Dataset · SSD failures · SYSTOR 2016

In an 18-month Berkeley study of 368 disks, 7 data disks failed against 13 enclosure backplane failures; disks were 90% of parts but about 4% of errors

Talagala and Patterson (UC Berkeley, 1999) logged 18 months of a prototype with 368 SCSI data disks. 7 data disks failed (1.9%), against 13 of 46 enclosures, and disks were about 90% of parts but around 4% of errors. Cross-check rows come from Schroeder and Gibson (FAST 2007) and Ford et al. (OSDI 2010).

Download CSV15 rows · 7 columns · CSV

File /data/datasets/disk-error-behavior-1990s-storage-prototype.csv · JSON metadata · All datasets · Human-readable note on hesela.com

Method

The CSV transcribes figures printed in Narayanan et al. (SYSTOR 2016) and two cross-check rows from NetApp (FAST 2020) and Backblaze (2016). It does not read chart values. AFR is the share of devices with failures divided by device years. A failure here is a fail-stop that takes a server down.

Limits

One small prototype built from 1996 parts; says little about modern drives.
Counts are small: 7 disk failures and 1 failure for several other types.
Failures are replacements over 18 months, not annual rates.
The error table and the text disagree slightly on data disk errors and machine counts.
The cross-check rows are other systems with other definitions.

Table

15 rows, the same rows as the CSV. Value low and value high are the printed range ends, or a single printed value in the high column. Units are in the unit column. Sources are the pages opened on 2026-10-11.
Measure	Value low	Value high	Unit	Scope	Note	Citation
SCSI data disks in the prototype	not stated	368	disks	20 PCs, 3.2 TB, 8.4 GB disks, hardware from 1996	Disks were about 90% of the components.	talagala-ucb99-pdf
Length of the failure record	not stated	18	months	Prototype operation	Error logs cover the last 6 months.	talagala-ucb99-pdf
Components replaced in the failure record	not stated	34	components	All component types	Printed as 34 absolute failures, nearly two a month.	talagala-ucb99-pdf
SCSI data disks replaced	not stated	1.9	percent of 368 disks in 18 months	7 of 368 disks	Lowest percentage of any component that failed.	talagala-ucb99-pdf
Disk enclosure backplane failures	not stated	28.3	percent of 46 enclosures in 18 months	13 of 46 enclosures	Backplane integrity failures on the SCSI bus.	talagala-ucb99-pdf
IDE system disks replaced	not stated	25.0	percent of 24 disks in 18 months	6 of 24 disks	Mounted in ordinary PC chassis, not the cooled enclosures.	talagala-ucb99-pdf
Enclosure power supplies replaced	not stated	3.26	percent of 92 supplies in 18 months	3 of 92 supplies	Each enclosure had two supplies, so a failure did not take it offline.	talagala-ucb99-pdf
Error instances in the logs	not stated	688	error instances	6 months of logs, 16 or 17 machines	The paper's text says 16 machines and a table caption says 17; instances are messages grouped within 10 seconds.	talagala-ucb99-pdf
Share of errors that were name-lookup and file-mount errors	not stated	40	percent of error instances (lower bound)	Same logs	Printed as over 40%; caused by dependence on external servers.	talagala-ucb99-pdf
Share of errors that were SCSI timeouts or parity errors	49	87	percent of error instances	49% of all errors; 87% once network errors are removed	Mostly one enclosure failure and loose or faulty cables.	talagala-ucb99-pdf
Share of errors that came from data disks	not stated	4	percent of error instances (about)	Same logs	Printed as around 4% overall, with about 3% recovered errors; the paper's table lists fewer.	talagala-ucb99-pdf
Restarts caused by data disk or IDE disk errors	not stated	0	restarts of 73	16 nodes, 6 months	The largest cause of restarts was external network failures.	talagala-ucb99-pdf
Disks as a share of hardware replacements	20	50	percent of replacements	HPC1 30%, COM2 50%, COM1 nearly 20%	Counts of replacements, not a rate per disk.	schroeder-fast07-pdf
Disk mean time to failure	10	50	years	Tens of Google storage cells	Counts software and hardware causes.	ford-osdi10-pdf
Node mean time to failure	not stated	4.3	months	Same cells	Most node failures are transient.	ford-osdi10-pdf

Columns

measure (string): The quantity as the source names it.
value_low (number): Lower end of a printed range. Empty when the source prints a single value.
value_high (number): Single printed value, or the upper end of a range. For 'over' or 'around' it is the stated figure.
unit (string): Unit of the value columns: disks, months, components, percent, restarts, or years.
scope (string): Population and window the value applies to.
note (string): What the value is not, or the source wording behind it.
citation_id (string): Id of the opened source in the citations list.

License

Small derived table of figures printed in Talagala and Patterson (1999), Schroeder and Gibson (FAST 2007), and Ford et al. (OSDI 2010), with credit. Not a Creative Commons license. The papers remain with their authors or publishers. This file is not a copy of the papers and not the data.