Dataset · Hardware failures · DSN 2017

Hard drives were 82% of 290,000 datacenter hardware failure tickets, and 2% of failed servers produced over 99% of the failures

Wang, Zhang, and Xu, DSN 2017, analysed over 290,000 hardware failure tickets from four years at one large Internet company. Hard drives were 81.84% of failures and 2% of failed servers produced over 99% of failures. Cross-check rows come from Schroeder and Gibson (FAST 2007) and Ford et al. (OSDI 2010).

Download CSV14 rows · 7 columns · CSV

File /data/datasets/datacenter-hardware-failure-tickets.csv · JSON metadata · All datasets · Human-readable note on hesela.com

Method

The CSV transcribes figures printed in Wang, Zhang, and Xu (DSN 2017) and cross-check rows from Schroeder and Gibson (FAST 2007) and Ford et al. (OSDI 2010). It does not read chart values.

Limits

One company's fleet and ticketing rules; the company is not named in the paper.
A failure is a ticket and includes SMART and error warnings, so shares are not failure rates per drive.
The 42.2 day figure is operator response time, not repair time.
The cause of the November 2015 batch is not known.
The cross-check rows are other systems with other definitions.

Table

14 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
Hardware failure tickets analysed	not stated	290000	tickets	One large Internet company, dozens of datacenters, hundreds of thousands of servers	Printed as over 290,000 reports.	wang-dsn17-pdf
Length of the record	not stated	4	years	Same company	Printed as the past four years; the batch analysis covers 1,411 days.	wang-dsn17-pdf
Hard drives as a share of all failures	not stated	81.84	percent of failures	Failure tickets excluding false alarms	Text says about 82%; 10.20% are manually entered miscellaneous tickets.	wang-dsn17-pdf
Memory as a share of all failures	not stated	3.06	percent of failures	Same tickets	Power 1.74%, RAID cards 1.23%.	wang-dsn17-pdf
SSDs as a share of all failures	not stated	0.31	percent of failures	Same tickets	Flash cards 0.67%.	wang-dsn17-pdf
Failures in out-of-warranty hardware that are not handled	not stated	25	percent of failures (lower bound)	Same tickets	Printed as over 1/4 of failures; operators decommission the totally broken servers.	wang-dsn17-pdf
Share of failed servers that produced over 99% of all failures	not stated	2	percent of servers that ever failed	Same tickets	Printed as 2% of servers that ever failed contribute more than 99% of all failures.	wang-dsn17-pdf
Fixed components that never repeat the same failure	not stated	85	percent (lower bound)	Same tickets	Printed as over 85%; about 4.5% of servers that ever failed had repeating failures.	wang-dsn17-pdf
Days with over 500 hard drive failures	not stated	2.48	percent of days	35 of 1,411 days	Batch failures, counted per day.	wang-dsn17-pdf
Servers of one product line reporting hard drive SMART alerts in a single night	not stated	32	percent of that product line's servers	November 2015 case study; 99% detected within about 6 hours	Cause not identified; about 28% of the drives were replaced and over 70% were out of warranty.	wang-dsn17-pdf
Mean time to respond to a failure ticket	not stated	42.2	days	Tickets that led to a repair order	The median is 6.1 days; this is operator response time, not a drive replacement time.	wang-dsn17-pdf
Disks as a share of all hardware replacements	20	50	percent of replacements	Three systems: HPC1 30%, COM2 50%, COM1 nearly 20%	Replacement logs of other organizations, FAST 2007.	schroeder-fast07-pdf
Node failures that are part of a burst of at least 2 nodes	not stated	37	percent of failures	Tens of Google storage cells, 120 second window	Printed as 37%; a different unit from servers with tickets.	ford-osdi10-pdf
Older data blocks that fail a checksum on scrubbing	0.0000001	0.000001	fraction of older blocks	Google file system, background scrubbing	Printed as 1 in 10^6 to 10^7 and concentrated on a small number of disks.	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 'less than' it is the stated bound.
unit (string): Unit of the value columns: tickets, years, percent, days, or fraction of blocks.
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 Wang et al. (DSN 2017), 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.