Dataset · SSD failures · SYSTOR 2016

In Alibaba's SSD fleet, 12.9% of failures were in the same node and 18.3% in the same rack within 30 minutes of another failure

Han et al., FAST 2021, studied nearly one million Alibaba SSDs over two years. 12.9% of failures were intra-node and 18.3% intra-rack within 30 minutes, and the strongest SMART attribute had a rank correlation of only 0.23 with them. Cross-check rows come from NetApp (FAST 2020) and Google (OSDI 2010).

Download CSV16 rows · 7 columns · CSV

File /data/datasets/ssd-correlated-failures-nodes-racks.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 operator's fleet and ticketing rules.
The 12.9% and 18.3% shares depend on the 30 minute threshold.
88.6% of nodes with two or more SSDs hold one drive model, so some clustering may be a model effect.
The redundancy results come from a trace-driven simulation.
The cross-check rows are other fleets with other definitions.

Table

16 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
SSDs studied	not stated	1000000	devices	11 drive models from 3 vendors, Alibaba data centers	Printed as nearly one million.	han-fast21-pdf
Nodes holding the SSDs	not stated	200000	nodes	Same fleet	Printed as 200 K nodes; 88.6% of nodes with at least two SSDs hold one drive model.	han-fast21-pdf
Racks holding the SSDs	not stated	30000	racks	Same fleet	Printed as 30 K racks.	han-fast21-pdf
Length of the record	not stated	2	years	January 2018 to December 2019	Daily SMART logs, trouble tickets, locations, applications.	han-fast21-pdf
Failed SSDs (trouble tickets)	not stated	19000	drives	Same fleet	Printed as about 19 K; whole-drive and partial-drive failures, each checked by an administrator.	han-fast21-pdf
Annual failure rate of all SSDs	not stated	1.16	percent per year	Same fleet	Computed from the tickets and drive-days.	han-fast21-pdf
Failures that are intra-node failures	not stated	12.9	percent of failures	Failures in one node within 30 minutes of each other	Default 30 minute threshold.	han-fast21-pdf
Failures that are intra-rack failures	not stated	18.3	percent of failures	Failures in one rack within 30 minutes of each other	Default 30 minute threshold.	han-fast21-pdf
Chance of one more failure in an intra-node failure group	26.3	64.3	percent	Group size 2 to 11 failures	If failures were independent this would be close to the annual failure rate of 1.16%.	han-fast21-pdf
Intra-node failures with an interval of one minute or less	not stated	10.0	percent of intra-node failures	Same fleet	The one month threshold gives 29.2%.	han-fast21-pdf
Intra-rack failures with an interval of one month or less	not stated	63.0	percent of intra-rack failures	Same fleet	The one minute threshold gives 14.4%.	han-fast21-pdf
Highest rank correlation of a SMART attribute with correlated failures	not stated	0.23	Spearman correlation coefficient	SMART attribute S187 (reported uncorrectable errors)	Same value for intra-node and intra-rack failures; the paper calls this limited.	han-fast21-pdf
Intra-node failure share by application	2.1	33.6	percent of an application's SSD failures	Eight largest applications	Intra-rack share ranges from 2.8% to 40.5%.	han-fast21-pdf
Chance of a drive replacement in a random week	not stated	0.0504	percent	NetApp RAID groups, about 1.4 million SSDs	Baseline for the next row.	maneas-fast20-pdf
Chance of a replacement within a week of a previous one in the same RAID group	not stated	9.39	percent	Same fleet	More than 180 times the baseline; 52% of consecutive replacements fall within a week.	maneas-fast20-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%; node unavailability, not SSD failures.	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 'nearly' or 'about' it is the stated figure.
unit (string): Unit of the value columns: devices, nodes, racks, years, percent, or correlation coefficient.
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 Han et al. (FAST 2021), Maneas et al. (FAST 2020), and Ford et al. (OSDI 2010), with credit. Not a Creative Commons license. The USENIX papers remain with their authors or publishers. This file is not a copy of the papers and not the data.