes that's exactly what's needed. Quite often, however, the problem has very little to do with the amount of storage available. A system can have terabytes of free capacity and still struggle to deliver acceptable application performance. The reason is that storage has several important characteristics, and capacity is only one of them.
Capacity, throughput, IOPS and latency
Capacity tells us how much information a storage system can hold. Throughput describes the volume of data it can transfer over time. IOPS measures the number of input/output operations it can complete per second, while latency describes how long those operations take. These measurements are related, but they are not interchangeable. A storage system designed to stream large sequential files might deliver excellent throughput while performing poorly with small random operations.
A transactional database, on the other hand, may generate thousands of relatively small reads and writes, often with strict latency requirements. It doesn't necessarily need enormous sequential throughput; it needs predictable response times. This distinction becomes particularly important in virtualised environments where multiple applications share the same underlying storage infrastructure.
SSD doesn't mean unlimited performance
Solid-state storage has transformed infrastructure performance, but it hasn't eliminated contention. Different SSD technologies offer different endurance, latency and write performance characteristics. Controller design, cache behaviour, queue depth, network connectivity and the storage software itself can all influence results. An NVMe-backed system may still experience high latency if it's overloaded or if network and replication overhead become limiting factors.
Distributed storage adds another layer. With replicated storage, a logical write may involve multiple physical operations. A three-replica design, for example, consumes roughly three times the raw storage for replicated data before accounting for other overheads. That redundancy has a purpose, but it must be reflected in capacity planning and performance expectations.
Don't forget what happens during failure
Storage should be sized for more than normal operation. When a disk or node fails, a resilient system may need to rebuild or redistribute data while continuing to serve applications. Those recovery operations consume bandwidth, CPU and storage performance. If the platform is already running close to its limits, the rebuilding process can place additional pressure on production workloads. Good storage design therefore includes operational headroom, failure recovery considerations and appropriate monitoring.
Not every workload needs the fastest available storage. A backup repository, an archive, a general-purpose virtual machine and a high-transaction database have very different requirements. Using premium storage for everything can be unnecessarily expensive, while using capacity-oriented storage for latency-sensitive applications can be equally problematic. At KORE, storage decisions are part of workload design rather than simply a price-per-gigabyte calculation. We look at the intended application, its I/O behaviour, growth requirements, redundancy expectations and recovery needs. Storage isn't just about how much data you can fit into a system. It's about what happens when that data needs to be accessed.