Fabric-based intelligence gets vendor nod

By Heidi Biggar

News last month that Brocade had signed up seven more independent software vendors (ISVs) as Fabric Application Development Partners signals a growing trend in the industry to move storage applications off hosts and disk arrays and into the fabric.

In early March, start-up switch vendor MaXXan Systems began production shipments of an intelligent switch that is designed to host storage management applications, and Cisco this week is expected to make clear its product/partnership strategy for this market.

Cisco has long stated its plan to work with leading ISVs to migrate their storage management applications to its MDS 9000 family of switches (see "IBM, HP first to resell Cisco storage switches," InfoStor, February 2003, p. 8).

"Many types of storage applications are ideal for hosting on the network or fabric, and Cisco is exploring a variety of options," says Ed Chapman, senior director of product management for Cisco's storage technology group.

Brocade, meanwhile, has set its sight on applications that require the movement and manipulation of a lot of data across heterogeneous storage environments. Specifically, Brocade is looking at migrating data replication, volume management, and data management applications into the fabric (or network) and onto SilkWorm fabric switches (see figure).

Moving applications (such as replication, SRM, and virtualization) off hosts and into the fabric onto "intelligent" switches has the promise of minimizing administration "touch points" and improving SAN scalability in mixed server/array environments.
Click here to enlarge image


Brocade acquired network switches from Rhapsody Networks earlier this year. From Rhapsody, Brocade inherited multi-protocol switching technology, advanced ASIC technology, and the software/APIs for ISVs to port applications to its network-based switch platform.

In conjunction with its initial partnership announcement, Brocade made its software developers kit (SDK) available to ISVs. Its objective: to accelerate the time to market by making APIs available to vendors looking to migrate storage management applications into the network.

Initial Brocade Fabric Application Development Partners represent a cross section of ISVs (see list). Some of these companies demonstrated their fabric-based applications at Brocade's annual user conference earlier this month. And early partner Hewlett-Packard continues to work with Brocade to bring HP's VersaStor virtualization technology into the fabric.

Earlier this year, HP detailed its virtualization strategy, which ultimately hinges on being able to run its VersaStor code in a fabric switch alongside an HP Continuous Access Storage Appliance (see "HP to merge virtualization platforms," InfoStor, March 2003, p. 8).

Sources close to Brocade say that the company will soon announce details of a partnership with Veritas to embed its volume management software in Brocade fabric switches.

While Veritas officials would not comment on the exact nature of the relationship, they said the two vendors were working together closely and would make a product announcement in the upcoming months.

Proponents of moving applications into the fabric cite improved management and scalability as primary motivators. Traditional host- and/or disk-based storage management techniques, in contrast, can be difficult to manage (requiring multiple management tools and many pieces of software) and to scale, and can be vendor-specific.

By moving storage management into the fabric, users can simplify storage management, improve scalability, and ensure investment protection, claims Tom Buiocchi, vice president of product management at Brocade.

"It improves storage management by reducing the number of administration 'touch points,' " says Buiocchi. "The virtualization, replication, or provisioning software spans all servers without actually putting any [additional software] on them...and it allows you to do so in a heterogeneous environment."

Take replication software. Today, it is most often installed on the storage array by the device vendor. "If you have multiple storage [types], then it is almost impossible for you to replicate among them," says Buiocchi. Fabric-based replication software, in contrast, allows administrators to replicate data among heterogeneous storage devices with minimal management hassles, he says.

Cisco's Chapman says that while there are good reasons for running applications from the host or the array, network-based models (a.k.a. fabric-based applications) have the advantage of providing many of the same benefits while eliminating many of the disadvantages of traditional methods.

"For example, running applications from the network will allow users to scale across both heterogeneous server and array environments and, as the central point in the SAN, it has management advantages over both host- and array-based applications," Chapman says.

However, vendors need to educate the end-user community about the pros—and potential cons—of the network-based model.

Click here to enlarge image


Explains Jacob Farmer, chief technology officer of Cambridge Computer, a systems integrator in Cambridge, MA: "I have never heard a client specifically ask for application logic in the fabric, but I believe most users will appreciate the value of system-wide logic being administered centrally."

The only downside to running applications from a switch, says Farmer, is the possibility of switches becoming less interoperable with one another, which could slow the adoption of next-generation SAN technology.

And there's the issue of availability, says Marc Staimer, founder of Dragon Slayer Consulting, in Beaverton, OR. "How do you ensure high availability of the application with a reasonable cost? What happens to the LUN map if the application blade or switch goes down?"

"Personally, I favor a centralized abstraction layer approach, i.e., virtualization software that resides either in-band between the fabric and storage devices or out-of-band with an agent residing on the host. The agent would handle block address redirection and would be fabric-agnostic," explains Farmer, "but I'm keeping an open mind."

"Those types of virtualization appliances [e.g., DataCore's or FalconStor's] aren't optimized for storage performance," argues Brocade's Buiocchi. "[They're basically] general-purpose servers [running virtualization code.]"

The debate over where best to run applications will likely continue for months—and even after fabric switches capable of running applications are generally available.

This article was originally published on April 01, 2003