Limitations
Disclaimer: What This Benchmark Does Not Deliver
Raw Block I/O
This benchmark measures raw block devices with direct I/OI/O: Input/OutputThe transfer of data between a computer and the outside world, such as reading and writing disks or sending and receiving network packets. A workload limited by its speed rather than by the CPU is called I/O-bound.WikipediaGlossary (Input/Output). It
does not measure filesystem performance, page-cache effects, or how a given
application or database uses the disk. Results show what the disk can do under
synthetic fiofioThe Flexible I/O Tester, an open-source storage benchmark and I/O workload generator. Each job describes an access pattern, such as random or sequential reads and writes, the block size, the queue depth, and the I/O engine, and fio reports the resulting IOPS, bandwidth, and latency.fio docsGlossary load, not what an application will see through a filesystem.
A Single Score
The output has one record per test and does not combine them into a single number. Spare Cores Navigator will publish bandwidth and IOPSIOPS: Input/Output Operations Per SecondThe number of read and write operations a storage device completes per second. It is the main measure of random-access performance, and depends on the block size, the queue depth and the mix of reads and writes.WikipediaGlossary (Input/Output Operations Per Second) at different block sizes, per disk and for each group of identical disks. The rest of the output, including any headline metric, is not decided yet.
Other fio I/O Engines
All tests use the libaio I/O engine. The image includes liburing, but this
benchmark does not test the io_uring engine.
Scope Decisions
Disk Scope
This benchmark focuses on a server type's bundled storage: the disk or disks provisioned with the server type by default. These can be a locally attached SSDSSD: Solid-State DriveA storage device that keeps data in flash memory instead of on spinning disks, so it reads and writes much faster than an HDD, especially at random locations. A controller with its own firmware maps logical blocks to flash pages, and many SSDs write incoming data to a fast cache first, then move it to slower flash. Cloud servers use SSDs as local disks and behind network block storage.WikipediaGlossary or HDDHDD: Hard Disk DriveA storage device that reads and writes data with magnetic heads on spinning platters. It offers cheap capacity, but each random access waits for a head to move and a platter to turn, so it completes far fewer random operations per second than an SSD.WikipediaGlossary (Hard Disk Drive), or network-attached storage. Network storage attached on demand is out of scope, because its performance is not a property of the server type.
Results reflect the bundled storage as provisioned with the server type, regardless of the underlying medium. A result may therefore be bounded by a constraint outside the drive itself, such as a provider-imposed IOPS or bandwidth limit on network-attached storage, or the mechanical limits of an HDD. This benchmark does not attribute results to a specific cause: it reports performance as measured, and the Navigator publishes those measurements.
The image does not detect which disks are bundled. It tests every real block device that passes discovery, including any volume attached on demand, and overwrites it if it can.
Read-Only Disks
A disk that is in use (the boot disk, for example) is only read, never written. Its read tests run over the whole disk, on whatever data it already holds, without a fill, and it skips every write and mixed test. See Disk Safety for how this benchmark decides.
Design Constraints
Partial Fill
The sequential fill runs for at most 300 seconds in total
(SC_STORAGE_FILL_BUDGET_S), split between groups. A disk that takes longer to
write is only partly filled. On the first production runs, the fill wrote the
following share of each NVMeNVMe: Non-Volatile Memory ExpressThe interface specification for accessing non-volatile storage, most often SSDs, over PCI Express. It is designed for the parallelism of flash memory, with many deep command queues, and NVMe over Fabrics extends it to network transports.NVM ExpressGlossary (Non-Volatile Memory Express) drive:
- GCPGCP: Google Cloud PlatformThe cloud computing platform of Google. It offers virtual machines (Compute Engine), object storage, managed databases, Kubernetes and machine learning services, among others.Google CloudGlossary (Google Cloud Platform)
c4a-standard-16-lssd: 26% of each of four drives - OVHcloudOVHcloudA French cloud provider, founded in 1999, with data centers in Europe, North America and Asia-Pacific. It offers dedicated (bare-metal) servers, public cloud instances based on OpenStack, and private cloud.OVHcloudGlossary
i1-90: 33% of each of two drives
A disk that is not fully written still has spare area, flash that holds no
data yet, which inflates random-write results. The fill record's
filled_fraction shows how much of each disk the fill wrote. Setting
SC_STORAGE_WS_TARGET_BYTES limits the fill, and all later tests, to a smaller
region of each disk.
Destructive Runs
This benchmark overwrites every disk it may write to, and does not
remount the filesystems it unmounts. Run it only on ephemeral servers, never on
production or similar systems. The SC_STORAGE_UMOUNT=0 setting leaves mounted
disks alone, but disks that are not in use are always overwritten. See
Disk Safety.
Time Budget
This benchmark targets an estimated 2400 seconds of tests. When the estimate exceeds that, it shortens tests and then drops low-priority ones to fit. See Run Duration for the order.