Return to ILM Table of Contents

  

NETWORK WORLD NEWSLETTER: 09/28/04
_______________________________________________________________

Today's focus:  Grids and optimized capacity

By Mike Karp

Last time, we talked about information lifecycle management and
the reasons why the time is now ripe for some companies to put
ILM into practice.  At the end of that article, I mentioned that
the technology behind storage grids is also likely to play a
significant role both in the enterprise and in enterprise ILM
implementations. Here's why.

For companies that make widespread use of very large IT systems,
the storage grid holds the promise of making real-time
scalability and near-infinite capacity always available.
Theoretically, at least, it can provide this capacity at any
tier of storage a manager might care to define.

The job of grids - and this applies to both storage grids and
compute grids - is to make capacity available when needed.  Note
however that capacity in this case needs to be redefined just a
bit.  With grids, capacity does not equal raw capacity, by which
I mean the sheer amount of storage or computing power a company
can have at its fingertips.  Rather, it refers to what might be
termed "optimized" capacity.

Optimized capacity?   In terms of storage, this means two
things:

First, it means that as additional capacity is brought online,
taken off-line or accessed, nothing is added to the management
load.  Thus, when provisioning or using storage devices, file
systems will not have to be mounted by users and admins will not
have to "twiddle the dials" on their management consoles to
bring additional systems online or to perform other management
tasks.  All decisions regarding rights of access and so forth
have already been determined by existing policies.

The first goal of a storage grid: if a qualified user needs
additional capacity, it is made available with no appreciable
management overhead.

Optimized capacity also means that whenever storage capacity is
added, throughput keeps pace.  The importance of this point
cannot be overemphasized; it does us little good to add new
capacity if, by adding the new storage, we also slow down the
system.

To place this second point in a context that will be familiar to
many of you, consider what happens on a 16-bit wide SCSI bus.
While you are certainly free to populate every address on the
bus with a storage device, throughput tops off when the fifth
device is added.  Thus, even though you may add devices to your
heart's content (up to 15), you can only access five of the
devices at a time at peak efficiency.

With grids, such a situation will be unacceptable:  capacity and
throughput must scale proportionally so that every time
additional storage is added, throughput also increases.

This objective of grids - matching easily managed and scalable
capacity with equally scalable throughput - addresses a storage
goal that we have been looking to achieve for over a decade.
Beginning in the mid-1990s, system designers began taking
compute cycles devoted to I/O processing away from the central
CPU and deploying them out to intelligent controllers (if you
have a sense of history, this is what the I2O initiative was all
about, and to a large degree what TCP/IP offload engines are
concerned with now).

Grids of course must do all this on a much larger scale, and in
an easy-to-manage fashion.  Also, their aim is to make all this
pervasively available, with no single point of failure within
the system.

How this might be accomplished is our subject for next time.
_______________________________________________________________

Mike Karp is senior analyst with Enterprise Management
Associates, focusing on storage, storage management and the
methodology that brings these issues into the marketplace.
Mike can be reached via e-mail
<mailto:mkarp@enterprisemanagement.com>.
_______________________________________________________________

Copyright Network World, Inc., 2004
 

Return to ILM Table of Contents