IOPS Glossary

Combined IOPS cap across all attached volumes.

Mean completion time for I/O operations.

Low-level storage abstraction used by operating systems.

Temporarily higher IOPS available for short durations.

I/O request served from cache without disk access.

I/O request requiring access to underlying storage.

Cost model based on provisioned or consumed IOPS.

Infrequently accessed data with low IOPS needs.

Trade-off between performance, availability, and correctness.

IOPS demand driven by transactional databases.

Storage resources reserved for a single tenant or workload.

Burst write IOPS when memory pages are persisted to disk.

Time required for an HDD head to locate data.

Storage system spread across multiple nodes.

Write amplification caused by parity calculations.

IOPS consumed by filesystem metadata operations.

IOPS capability of spinning disks, typically much lower than SSDs.

Distinction between IOPS observed at the host (VM / bare metal) and IOPS at the storage backend, which may be higher due to RAID, erasure coding, or write amplification.

Frequently accessed data requiring high IOPS.

I/O processing layer between VM and physical storage.

CPU idle time spent waiting for I/O completion.

Additional IOPS caused by maintaining database indexes.

Per-instance cap that can throttle IOPS.

Mechanism allowing burst IOPS based on accumulated credits.

Storage-specific delay per I/O request.

Buffer holding pending I/O operations.

OS component that orders and prioritizes I/O requests.

Amount of data transferred per I/O operation, typically 4K–256K.

Measure of how many read or write operations a storage system can perform per second.

Specific mix of block size, queue depth, and access pattern.

Maximum achievable IOPS under ideal conditions.

Stability of IOPS delivery over time.

Cost paid per unit of delivered IOPS.

IOPS per GB of provisioned capacity (e.g., IOPS/GB), used to compare how “performance-dense” different storage tiers or volume types are.

Reserved IOPS capacity for traffic spikes.

Continuous tracking of IOPS usage and limits.

Allocating more IOPS than actually required.

IOPS that can be processed per CPU core, including kernel, hypervisor, and protocol overhead, often a limiting factor for very fast NVMe arrays.

Normalized metric to compare storage value.

Maximum aggregate IOPS allowed by a compute instance.

Maximum IOPS supported by a single storage volume.

Safety buffer added during capacity planning.

Matching IOPS capacity to real workload demand.

Load level where additional demand no longer improves performance.

Contractual guarantee for IOPS performance.

Artificial limiting of IOPS to enforce fairness or pricing tiers.

Fluctuation in IOPS due to contention or burst behavior.

Scenario where storage limits performance before CPU.

Scenario where storage access dominates due to insufficient caching.

IOPS measures operation count, throughput measures data volume.

Application limited by storage I/O capacity.

Scaling storage or instances based on IOPS consumption.

Filesystem that increases write IOPS to ensure consistency.

I/O with larger blocks that emphasizes throughput over IOPS.

Time taken to complete a single I/O operation.

Relationship showing latency increase as IOPS approach saturation.

Application sensitive to I/O delays more than throughput.

IOPS measured under combined read and write workloads.

Shared storage infrastructure across multiple users.

IOPS degradation caused by contention from other workloads.

High IOPS enabled by NVMe protocol over PCIe.

Number of in-flight I/O requests at a given time.

Ensuring one workload’s IOPS usage doesn’t affect others.

India-relevant metric comparing cost vs IOPS delivered.

Performance cost introduced by storage protocols.

Explicitly reserved IOPS guaranteed by the provider.

Difference between allocated and actual IOPS usage.

Number of outstanding I/O requests waiting for execution.

Point where increasing queue depth no longer improves IOPS.

IOPS generated by non-sequential access patterns, common in databases and OLTP systems.

Additional reads performed internally by storage systems.

Number of read operations completed per second.

Background I/O triggered to fix inconsistent replicas.

Proportion of read vs write operations in a workload (e.g., 70/30, 90/10), which strongly influences achievable IOPS and latency on a given storage system.

IOPS observed under production workload patterns.

Sequential write IOPS critical for database durability.

Extra IOPS consumed to maintain data replicas.

Delay caused by disk platter rotation in HDDs.

IOPS generated by sequential access patterns, common in backups and streaming workloads.

Service time is the time the storage device spends actively processing an I/O; response time (latency) is service time plus all queueing and protocol overhead.

I/O with small block sizes that stresses IOPS capacity.

IOPS capability of solid-state drives.

IOPS after burst credits and cache are exhausted.

Maximum data transfer rate supported by storage.

Memory layer used to accelerate I/O operations.

Component managing I/O requests between compute and storage media.

Classification of storage based on IOPS and latency.

Enforcement of IOPS fairness across workloads.

IOPS that can be maintained continuously.

Artificial workload used to measure raw IOPS capability.

Worst-case latency affecting user experience under load.

Total data transferred per second, measured in MB/s or GB/s.

Performance drop when IOPS saturation is exceeded.

High write IOPS generated by database commit logs.

Maximum number of volumes attachable to an instance.

Artificial IOPS inflation due to caching.

Extra internal writes generated beyond application writes.

Number of write operations completed per second.

Logging mechanism that heavily influences write IOPS.