randwrite

Randwrite is a workload pattern used in storage benchmarking to generate random write operations across a storage device. Unlike sequential writes, randwrite issues write I/O to arbitrary logical block addresses, simulating irregular access patterns typical of databases, virtual machines, and many file systems.

Characteristics include the randomness of addresses, often uniform across the device, though some tests use skewed distributions (Zipf) to model hotspots. Block size commonly ranges from 4 KB to 128 KB, and the depth of outstanding I/O (queue depth) affects observed IOPS and latency. Random writes can cause higher write amplification and more wear on flash-based devices due to wear leveling and garbage collection.

In benchmarking tools, randwrite is a predefined I/O pattern. For example, in the fio tool, rw=randwrite specifies a random-write workload; tests may be configured with additional parameters such as iodepth, bs, and size to tune intensity. Some benchmarks combine randwrite with random read (rw=randrw) or with a fixed percentage of reads (rwmixread) to model mixed workloads.

Interpreting results requires awareness of caching and virtualization layers. Operating system caches, file system journaling, device caches, and hardware RAID controllers can mask or exaggerate raw device performance, so measurements should be repeated with caches disabled or explicitly controlled.

Randwrite is a standard component of storage benchmarking and capacity planning, providing insight into I/O performance under random-write pressure.