Hedge fund virtualizes 200 TB with Incipient

A major Wall Street firm based in Stamford, Conn., is preparing to put two thirds of its data behind Incipient's intelligent virtualization switch.

As storage virtualization moves, slowly but steadily, from hype into real-world deployments, one of the newest virtualization startups, Incipient Inc., has a new customer to report: a hedge fund based in Stamford, Conn., that is getting ready to take the plunge with the Incipient Networked Storage Platform (iNSP).

The fund, which director of systems architecture Bart McDonough said he is not at liberty to identify, has some 300 terabytes (TB) of data spread across three data centers -- two a few hundred yards apart at Connecticut headquarters and at a disaster recovery hot site 70 miles away. The headquarters data centers each have the bulk of production data spread evenly between them, with 100 TB at each, and the disaster recovery hot site has another 100 TB of data.

The fund will be using the Incipient software to virtualize all of its production data at headquarters, as well as blend the two data centers together under a central management console at the end of this month, McDonough said.

More storage virtualization info
Next-generation data protection, iSCSI and storage virtualization on the rise in '07

Colonial Williamsburg virtualizes with Acopia

Storage virtualization moves from hype to reality

Resellers come around to storage virtualization

The main data center is connected to a Cisco Systems Inc. MDS 9513 director at the "core," with "edge" MDS 9216 switches making up the rest of the data center's fabric, as well as the "core" of the second data center. The Incipient software currently can only be run on a Storage Services Model (SSM) blade within the Cisco MDS chassis.

McDonough said the hedge fund will put one instance of Incipient's software on each "core" switch at the two data centers, but will manage both from the primary site through Incipient's NSP Administrator. All data at the main campus' data centers will eventually run through Incipient.

Behind the Incipient system there will be eight Hewlett-Packard Co. (HP) Enterprise Virtual Arrays (EVA). "We love the virtualization the EVAs give us," McDonough said, "but they only virtualize at the disk-drive level." Meanwhile, he said, the company wants to make it so that the loss of an entire EVA is "a nonevent." Incipient, over the other virtualization and replication products we looked at, will make that happen."

The goal once iNSP is fully up and running will be to make it so that all the data can be "failed over" automatically to another EVA in the event of an outage and can also be transferred manually if one of the arrays needs maintenance.

"For example, this weekend, we have to add six shelves of disk to one of the EVAs," McDonough said. "There will be downtime associated with that. Our hope is that once the Incipient is in place, we won't have to deal with that issue."

Another benefit from the use of the Incipient product will come from melding the hedge fund's EVA storage with the 1 TB RAM storage area network (SAN) by Texas Memory Systems, a solid-state disk cache used for processing the most performance-intensive data in the environment.

With the Incipient software in place, McDonough said, he plans to mirror the database data between the RAM SAN and an EVA, and then split the writes and reads between the two to optimize performance, as well as reliability. The RAM SAN handles reads with an average latency of 70 microseconds, he said, compared to 200 milliseconds with the EVA.

"So if all goes well, the RAM SAN will boost database read performance better than if it's handling all the database data," McDonough said. "And if the RAM SAN blows up, I'll have this highly intelligent, highly available pool of arrays sitting right next to it."

Proceeding with caution

McDonough said that at this point he is gung ho about moving forward with the virtualization plan, but that most of last year was taken up with very careful testing of Incipient's product, as well as evaluation of its competitors.

McDonough admitted he didn't do the kinds of hands-on testing with other virtualization products such, as IBM's SVC or EMC Corp.'s Invista, but that he did do extensive research and saw product demonstations from both companies. The hedge fund also evaluated replication software from Double-Take Software Inc.'s host-based products and replication software from Sanovi Technologies Corp.

With the SVC, McDonough said, he was put off by its architecture even from a conceptual point of view. "I don't want all the storage traffic in my environment running through some Linux box," he said. He evaluated Invista "briefly," he said, but at the time felt it still had some bugs to work out. As for the replication products, McDonough said he began to see replication as complementary to virtualization rather than a competitor for the same place in the data center and as an HP customer said he is satisfied with HP's Continuous Access (CA).

"We will continue to use CA for intersite replication," he said, since the remote site will not be brought under the Incipient product due to the 70-mile distance between data centers. Incipient can also do replication, he said, but the company will shy away from using it at first as it will with Incipient's provisioning feature.

"Incipient can also carve out LUNs for you on the back end," McDonough said. "But we'll be keeping that a separate, manual process for the foreseeable future, just in case."

Before signing the purchase order for the NSP, the company tested it intensively against test instances of its databases for six weeks, McDonough said, and spent more than six months in discussion with Incipient last year.

"We had a 40-page testing plan," McDonough said. He added that the company is very performance sensitive and used IOMeter throughout the testing of Incipient to gauge what kind of latency comes with the product.

"Consistently, we saw about a one millisecond difference," he said. "That's something we can live with."

Finally, once the installation begins, the process of rolling it out will probably consume between three and six months. "We're going to take our sweet time getting it integrated into our environment," he said. "We're going to present and test and double check one LUN at a time."

HP backs up Incipient

Given the risks, then, how does McDonough justify going with the product in the first place?

Services engineers at HP are in charge of troubleshooting the hedge fund's storage environment, and they have agreed in writing to continue to support the data centers once the Incipient was in place. HP already supports the Cisco environment as well, according to McDonough and will be responsible for calling in Incipient engineers should they be needed.

"That went a long way toward convincing my upper management," McDonough said.

Finally, he added, his firm is less conservative than most when it comes to technology. "We make it a point to invest in technology," he said. "We were doing electronic trading before most people were doing that, too."


Dig Deeper on Storage virtualization

Start the conversation

Send me notifications when other members comment.

Please create a username to comment.