You are on page 1of 4

S T O R A G E

E N V I R O N M E N T

Multiservice Access for SAN-to-SAN Connectivity


By Matthew Brisse and Michael Harris

SAN-to-SAN connectivity promises interoperability among geographically dispersed storage area networks. This article describes several methods for SAN-to-SAN connectivity and outlines the benefits of a multiservice access approach.

ne of the most promising areas of storage technology today is SAN-to-SAN connectivity. Extending storage area networks (SANs) beyond the local fabric will greatly improve data availability, disaster avoidance, and storage consolidation. Longhaul SAN transportthe ability to connect geographically disparate SANs and make them look like one fabricprovides a simplified management model and allows consolidation of key human resources. Long-distance SAN-to-SAN connectivity is also important for storage service providers (SSPs) because it allows economic and business needs, rather than location, to dictate the placement of consolidated storage facilities. Unfortunately, confusion exists in the marketplace regarding SAN-to-SAN connectivity. Multiple protocols (some based on open standards and some proprietary), competing technology camps, and a range of opinions on what is considered the best option for SAN connectivity are sending mixed signals to the IT community. This uncertainty threatens to delay deployment of this useful technology to the detriment of business operations. However, a proven technology, which has been in existence for years, is mirroring from one SAN to another. Adding a relatively new technologysnapshot and virtualizationto mirroring can create a powerful combination that many IT organizations can implement today. (See Figure 1.)

fabrics is either server or SAN appliance based. Integrated solutions based on existing local area network (LAN) and wide area network (WAN) infrastructures, Fibre Channel over Asynchronous Transfer Mode (ATM) and Synchronous Optical Network (SONET), and Fibre Channel-to-T1/T3 are quickly emerging. The viability of each methodology depends on organizational data availability needs and the requirements of different storage applications. The most common of these applications includes storage consolidation, disaster avoidance/recovery, backup, and virus protection. Server-based Server-based approaches employ a server with host bus adapters, network interface cards (NICs), and specialized software for routing either Fibre Channel packets or Small Computer System Interface (SCSI) blocks to a LAN for transport over an existing WAN. Serverbased solutions can be deployed immediately but often complicate long-distance SAN transport. Server-based SAN transport can be costly and difficult to manage because it typically requires a high-end server with vast amounts of memory, powerful processors, and specialized software. These powerful servers usually support mission-critical applications and cannot be dedicated only to data transportation. Appliance-based Appliance-based approaches employ a SAN appliance specifically designed and architected for the virtualization and movement of

Current approaches to SAN connectivity


Several distinct methodologies enable SAN-to-SAN connectivity. The most widely deployed method of extending SANs past local

www.dell.com/powersolutions

PowerSolutions

S T O R A G E

E N V I R O N M E N T

PSTN

Alarm/management call

Remote management

Servers

SAN

Ethernet SNMP SAN management

...
Fibre Channel switch

...
Fibre Channel Director OC-48

T3

OC-3 WAN

DR backup SAN

...
Disk arrays

...
Enterprise SAN

IP network ATM Router

Gigabit Ethernet

SSP SAN

ASP SAN

Figure 1. Protocol-neutral multiservice access allows flexible connectivity options

data within a SAN and Internet Protocol (IP) environment. SAN appliances include specialized hardware and software to provide logical unit number (LUN) masking and mapping techniques while combining high-end features such as LUN virtualization, LUN concatenation, sub-LUN partitioning, snapshot, and mirroring functions. Advanced appliances such as the Dell PowerVault 530F can perform synchronous Fibre Channel mirroring as well as synchronous and asynchronous IP-based mirroring. SAN appliances combine virtualization techniques with snapshot and mirroring technologies. This combination enables administrators to create point-in-time clones of LUNs and then mirror the LUN to a remote site where the targeted LUN can be assigned to a server for backup or testing, or staged in the event of a disaster or virus attack. The mirroring transport for appliances is typically Gigabit Ethernet or Gigabit Ethernet over ATM. For online transaction processing (OLTP) and data-sensitive applications, administrators should profile the latency requirement to ensure adequate data response. Gigabit Ethernet and SONET (OC-12 or higher) are adequate transports for most applications, depending on the distance that the data will travel. Gigabit Ethernet Another approach is to deploy devices that leverage current Ethernetbased network infrastructures to bridge Fibre Channel SANs and existing LAN-to-WAN routers and switches. Do not confuse this

method with storage over IP systems in which all storage traffic, including SCSI block I/O, is run over an Ethernet or Gigabit Ethernet network. The obvious advantage of transporting Fibre Channel packets through existing LANs and WANs is that most of the infrastructure is already in place. In addition, the IT community is very familiar with its current infrastructure, resulting in faster implementation and easier management. The major disadvantage is congestion and latency. Running block I/O traffic with other traffic can overload the primary network and, during periods of heavy usage, prevent storage traffic from arriving in order and on time. Data must be routed through multiple switches and routers before it enters the WAN, which compounds latency problems and complexity. For these reasons, the Fibre Channel-toGigabit Ethernet approach best fits the following scenarios: organizations that have existing infrastructures, emergency failover from a dedicated storage WAN link, periods when network traffic is light (for example, after business hours), and situations that require lowto moderate-speed bandwidths for financial considerations. SONET In many cases, a dedicated connection to remote storage resources is preferred, especially when this connection is constantly available or when large amounts of block I/O traffic are expected to flow between geographically dispersed SANs.

PowerSolutions

Issue 4, 2001

SONET (OC-1 and up) provides effective SAN-to-SAN connectivity because of the high throughput (up to 2.5 Gbps), multipoint service, and more reliable error handling. Disadvantages of SAN transport via SONET include its relatively short range (restricted to metropolitan area networks), high cost, and lack of availability to many enterprises. A typical list price for SONET (OC-12 to OC-48) in the New York City area can be as high as $300,000 per month. SONET (OC-48) is a good choice when running I/O-intensive applications over short distances. However, the storage application must be mission-critical to justify the expense of dedicated solutions.

protocol for transporting Fibre Channel SANs over IP, ATM, or SONET networks. Protocols specific to a particular vendors product also exist. This approach works well if a customer is solving a specific need and is an early adopter of technology. However, proprietary systems can lock IT managers into a specific vendors solutions, which may not serve all of their future needs effectively. Proprietary systems offer little protection against technology obsolescence.

Multiservice access can support different protocols


Faced with so many compelling and competing approaches, vendors, integrators, and end users are challenged by questions such as: Which type of connection should I use? Which protocol is the right one? Whose approach will dominate in the end? Many IT professionals with responsibility for multiple storage applications believe they have no choice but to deploy different vendors products, many of which are not interoperable. This decision is a risky career move unless purchased from a tier-one player that has tested, certified, and promised to stand by the entire solution. An alternative to the multivendor solution is the multiservice access approach. Multiservice access provides a variety of WAN, LAN, and metropolitan area network (MAN) interlinks that support different protocols. Such access offers IT managers the opportunity to make decisions based on need and budget. Multiservice access frees IT professionals from being at the mercy of a particular vendor. The applications, rather than the router or switch vendor, can now drive the connectivity planning. By deploying a variety of open protocols, IT managers can interoperate effectively and efficiently with other facilities, customers, and suppliers. (See Figure 2.) The multiservice access concept has been a fundamental element of traditional networking. When applied to SAN-to-SAN connectivity, multiservice access provides IT managers with the right tools for meeting the constantly escalating storage demands of their enterprise environments. Multiservice access offers five advantages: Maximized performance. The needs of mission-critical applications can determine WAN-link deployment decisions. Multiservice access also can maximize application performance. Tighter cost control. Bandwidth can be purchased based on need, which supports tighter cost control. WAN links are a recurring monthly expense. Investment protection. IT managers can order incremental upgrades in connectivity with minimal changes and disruptions to the overall system, ensuring investment protection. As the system requirements grow, decisions can be based

Network latency considerations


The issue of network latency is fundamental to the intrinsic nature of data networks. Distance, packet loss, network congestion, link errors, and network errors cause latency. Storage applications such as databases and OLTP are extremely sensitive to the effects of data latency. Even in a perfectly engineered network, link errors occur. For example, consider the Fibre Channel objective of 10-12 bit error rate. For a 10 Gbps link, this metric equates to one error every 100 seconds. Network errors occur with significant frequency in IP networks. Transmission Control Protocol (TCP) catches network errors that occur at a ratio between 1 packet in 1,100 and 1 in 32,000. Cyclical redundancy check (CRC) does not catch these errors. Typical IP network characteristics that permit 3 percent to 5 percent packet loss are potentially devastating not only to data latency, but also to the overall throughput of the application solution. The amount of possible errors in a typical data network can be exponential when compared to that of a Fibre Channel network. In a direct-connect (server to storage) scenario, the average latency is 3 milliseconds to 5 milliseconds with most of the latency associated with the disk drive mechanics. Additional latency from errors or distance associated with IP networks might affect the storage applications themselves. Storage applications such as OLTP have very stringent latency requirements. Proper throughput sizing is imperative when considering SAN-to-SAN connectivity.

Open vs. proprietary protocols


Transport protocols are another important element of SAN-to-SAN connectivity. Open standards-based systems compete with proprietary protocols. Many vendors are backing the open protocol standards developed by the major standards bodies. These protocols include Fibre Channel (FC) over Internet Protocol (IP), which is currently under discussion by the Internet Engineering Task Force, and Fibre Channel Back Bone, which is under revision by the American National Standards Institute T11.3 committee. Both proposed standards represent an attempt to develop an open

www.dell.com/powersolutions

PowerSolutions

S T O R A G E

E N V I R O N M E N T

Site A
Storage disk mirroring Local Leased line, Dark Fiber DWDM, ATM, and/or IP Disaster recovery Long haul

Site B

SAN

SAN

SoIP Storage Switch, and the Entrada Networks Silverline-222 SAN-over-IP transport switch. The Silverline-222, for example, can connect SAN islands across a variety of network topologies. Connection options range from T3 for ATM-based WANs, OC-3 and higher feeds for WAN/MAN networks, and Gigabit Ethernet for transporting SANs over existing high-speed IP networks. With products like the Silverline-222 and others on the horizon, the outlook is promising for IT professionals faced with the daunting task of connecting SAN islands in todays environments.

Remote backup

Multiservice access offers viable solution


The SAN-to-SAN connectivity landscape is new and fragmented. Several different schools of thought have emerged regarding the correct way to allow geographically dispersed SANs to interoperate. Advancements in Fibre Channel, Gigabit Ethernet, WAN infrastructure, and other key storage networking technologies are making the multiservice access approach a reality, and one that customers should consider. Matthew Brisse (Matthew_Brisse@Dell.com) is the SAN product manager for PowerVault 530F, Dells SAN virtualization appliance. He has worked in the storage industry for more than 19 years in engineering development and product management roles. He holds a B.S. in Sociology, Statistics and Analytical Research from the University of Wisconsin. Michael Harris (MAHarris@entradanet.com) is senior vice president of Entrada Networks, where he is responsible for marketing, business development, and investor relations. Michael holds an MBA from Vanderbilt University and a B.S. in Communications from the University of Tennessee.

SAN

Site C Figure 2. Multiservice SAN transport supports many networking applications

on economics rather than a particular vendors product schedule. Efficient service. Various groups within the enterprise can use a solution best suited to their scenario; for example, SONET (OC-48) is not universally available. Even within a single company, WAN infrastructure varies widely. Accordingly, relying on one approach for SAN-to-SAN connectivity means that some locations cannot use an application until that service is available. If an organization implements a slower interconnect such as a T1, some bandwidth-intensive applications will either run at an unacceptable slower speed or, even worse, fail because of unexpected latencies that were not anticipated by the application designer. Enhanced interoperability. As SAN-to-SAN connectivity becomes increasingly important, the interoperability of equipment from multiple vendors and different connectivity options is vital. Moreover, in a time when mergers and acquisitions are commonplace, interoperability is an important part of IT planning and should be considered when deploying SAN-to-SAN applications. Vendors have announced multiservice access devices: the Cisco SN 5420 Storage Router, the Nishan Systems IPS 3000

FOR MORE I NFORMATION


Cisco Systems: www.cisco.com Dell Computer Corporation: www.dell.com Entrada Networks: www.entradanet.com Nishan Systems: www.nishansystems.com American National Standards Institute: www.ansi.org Internet Engineering Task Force: www.ietf.org

PowerSolutions

Issue 4, 2001

You might also like