Scaling Up In Servers? Back Up To A SAN
With data from 1,500 servers to fit through the backup window, a hosted services provider moved many of its customers' backups to SANs (storage area networks).
Every corporate IT director knows the feeling of being responsible for the organization's data. Crashed systems must be restored; lost data must be recovered. But, the pressure of delivering data for one's own company, though demanding, is not nearly as anxiety inducing as delivering it for someone else. Just ask an ASP (application service provider).
For ASPs, promises about data delivery are not made to be broken. USinternetworking, Inc. (USi) (Annapolis, MD) is no different in that regard.
For its customers, USi houses and manages the hardware and software needed to serve up major enterprise applications from vendors such as Ariba, BroadVision, Lawson, Microsoft, PeopleSoft, and Siebel. In its SLAs (service level agreements), USi promises to retrieve customer data within specified time periods. It guarantees the return of lost files in less than an hour for urgent needs and in no more than 24 hours for less time-sensitive applications.
To make good on those promises, USi must have reliable, scalable data backup systems. And, those systems must be able to accommodate the increasing data storage demands of an expanding customer base. Over the past three years, USi's client base has nearly doubled, growing from approximately 80 customers to more than 150. Three years ago, when it forecasted future growth, USi recognized limitations in the storage backup systems then on the floor. Says James Dean, director of enterprise backup systems at USi, "What we had at the time couldn't handle 100 to 200 application servers per master backup server. Plus, we knew we were eventually going to have to meet the needs of 5,000 or 10,000 servers." Three years later, USi is already housing more than 1,500 application servers for its customers. As it has grown, USi has solved the scalability issue by moving more and more of its customers' data storage, including tape backup, to SANs (storage area networks).
Backup Servers Watch Each Other's Backs
In each of its data centers, USi now offers SAN (storage area network)-based storage in which its customers' application servers and primary storage devices are backed by DLT (digital linear tape) drives. A key example is its original data center in Maryland. In that center, USi implemented a backup solution built around VERITAS NetBackup software and a Quantum/ATL (Irvine, CA) P3000 DLT (digital linear tape) library.
The Quantum/ATL library backs the data center's server and application infrastructure with 16 DLT (digital linear tape) 7000 drives and 326 DLTtape cartridges. An integrated Fibre Channel bridge, also from Quantum/ATL, provides a direct connection between the Fibre-based SAN and each of the 16 SCSI (small computer system interface)-based drives in the DLT library. This configuration enables any of the backup servers to perform backups over the SAN to any drive in the tape library. If one backup server's performance is slowed by simultaneous backup events coming from multiple application servers, another backup server can assume some of the load and continue the backup process to the same drive.
SAN And IP Storage Cover The Range
A typical USi customer runs four to six enterprise applications and generates a lot of data, anywhere from 2 GB to several hundred GB. For its largest customer, USi backs up as much as 130 GB from each server. Combined, USi's customers put a daily 3.5 to 5 TB demand on the storage backup systems across its various data centers.
According to Dean, backup in a SAN is enhanced by zoning based on platforms rather than on applications. This approach allows USi to send data from applications on different platforms in separate zones to the same storage frame. Grouping each customer's servers in their own pod is another key to efficient zoning. "Even though our customer pods are on a shared network, they function as though each is in its own virtual data center," Dean explains. He notes that the topology of a USi data center SAN ensures seamless zoning. The primary storage devices, the backup servers, and the tape drives are connected to the core of the SAN, made up of multiple Brocade switches. Additional Brocade switches on the outside of the core make up the customer pod connections. "Having each customer pod connected to its own switch for connecting to the SAN makes it easier to zone data for storage and backup," Dean says.
Not all of USi's customers are connected to the SAN environment. Those clients with storage demands of less than 50 GB per server have their data routed to and from IP (Internet protocol)-based direct attached drives. In the original Maryland data center, additional Quantum/ATL libraries provide non-SAN-based backup. "We try to limit the load on the IP-based backup process because it's somewhat slower," Dean says.
For all customers, USi offers a backup window between 10 p.m. and 8 a.m. To ensure that all customers stay up and running during that time, USi monitors performance and capacity on the backup system. If an application demonstrates a high I/O (input/output) rate, USi either adds more processing power or limits the bandwidth the customer can use for backups. "In the entire time we've been monitoring backups, we've had to throttle back only a couple of servers during backup," Dean asserts.
For Dean, the major benefit of performing most customers' tape backup over a SAN is the relief it brings to the network on which the server pods run. "Because we do backup across Fibre Channel SANs, we are able to enhance performance and move more data without clogging up the primary network," says Dean.