35
WWW.7X24EXCHANGE.ORGAPPENDIX A
From 2010 Uptime Institute
6
Tier I Assumptions:
Two 12-hour scheduled outages for maintenance
1.2 failures per year, 4 hours per failure
Tier 1 Availability:
2x12 + 1.2*4 = 28.8 hours down per year => (8760-
28.8)/8760 = 99.67 % availability.
Tier 2 Assumptions:
One 18-hour scheduled outage per year
1 failure per year, 4 hours per failure
Tier 2 Availability:
18 + 4 = 22 hours down per year => (8760-22)/8760
= 99.75% availability
Tier 3 Assumptions:
No planned outages
1 failure per 2.5 years, 4 hours per failure
Tier 3 Availability:
4 / 2.5 = 1.6 hours down per year => (8760-1.6)/8760
= 99.98% availability.
Tier 4 Assumptions:
No planned outages
1 failure per 5 years, 4 hours per failure
Tier 4 Availability:
4 / 5 = 0.8 hours per year downtime => (8760-
0.8)/8760 = 99.99% availability
APPENDIX B: RELIABILITY AND
AVAILABILITY
In some industries, losses due to facility downtime
are proportional to downtime:
Lost $ = Rate * Downtime
In data centers and many other critical facilities, there
are substantial losses incurred during the first instant of
downtime. For example, A banking facility might
require 16 hours to restore normal operations after a
major data center outage disrupts ATM operations.
The losses in terms of business, customer satisfaction,
regulatory effects, etc., all happen in the first instant;
even if power is restored in 1 second, it takes 16 hours
to restore the full ATM network.
In these facilities, losses due to facility downtime
include an initial loss term, represented as D:
Lost $ = D + Rate * Time
The amount of loss probably increases with duration of
the outage
7
, but the penalty D is always incurred.
If a planner wants to estimate the total losses for a
facility over its operating lifetime, and the facility
does not have large initial losses, the calculation is:
Total Loss =
∑
(Rate * downtime) = Rate *
∑
downtime
This is true because Rate is constant. The symbol
∑
downtime means “the sum of all downtime.”
If this planner has a valid figure for availability of
the system, they can calculate the expected
downtime over a given interval:
∑
downtime = (1 – Availability) * mission_time
If availability is 99%, and mission time is 10 years, then
∑
downtime = (1 – 99%) * 10 * 8760 hours = 876 hours.
The calculation is different if the facility suffers a
significant loss for each event, regardless of its
duration.
Total Loss =
∑
(D + Rate * downtime) = n*D +
Rate*
∑
downtime
Where n is the number of outages expected over the
mission time.
Availability cannot be used to predict
the number of events. Reliability must be used to
determine the frequency of outages.
1 In some facilities it is acceptable to interrupt cooling for short periods.
2 “Reliability of Data Centers by Tier Classification”, Arno, R., Friedl, A., Gross, P., Schuerger, R.J., Industry Applications, IEEE Transactions on
(Volume:48 , Issue: 2 ), March-April 2012, pp. 777 - 783
3 “Reliability of Example Mechanical Systems for Data Center Cooling Selected by Tier Classification”, Arno, R., Githu, G., Gross, P., Schuerger, R.,
Wilson, S., Industry Applications Society Annual Meeting (IAS), 2010 IEEE
4 See Appendix A
5 “IEEE Recommended Practice for Determining the Reliability of 7x24 Continuous Power Systems in Industrial and Commercial Facilities”, IEEE
Std 3006.7-2013
6 Insert citation
7 The rate of loss need not be linear with time, for example food or drug spoilage might not occur for some time after an outage begins. The analysis of
availability versus reliability still applies.
Steve Fairfax is the President at MTechnology, Inc. He can be reached at
[email protected]