An iSCSI SAN has solved a server-to-server data transfer problem at Southampton Football Club. The club sells around 700,000 tickets a year and recently bought a new ticketing application. Ticket sales typically have a tremendous peaks and valleys effect in terms of system usage and performance was paramount. The new application also required additional storage and this prompted a review of the existing infrastructure.
At that point Southampton FC used a cluster of servers with direct-attached storage. It stores five days' worth of ticket sales in its customer relationship management database. Each day represented 15GB of ticketing data, so the total for five days is 75GB. This data is backed up after hours.
This clustering entailed the daily transfer of ticketing data from server to server - there was no shared cluster storage - and that used up server compute cycles. Mark Wierzbicki, IT director for the club, explained; "Clustering with servers would have been reliable, enabling us to switch to the next server should one fail." But data copying was problematic.
There are four server rack rooms, one at each corner of the stadium, and the servers had to be synchronised. With the DAS this was difficult. Wierzbicki describes the problem; “We had to go across the network, find the right machine and bring up the accessible drive. If the drive wasn’t available, we had to physically go across the field to that machine in order to locate the drive and copy the data,” This data copying was the problem, because of server CPU cycle consumption.
The risk was that the ticketing application performance would be hit and ticket sales affected. Wierzbicki explained; "We have a limited window in which to sell tickets and, unlike a theatre performance, a sporting event can only be played once. The system simply cannot fail or we run the risk of creating thousands of very unhappy fans. If we transfer from one server to another server, we utilize CPU time that is needed to process other things."
The club decided to stop copying data and use shared storage instead. It went for an iSCSI SAN based on an Adaptec iSA1500 Storage Array with an iSCSI-to-Serial ATA RAID subsystem and standard Ethernet LAN links.
There is central management of the servers and no need for the cross-stadium treks previously involved in synch'ing the servers. Wierzbicki said, "With (this) we don't risk slowing down any of our other information systems."
Backup performance improved. Previously, an engineer spent two to three hours every day—after hours—backing up data. Now, it takes just 30 minutes and does not require the 4-man team to stay late. “With a small team like ours, we can’t afford to work anyone too many hours,” says Wierzbicki.
There were cost-savings as well. Wierzbicki estimates the club paid two-thirds of what it would have paid for server clustering, not including licensing fees or setup and maintenance. Also the savings would be massive should the system ever go down; “For starters, we could loss up to one million pounds in ticket sales for just one game. Then there are fines from the games ruling body. Not to mention the cost of the loss of good will from our fans. Given that there are only 22 of these play-days each year, disrupting one of them would be disastrous.”
A Fibre Channel SAN would have been more expensive. The iSCSI SAN uses well-understood Ethernet technology and its implementation was straightforward; Wierzbicki said, “I just showed the team once and away they went.”
Is this football club-specific SAN story of general applicability?
Where clustered servers with DAS are having to copy data across a network and performance is critical then an iSCSI SAN would appear to be something to be checked out. It provides storage consolidation and could save server CPU cycles as well as having other benefits that could raise IT performance up the league tables.