Optical Networks for The Enterprise: The Broader View.

Storage Area Networks (SANs) and Network Attached Storage (NAS) are the buzzwords in today's Enterprise and Service Provider space. Their popularity is due to the increased demand not only for access to the storage facility itself, but also for the flexibility they provide in controlling information.

In the midst of all of this hoopla about storage and SANs and NAS are the basic business requirements for moving, storing, accessing, and managing information--not only within a single data center, but between two or more data centers. Driving the movement toward integrating and leveraging multiple data centers are applications such as high availability, backup and recovery, data vaulting, and information sharing.

Making the Connection with Fiber Optics 

To move information between sites efficiently and effectively within a metropolitan area, many businesses are taking a serious look at fiber optic connectivity. Until now, most storage vendors and solution providers have been taking the narrow view that the killer application for optical networks within the metropolitan area is storage access.

While it is true that remote storage access for a Disaster Recovery application can provide a reasonable justification for building a Storage Wide Area/Metro Area Network (SWAN or SMAN) using optical connectivity, this is a limited view. Looking at it more broadly, a fiber optic network connecting multiple sites allows a business to build a much more powerful business tool--a geographically dispersed Systems Area Network.

Optical Systems for the Enterprise

The optical multiplexing technology being used today for metropolitan area connectivity for the enterprise is Wave Division Multiplexing (WDM) or its more current adaptation, Dense Wave Division Multiplexing (DWDM). This type of connectivity can provide a business with much more than simply access to massive amounts of storage at high speed across a region. DWDM optical networks offer tremendous capacity, scalability, availability, and protocol transparency that other technologies can't provide. The underlying value of a DWDM network is in understanding how to leverage those features effectively, and what the implications are to the business and the applications in using this technology.

Major Features of a DWDM Optical Network

A DWDM optical network offers a substantial number of features, including:

* Capacity, Bandwidth, and Speed, and lots of it--up to about 640Gbps (Gigabits per second) per system today. Protocols and applications can all run at their native speeds, i.e., 17MB (MB/sec) for ESCON, 1.062Gbps for Fibre Channel (FC)--2.12Gbps soon--and 1.25Gbps for Gigabit Ethernet (GbE). Relative to traditional carrier circuit technologies, a DWDM optical network today has a capacity 14,222 times that of a D53/T3 circuit (45Mbps), about 4,129 times that of an OC3 circuit (155Mbps) and 512 times that of a single Gigabit Ethernet or Fibre Channel connection (1.25Gbps or 1.062Gbps respectively).

* Scalability--some vendors can support up to nine sites in a protected optical ring--a highly available topology, and up to 160 channels. There's nothing to stop you from building multiple optical networks and tying them together with devices such as Fibre Channel, Ethernet, or ESCON switches or directors.

* Protocol Transparency/Low Latency--Most DWDM systems support Fibre Channel, ESCON, ATM, FICON, GbE, SONET/SDH, Sysplex Coupling Facility timers, and video. Little latency through the device is noticed because only modest processing is required to direct the optical signal from the input to the appropriate laser or vice versa.

* Availability--Different vendor architectures provide different levels of availability, but overall, redundancies are either built-in or can be implemented externally. There are costs and implications associated with each option.

* Long life span--DWDM optical connectivity in the Enterprise is in its infancy yet is proving to offer an effective competitive edge. Along with wireless, communication over fiber optics will be around for a long time to come.

Applications

With all these features available, it becomes fairly obvious that an extensive array of both old and new applications can now be deployed between multiple data centers across a region using a single, consolidated infrastructure that before was either ineffective, inefficient, or impossible to implement. This applies to both mainframe and open-systems applications.

* Disaster Recovery. Backup and recovery processes become much more effective providing a real opportunity to reduce downtime, thereby lowering risk and reducing losses. Imagine the decrease in recovery time using several 1Gbps Fibre Channel links to restore data to a primary disk subsystem from one that's remote, rather than across a single ATM connection at 155Mbps typically in use today. A single Fibre Channel link provides about a five-fold decrease in the time to recover versus a 155Mbps connection. And it's quite possible to have multiple Fibre Channel connections transferring data simultaneously.

If you are outsourcing your Disaster Recovery application, DWDM gives you several choices. You may be able to tie the outsource facility into the network, or because of the better facilities and lowered cost, bring it back in house--a method gaining popularity.

* Business Continuity. Applying clustering technologies between locations for application or server failover gives you the ability to provide greater uptime and better leverage your open systems resources. Load balancing may also be applied where appropriate and available.

If you're a Storage Service Provider (SSP), why be limited to providing only storage? With an optical network in place, you can easily add more value to your offerings. Using Microsoft or VERITAS Cluster Server software, for example, you can provide customers with a high availability service by placing servers in the POP (point of presence) to support remote application or server failover functions.

* Centralize LAN-free and server-free backup and restore functions. One tape silo may be able to support an entire system, significantly reducing backup and overhead costs. The time to complete a backup or restore function to or from a storage device decreases tremendously vs. other remote methods.

* Circuit/Fiber Consolidation. The cost for fiber has dropped dramatically over the past year or so depending upon your ocation. One client in the NYC area reports the total cost of his 4-DS3 and 1-OC3 circuits as $75,000/month. A recent quote he received from a dark fiber supplier for dual circuit pairs of 46 and 65 miles respectively is $19,000/month --a savings of $56,000 per month, redundancy included. Some of the savings will necessarily get invested into the DWDM hardware but this business will still come out way ahead.

* LAN connectivity. Gigabit Ethernet is here with 10GbE on its way. You can increase performance between locations by connecting LAN switches together at gigabit Ethernet speeds over the optical network providing a faster, flatter, and more easily manageable network.

* Data Movement. This is a catch-all category for applications such as multimedia delivery and video streaming that require the storage, movement, and delivery of massive amounts of information quickly.

* Storage Virtualization. This new technology has the capability to allow you to manage a pooled multi-vendor storage resource through a single policy-based framework across multiple sites. Optical connectivity can make this application transparent from a performance standpoint.

* Future options and long-haul integration. Deploying an optical network can also prepare your business for future applications such as storage over IP or any of the up-and-coming storage or data-movement related protocols (iSCSI, FCIP, SoIP, VIA, InfiniBand, etc). Other types of communications facilities can also be plugged into the network such as channel extenders for access to ESCON/FiCON devices beyond the metropolitan area, and for integration into the optical long-haul network when available.

Performance

With all of that said, one issue that needs to be clearly understood is how your applications will perform at these extended distances Many applications and protocols were not originally designed to operate over distances greater than a few meters or kilometers. For example, the maximum distance suggested for ESCON is 43km. At the Hitachi Data Systems interoperability laboratory we found that ESCON can operate error free up to at least 90km using a major DWDM vendor's optical multiplexer, but the performance of the application at that distance dropped to about 20% of the performance at 0km. We measured a slow drop-off in performance from 0km to 50km at which point a steep curve begins and performance drops precipitously.

Fibre Channel is another story. Since the Fibre Channel protocol has the ability to manage data flow through the use of buffer credits, much greater performance can be achieved over longer distances. Tests measured nearly 100% performance at 87km.

Storage over IP across an optical network is an area scheduled for study. The concern is latency through the network and how the performance of an application is affected by the TCP/IP protocol. We'll be looking closely at application performance as the distance between endpoints is extended across multiple topologies.

Regarding the applications themselves, there is a distinct difference in the behavior between synchronous and asynchronous applications. That is to say, synchronous data transfer requires an acknowledgement that a command has been received at the distant location before the application can resume transferring data; asynchronous transfer does not require this delay in transmission. Even though optical networks can offer transmission at wire speeds, when traversing tens or hundreds of miles, even at the speed of light (light traveling through a glass fiber moves at about two-thirds the speed of light in a vacuum), propagation delay and device turn-around-time become the new limiting factors.

For business requirements where losing a single transaction is significant and application performance is paramount, synchronous data transfer or synchronous remote copy programs are the applications of choice. But, as distance to the remote site increases, there comes a tradeoff point where you will have to decide whether the decrease in performance caused by the added distance is worth preserving every transaction in case of a failure, or if asynchronous data transfer should be used to maintain high performance, even though there is a greater risk of losing multiple transactions.

It is now possible to deploy an effective geographically dispersed Systems Area Network around a region built upon fiber optic technology. Performance and scalability are inherent, costs are dropping and availability can be designed into the system. By consolidating multiple, disparate systems across a single platform, a business can save circuit costs, grow the business easily because the infrastructure is in place, and more easily manage the entire network.

Jerry Zeisler is a business solutions development manager at Hitachi Data Systems (Santa Clara, CA).