Datasheet

overvieW of availabilitY MechanisMs
|
5
synchronously mirrored storage arrays provide even higher performance as well as resiliency. In
this case the entire storage cabinet, including low-level controllers, power supplies, and hard
drives, are all duplicated, and the two arrays mirror each other, usually in a synchronous manner
where both arrays receive the data at exactly the same time. The production servers are not aware
of the duplicated arrays and can therefore equally access either autonomous storage solution.
So far, this sounds pretty good. But there are still some challenges, though far fewer chal-
lenges than there used to be. Back then, disk arrays were inordinately more expensive than
local storage. Add to that the cost and complexity of storage area network (SAN) fabrics and the
proprietary adapters for the server(s), and the entire solution became cost-prohibitive for most
environments. In 2002, Gartner’s “Study of IT Trends” suggested that only 0.4 percent of all IT
environments could afford the purchase price of synchronously mirrored storage arrays. For
the other 99.6 percent, the cost of the solution was higher than the cost of the problem (potential
data loss). Of course, that study is now eight years old. The cost of synchronously mirrored stor-
age has gone down and the dependence on data has gone up, so it is likely that 0.4 percent is
now too low of a number, but it is still a slim minority of IT environments. We will discuss this
statistic, including how to calculate its applicability to you, as well as many metrics and decision
points in Chapter 2.
While you could argue that the parity bits in a RAID configuration are about preserving the
integrity of data, the bigger picture says that mirroring/striping technologies are fundamentally
about protecting against a component-level failure — namely the hard drive. The big picture is
about ensuring that the storage layer continuously provides its bits to the server, OS, and appli-
cation. At the disk layer, it is always one logical copy of the blocks — regardless of how it is
stored on the various spindles.
This concept gets a little less clear when we look at asynchronous replication, where the data
doesn’t always exactly match. But in principle, disk (hardware or array)-based “data protection”
is about “availability.”
de c I s I o n Qu e s t I o n : Is It re A l l y MI s s I o n cr I t I c A l ?
The first decision point, when looking at what kinds of data protection and availability to use, is
whether or not the particular platform you are considering protecting is mission critical (we’re
ignoring cost factors until Chapter 2). But in principle, if you absolutely cannot afford to lose
even a single mail message, database transaction, or other granular item of data, then a particu-
lar server or platform really is mission critical and you’ll want to first look at synchronous stor-
age as part of your solution along with a complementary availability technology for the other
layers of the server (for example, application or OS).
Note that crossing the line between synchronous and asynchronous should be looked at
objectively on a per-server or per-platform basis — instead of just presuming that everything
needs the same level of protection.
Even for key workloads, the idea that they are mission critical and therefore immediately
require synchronously mirrored disks and other extraordinary measures may not be univer-
sally justified. Consider two of the most common application workloads — SQL Server and
Microsoft Exchange.
In a large corporation with multiple Exchange Servers, you might find that the Exchange
•u
Server and/or the storage group that services email for the shipping department might be
considered noncritical. As such, it may be relegated two nightly or weekly tape backups
only. In that same company, the executive management team might require that their email
572146c01.indd 5 6/23/10 5:42:19 PM