Let's get straight to the point: We think the term “storage networking” will mean something entirely different in just a few
short years. Let us explain.
The Storage Networking Industry Association today defines "storage networking" as the practice of creating, installing, administering, or using networks whose primary purpose is the transfer of data between
computer systems and storage elements, and between storage elements.
This definition is good for now, but the idea of a single network for storage will slowly fade. No, we are not saying Fibre Channel is going away; we're saying the architecture of a data center is changing. And as that change occurs, so will storage’s
relationship to servers.
The underlining premise of this argument is the idea of discrete compute elements that can be assigned, provisioned, and configured
for applications on an as-needed basis. This is a giant departure from today's storage architecture, which features expensive
servers connected to dense disk-based storage systems.
The datacenter of tomorrow, according to IBM and Sun Microsystems, has a common pool of storage, compute, and network resources
that can be used and reused as application needs change. To illustrate, if a database application required more processing
or more storage, an administrator today would have to physically add another server and more storage capacity. With a shared resource architecture, an admin could simply reassign an existing server and allocate more storage from a common pool of
capacity.
Some of this is already happening today. The most obvious trend is the move to low-cost Linux-based blade servers. These servers
are cheap and can be easily strung together to replicate a high-end server, eliminating the need to purchase expensive multiprocessor
servers. In effect, with blade servers, we've already begun to disaggregate the different elements.
The next step, which is currently underway at several startups, is the creation of a platform that interconnects these shared
resources. Using 10Gbps Infiniband interconnects, these platforms provide high bandwidth and low latency, and are being positioned to connect servers to servers
and/or servers to storage, using FC.
Applications to take advantage of this architecture are already being built. For example, with Oracle9i Real Application Clusters
(RAC), system administrators now have the ability to run a single database on a group of servers clustered together.
It is feasible that by year’s end IBM and Sun will introduce platforms that also connect discrete I/O, CPU, and storage resources
to each other via a new platform. And when this happens, “storage networking” will finally mean storage is a true part of
a network - and not something you link as an after thought.
E-mail
Printer Friendly
Reprints
(InfoWorld) - In the litigious world we live in, deploying a unified communications platform in your enterprise could...
