Tim Hesson, director of storage management at Kindred, said this week that two new storage area networks (SANs) using four Cisco Systems Inc. MDS 9509 Multilayer Director Switches (MDS) have replaced the old one, which was made up of McData Corp. directors and switches. The new SANs provide 800 ports.
The consolidation freed up 100 switch ports that had been used in the old network just to link all the switches together.
Hesson said the new SAN design lowered the company's Fibre Channel per-port costs by $200 each and gives every server two data paths -- so if one director goes down, it automatically fails over to a secondary network.
"I'm protected from a blade failure. I'm protected form a [switch] chassis failure. I'm protected from a cable failure, and I'm protected from a [host bus adapter] failure. I have end-to-end redundancy," Hesson said. "I think right now we're getting really close to four 9s [storage reliability]."
In 2004, with more than 50,000 employees in 350 facilities, Kindred found its data needs quickly outpacing its storage networking capabilities. The company also needed to set up a more proactive disaster recovery plan, Hesson said.
The Fortune 500 company's data center is located at its Louisville, Ky. headquarters, with remote servers on a WAN deployed across the country.
For its disaster recovery plan, the company ships backup tapes to a Sungard facility. But server rebuilds only occur once an emergency declaration has been issued by Kindred, which would slow its ability to get back online.
Kindred wanted to be able to replicate data over the WAN to Sungard where data could be stored on disk and made available at a moment's notice.
With that goal in mind, Hesson's storage team this year plans to complete a dual network remote disaster recovery SAN based on two modular Cisco 9216i Multilayer Fabric Switches.
With Gigabit Ethernet capability in each storage network, Kindred will be able to connect the main SAN with the disaster recovery SAN by using the Fibre Channel over IP (FCIP) protocol to transmit data from its main data center in Louisville to the Sungard disaster recovery site in Philadelphia.
Kindred also wants to use the MDS switch's iSCSI capabilities for Ethernet connectivity to eventually consolidate backup on some of the company's 1,500 Wintel servers.
"We've always tried to take a look around the corner," Hesson said. "If we are going to spend this money, is there any possibility to do something different or to better improve the infrastructure tomorrow, even if we can't leverage it today?"
Kindred is an all-EMC storage shop, and over the past year installed about 225TB of capacity on multiple high-end Symmetrix DMX arrays, a large number of midrange Clariion arrays and one Celerra network-attached storage (NAS) array. The company plans to install an EMC Centera content-addressed storage array this month.
Hesson said that what sparked his interest in the Cisco director is its modular design, which allows users to expand the number of ports and even networking protocols -- such as iSCSI and FCIP -- by switching out blades from a chassis.
Kindred had been using two director-classs switches and 26 edge switches from McData Corp. for about four years. But the architecture could no longer grow, limiting the company to 1Gbit/sec Fibre Channel networking speeds -- even though it had replaced the server's host bus adapters with new 2Gbit/sec technology.
The large numbers of "edge" switches -- used to tie application servers to the centrally located director -- also took up ports by using them as interswitch links (ISL). Hesson said he also liked reducing the number of vendors he had to deal with by choosing Cisco, which supplies his LAN network.
The transfer of data to the new SAN environment took about 70 days and did not affect Kindred's users because Hesson was able to take advantage of the redundant network to keep systems online.
"It was probably one of the best project experiences I've had, and I don't say that lightly," Hesson said.