The Uptime Institute
Monday, October 21, 2002
Jonathan G. Katz, Secretary
Securities and Exchange Commission
450 5th Street, NW
Washington, DC 20549-0609
RE: File No S7-32-02
The Uptime Institute maintains the nation's largest database of how
data centers physically fail. Started in 1994, we have tracked
statistics on 3,000,000 square feet of electrically-active raised floor
for our membership of mostly Fortune 100 companies (see 2002 membership
roster attached). We estimate our members represent 60% or more (based
on assets) of the total mission-critical raised-floor space of the
nation's financial institutions.
Briefly, this is our data:
- In eight years, none of our 48 members has experienced physical
failures that would require activation of their backup site. I think
this is for two reasons.
- Members are in Tier 3 (1985) and Tier 4 (1994) facilities (our tier
ranking system is defined on our website at
http://www.upsite.com/TUIpages/whitepapers/tuitiers.html)
- Members are very serious in the staffing and maintenance resources
they devote to assuring equipment and facility uptime
- Over the past eight years, the collective facility failure rate of
our members has dropped from an average of once every eight months to
once every three years. We have every reason to believe that the failure
rate for non-members has remained at once every six months and that
those companies doing processing in commercial collocation or web
hosting facilities will be subject to an increasing frequency and
duration of outages (Many of the collocation sites built in the last 36
months are Tier 1, which means they cannot perform necessary maintenance
without a processing interruption. The age of these sites is nearing the
sharp rise in the bathtub curve failure rate because electrical
maintenance has not been performed).
- There is no perfect place to build a data center. Every location
within the continental US has an element of natural disaster risk
(earthquake, snow, ice, tornado, hurricane, wind, flood plain, etc., as
shown in a Natural Risk Locations Map which is available through The
Uptime Institute). In addition, there are man made risks of being near
an interstate highway, railroad, or chemical plant. There also are risks
with where the data center is located within the building. The classic
is being located under the cafeteria, which exposes the date center to
water leaks and to evacuations due to cooking fires on the floor above
(these are further explained as part of the risk map).
- 95% of disaster recovery hot site activations are not as the result
of an act of God, but are the result of management decisions to take
risk (often these decisions are made at a junior level without senior
management understanding the full business consequences), either by
building in an inadequate location, designing to Tier 1 (1965) or Tier 2
(1975) standards, or not maintaining electrical equipment.
It is from this perspective that we offer the following suggestions:
- Specify the minimum regional separation between primary and backup
sites. Unfortunately, fiber distance limitations are still too
restrictive to deal with a regional event like the ice storm several
years ago that affected most of Maine and parts of New Hampshire and
Vermont. From a facility perspective we suggest the primary and
secondary sites be served by two different parts of the electric
transmission grid, being served by separate water grids (because large
data centers require large amounts of water for cooling evaporation),
being in different regions from a weather perspective, diversity for
data communications, and being served by a different transportation
network. We have seen repeated problems of depending upon cell phone for
emergency communications because of insufficient provider capacity when
a regional event occurs.
- The fault tolerance and maintainability of the primary and backup
site infrastructure needs to be specified. We would suggest at least
Tier 3.
- The competence of the facility staff running the infrastructure
needs to be specified to especially assure the backup site is being
maintained and will really work in an emergency. The last publicly
reported test of engine generators actually working in an emergency
occurred in the early nineties when the South Street Substation outage
left Wall Street without power for several weeks. Twenty-five companies
declared disasters because their engine generator equipment either
failed immediately, in the first hour, or in the first 24-hours.
Unfortunately, the painful lessons learned about the need for testing
performance under real load regularly as opposed to just exercising
whether the engine will start, have mostly been forgotten in the
facility department cost reductions of the last ten years.
Respectfully yours,
Kenneth G. Brill
Executive Director
The Uptime Institute
1347 Tano Ridge Road
Santa Fe, NM 87506
Tel: (505) 986-3900
Fax: (505) 982-8484
Members of the Site Uptime® Network
(2002 Roster)
A.G. Edwards
Alltel
American Express
BellSouth
BP North America
Boeing
Capital Group
Caterpillar
ChevronTexaco
Conagra Foods
Confidential (2)
Consonus
Deere & Company
Defense Information
Systems Agency
Depository Trust
DST Systems
E*Trade
Exodus Communications
Fidelity Investments
Hewitt Associates
Hewlett-Packard
Household International
Inovant (Visa)
(i)Structure
Johnson & Johnson
JP Morgan Chase
Lexis-Nexis
Lowe's
MasterCard
Microsoft
Nationwide
Northwest Airlines
Orange
Philadelphia Stock
Exchange
Procter & Gamble
Salomon Smith Barney
SIAC
Social Security Admin
Sprint
Sun Microsystems
SunGard Recovery Systems
Target
United Airlines
United Parcel Service
USAA
Verizon Communications
Wachovia Bank
|