SANs For The Internet.

ISPs, ASPs and SSPs all require SAN solutions.

This article is the first in a two-part series. The second part will appear in the July issue of CTR.

The development of the World Wide Web has profoundly changed both information access and information content. Access, which was previously restricted to private computer systems, is now universal. Any platform with Internet connection and a browser can access information anywhere in the world. Content, which was previously text-based data, is now information in the form of text, graphics, sound, and video, often on a single screen. Institutions and enterprises have unprecedented visibility via the Internet to provide information, advertising, and online transactions and have effective content formatting tools for creating interest in their services or products. Ubiquitous access to knowledge bases, entertainment, and products has proven to be so successful that the rate of development of Internet infrastructure and content generation have enjoyed explosive growth.

As academic and government institutions, enterprises, and businesses of all sizes are implementing Internet technology, new Internet services have emerged to facilitate information access. Internet Service Providers (ISPs), for example, allow enterprises to connect to the Internet backbone over high speed lines. Without ISPs, companies would have to install and administer their own Internet links at significantly higher cost. ISPs may also provide Web hosting services. By provisioning its Web servers at the ISPs site, an enterprise can reduce administrative overhead and benefit from higher performance due to colocation to the ISP's Internet link. Web hosting services may include additional offerings such as Web content generation, content updates, tape backup, and bill-back services for individual departments within an enterprise. By providing the technical expertise required for Web content generation and administration, ISPs can offer significant savings over in-house implementations.

In addition to ISP functions, Internet connectivity has enabled completely new services to off-load technical and administrative overhead. Application Service Providers (ASPs) have emerged to solve a persistent and expensive problem faced by all corporations: maintaining and updating applications for their users. If an enterprise buys and maintains its own software applications, it bears the burden of licensing costs, administration, allocation to end users, and maintenance of application version levels. An ASP assumes all these responsibilities and, for a monthly or annual fee, makes software applications available, often via the Internet. This arrangement puts powerful software like Oracle into the hands of companies that otherwise could not afford to staff and maintain a major business application. Microsoft Backoffice, Office, and other suites can be made available to end users and automatically maintained by the ASP as new versions or components are introduced. The ASP market is expected to grow to a pot ential $8 billion by the year 2003.

Applications are used to generate user data, which in turn, must be stored and archived. Even without the additional storage requirements of Web content, enterprises depend on massive amounts of customer, order, inventory, manufacturing, administration, and other information that is essential for a company's existence. Loss of data inevitably means loss of the enterprise. The addition of Web content is driving storage requirements ever higher to the point that companies find it difficult to satisfy their systems' ravenous appetite for servers and disk arrays. More than any other single factor, e-commerce is pushing the limits of previous storage infrastructures. Servers must be faster and more numerous. Disk arrays must be larger and more accessible. And 7x24x365 availability requirements demand new technologies for more efficient and reliable data access. Storage Service Providers (SSPs) represent another solution for storing and securing enterprise information.

Like ASPs, the SSP service model relies on Internet transport to off-load a company's storage requirements. The SSP assumes the responsibility for provisioning sufficient storage, providing the technical talent for storage administration, and safe-guarding data via tape backup. This allows the client company to focus on their core business and to reduce their IT staff and storage administration overhead significantly. The infrastructure of an SSP assumes high speed access between the SSP and the client, high speed transport between the SSP's servers and disks, and an enormous storage capacity to service multiple clients.

Although their service offerings may vary, ISPs, ASPs, and SSPs have common requirements that are driving each towards Storage Area Networking (SAN) solutions. These requirements include: High availability for servers; high availability for storage; increased storage capacity; high speed storage access; shared storage; support for high populations of devices; flexibility in server and storage placement; and data security via tape backup.

The simultaneous emergence of the Internet's storage-related needs with SAN-based solutions is a fortunate coincidence. SAN technologies and, in particular, Fibre Channel SANs, were under development well before the sudden explosion in e-commerce and Web-inspired content. Consequently, just as the Internet is bursting through the confines of previous server and storage architectures, SANs can offer new and viable solutions to move information access and content to the next level.

The Traditional Internet Storage Model

A Web site, hosting site, ASP, or SSP may be configured as a small network segment as shown in Fig 1. One or more high speed communications links provide access between the site and the Internet backbone. The speed of these links determines how fast content can be served up for individual requests, as well as how many concurrent requests can be satisfied. Before high speed links became available, the link itself was the gating factor for Internet expansion. Few users would use the Internet if each browser screen took several minutes to update.

Internet links are handled by IP (Internet Protocol) routers. Routers provide the interface between the "cloud" of the telecommunications infrastructure and the Web site's local LAN segment. The router's link to the cloud may be ATM (Asynchronous Transfer Mode), Frame Relay, or point to point, and offers bandwidth in the megabit or hundreds of megabit range. Multiple links to the cloud can provide much higher aggregate bandwidth. On the far side of the cloud, an individual end user making a request may have a much slower link speed, but taken collectively, thousands or millions of these relatively slower end user requests may flood simultaneously into a single site. Web service sites with high speed Internet access can help insure that user requests do not get dropped or blocked and discourage site hopping by impatient or disappointed users.

Access to a web site (URL) is based on Domain Name System (DNS) rather than physical addressing. DNS allows multiple Web servers and multiple sites to service user requests using the same content. Different Web servers, for example, may satisfy concurrent user requests, although to the end users, it appears that the content is being delivered from a single resource. This configuration has a number of benefits. The workload is distributed, so that no single server bears the brunt of user access. With multiple servers responding to a single URL, hundreds of hits can be serviced simultaneously and, in the event of a server failure, user requests are transparently redirected to other available servers.

Traditional server configurations with direct SCSI-attached storage, however, undermine the efficiency of the DNS scheme. Each server retains exclusive access to its attached storage. If the server fails, access to its data is lost, but even if all servers are operational, each must maintain an exact copy of the URL content. Updates to Web screens must, therefore, be coordinated among multiple servers. This is less of a problem if the web content is static, as with sites that provide read-only information. It is a significant issue for dynamic sites that offer variable content (e.g., news) or provide e-commerce services. The evolution of web content from read-mostly to online transactions has encouraged the separation of storage from individual servers. Just as any server may provide content to a user request, storage must have the flexibility of providing content to any server.

Additionally, as the amount of content grows, it is increasingly difficult to provision sufficient storage by traditional parallel SCSI attachment. Adding multiple SCSI strings to a single server may provide a temporary solution, but at the expense of additional administrative overhead and downtime. Attaching more drives requires taking the entire SCSI string off-line and, since each server must maintain the same content, all servers in a cluster would have to be serviced at the same time to increase storage capacity. For 7x24 operations, even planned downtime may be unacceptable. The distance limitations of parallel SCSI cabling may also be an issue as more drives are attached. Co-location of servers and storage becomes increasingly difficult as racks fill with equipment and new chassis must be accommodated within parallel SCSI's 15-25 meter limitations even if the original floor plan did not allow for expansion.

Depending on the type of content a site provides, storage bandwidth may be problematic. Concurrent requests for streaming audio and video content require higher bandwidth between the provider and the Internet and between the servers and storage. While bandwidth is somewhat mitigated by the link speed of the requestor, it is a significant issue for business-to-business and intranet applications that typically enjoy higher speed connections.

SAN Solutions For The Internet

The geometric growth of Internet and e-commerce applications is driving the contradiction between old and new technologies to the forefront. Traditional Tl and T3 speed links have given way to much higher capacity optical connections. Intelligent routers and switches have replaced first generation products. More powerful servers optimized for high availability have enabled consolidation of server resources and, for storage, Fibre Channel-based SANs are resolving the majority of infrastructure issues faced by Internet, application, and storage service providers. As shown in Fig 2, a storage network overcomes the constraints of traditional storage by introducing the flexibility of networking between server and storage.

Fibre Channel SANs offer the bandwidth, distance, hot-plug, and device support required for the dynamic growth experienced by Internet providers. In Fig 2, a' Vixel 7000 series switch delivers 100Mbps per port for high-speed Access between servers and Fibre Channel-attached disk arrays. The servers can be up to 500 meters (using short wave fiber optics; up to 10km using long wave) from the switch, with up to 500 meters between the switch and each array. Servers and arrays can be hot-plugged into the SAN without disrupting ongoing transactions, which eliminates downtime for server or storage expansion and an extended SAN can be built with 1OOs or 1,000s of devices in a single configuration. The flexibility of Fibre Channel SANs not only overcomes the limitations of legacy storage connectivity, but also provides new network designs that enable new businesses such as SSPs.

High Availability For Servers

By making servers and storage peers on a common network, service providers can implement clustering configurations to maximize server availability. Individual servers can be inserted into or removed from the SAN without disrupting ongoing transactions and, since no server is the exclusive owner of storage, server resources can be manipulated without affecting storage availability or administration. Fail-over software adds a supplementary level of reliability, since, although DNS may treat a server pool as a single resource, fail-over can insure that clustered servers assume the workload if a problem occurs.

High Availability For Storage

In a SAN architecture, storage plays an independent role. Since an array is acessible from a number of servers, storage is no longer dependent on the availability of any single access point. This facilitates a number of availability designs. In Fig 3, each server is provisioned with two Fibre Channel Host Bus Adapters (HBAs), with each HBA inserted into a separate SAN segment. Fibre Channel disk arrays offer primary and secondary ports, which allows for dual attachment to a single resource. By installing two Vixel 7000 series switches, the servers have two data paths to the same content.

Unlike unmanaged SAN configurations, the status of each path is constantly monitored by the Vixel 7000 series switches and reported to SAN InSite 2000 management software. If a path fails, traffic continues on the available path while the administrator is notified of the failure by SAN InSite via page, email, or SNMP trap to an upper layer management framework. As shown in Fig 3, it is also possible to provide redundant storage on the SAN. Mirroring from one array to another can provide additional insurance for availability, even if an entire array is taken out of service. Fibre Channel-attached RAID systems and JBODs (Just a Bunch of Disks) are available in the market, which gives the service provider additional storage options to meet reliability, capacity, and budget requirements.

Tom Clark is the technical marketing director at Vixel Corporation (Bothell, WA) and the author of "Designing Storage Area Networks: A Practical Reference for Implementing Fibre Channel SANs" (Addison Wesley Longman).