A software-based IP storage and virtualization package enables Bell to replicate data cross-country for remote disaster recovery, consolidate storage resources, and reduce storage administration costs.
BY DAVE SIMPSON
Early this year, distributor Bell Microproducts wanted to expand its data processing and storage facilities, while enhancing its disaster-recovery and business resumption program. In addition, Bell Micro wanted to shift the hosting of its data-center operations-including all application servers and data-from its headquarters in San Jose, CA, to a facility in Montgomery, AL. In this scenario, the Montgomery site would become the company's primary data center, with the San Jose facility functioning as a passive hot site holding replicated data from the Montgomery site. Data would be continuously replicated between the two sites over an IP WAN.
"Our primary challenges involved storage networking, data management, replication, business continuity, and procedural integrity between the two locations," says Bob Sturgeon, CIO at Bell Micro. "Our base requirements were absolute redundancy in every aspect of communications, data protection and recovery, and performance."
The project began early this year with the evaluation of a number of solutions from high-end disk array/software vendors. Bell Micro also explored the option of contracting with a third party for backup and disaster recovery. However, the proposal from one of the leading RAID vendors approached $3.5 million, which was unacceptable. Likewise, the option of outsourcing to a third party also proved to be prohibitively expensive.
Bell Microproducts' Montgomery, AL, site includes two Linux servers running IPStor software connected to a pair of RAID arrays in an active-active fail-over configuration.
Eventually, Bell Micro opted for a product that it resells-FalconStor Software's IPStor, a software-based approach to IP-based storage networking and virtualization. The code also enables storage services such as active-active fail-over, snapshot, mirroring, and replication.
According to James Dyches, director of computer operations at Bell Micro, the IPStor approach cost approximately 20% to 30% of the high-end disk array vendor's proposal.
Bell Micro's Montgomery site includes 14 Windows NT 4.0 servers (Bell Micro's Trademark brand), each with two 18GB hard drives for the operating system and applications code (see diagram). All production data resides on two Galaxy 55 RAID arrays from Bell Micro's Rorke Data division. Total capacity is 1.7TB, of which 850GB is currently being used. Each of the 14 NT servers includes two Gigabit Ethernet cards from Intel that support a pre-standard version of the iSCSI protocol.
Two Linux servers running IPStor server code control the two RAID arrays. IPStor client code runs on all of the attached NT servers. The redundant configuration includes full fail-over capability between the two IPStor Linux servers, and each server has two Fibre Channel host bus adapters (HBAs).
The IPStor Linux servers provision virtualized storage to the 14 Windows NT servers, which in turn, host e-mail, Internet, and intranet applications.
All connections between the two IPStor Linux servers and disk arrays are Fibre Channel, while all connections outside the two Linux servers are IP/Ethernet. The servers are connected to a Cisco 6500 Series switch with four Gigabit Ethernet blades. Data is replicated between the Montgomery and San Jose sites over an IP WAN. For increased data protection, the Montgomery site data is backed up to an ADIC DLT tape library.
The Montgomery site went "live" this summer, and Bell Micro has replicated all of its data and applications from San Jose to Montgomery.
According to Bell Micro's Sturgeon, the primary benefits of the IPStor approach versus competing alternatives are the following:
- Price/performance. "We had significant savings on up-front investments, with no sacrifice in performance relative to the more expensive alternatives we considered," Sturgeon reports. On the Fibre Channel part of the configuration, Sturgeon says he's getting a sustained 85MBps to 95MBps throughput rate;
- Lower cost of administration and simplified storage management (via storage consolidation and administration services provided by IPStor); and
- Remote disaster protection and data recovery.
In addition, the IP storage approach eliminates the need for a Fibre Channel-based approach to long-distance storage transfers, which would have added considerable expense due to the cost of routers or gateways needed to translate between Fibre Channel and IP. In addition, at the time, most of those solutions were proprietary.
IPStor also provides a number of features typically associated with storage virtualization engines. For now, however, Bell Micro is using virtualization primarily for RAID storage management. "We use virtualization in that all the data is stored on the SAN-not on the servers-so data is virtualized in the SAN 'data pool,' " Sturgeon explains. (In addition to SANs, IPStor can be used in NAS environments.)
Virtualization also enables administrators to provision volumes to other servers and to add capacity on-the-fly without any downtime. In addition, virtualization facilitates implementing storage services such as mirroring and replication. In the case of Bell Micro, synchronous replication takes place on-site at the Montgomery facility as well as asynchronously between the San Jose and Montgomery facilities.
Coming in INFOSTOR...
- Storage virtualization, part III: SANs
- Moving from SAN to NAS virtualization
- Backup strategies for networked storage
- Continuous replication enhances tape backup
- Bandwidth vs. latency in SAN extensions
- Make way for the serial ATA interface