Using storage area networks to reduce congestion and increase performance

Increasing amounts of data are clogging existing networks. Storage area networks offer a way of reducing congestion and...

This Article Covers


Increasing amounts of data are clogging existing networks. Storage area networks offer a way of reducing congestion and increasing performance.

Back when a network meant a mainframe and a collection of dumb terminals, storage wasn't too much of a problem. Data was stored in magnetic disk packs called DASD - direct access storage devices. When you ran out of space, you simply added another pack. There were time implications in accessing data in this way, but, because of the delays in processing inherent in mainframe technology, this wasn't noticeable to the end user.

The advent of the PC brought with it a revolution in storage. Information was either kept on the hard drive - a cut-down version of DASD ( or in RAM, where it could be accessed immediately. Both storage methods speeded up data access considerably, since it was all occurring within the PC itself.

When distributed client/server networks came into their own, storage once again moved out of the PC and into a centralised location. Unless you were running a small network in which the data was held on the server, the chances were that your network data was being held on a dedicated file server or on some kind of storage device, such as tape, or magnetic disk. Soon, data ceased to be the property of individual users or even individual departments, and became an enterprise-wide commodity. For this to work, however, the storage devices had to be available for access across the entire network. This created even more complexity in the network.

This situation was all well and good until people actually started to use their networks with a vengeance. However large and advanced a network is, there is only a finite amount of bandwidth available at any given time. Furthermore, there is a discernible delay between the request for information being made from a PC, and that data being delivered from the server or storage device. With complex cross-connections between PCs, servers and data storage devices, data was very soon traversing the entire network.

When data was predominantly flat text and numeric data, the processing power of the network server and the PC was sufficient to make any delays negligible. However, since the rise in popularity of the Internet, and the increased use of multimedia files, two things are happening. The available bandwidth is fast becoming swamped by this more complex data, and the time necessary to access that data from storage devices is increasing.

The Mongolian hordes' approach of throwing more bandwidth and more processing power at the network in the hope that this would solve the problem is sadly wrong. The problem doesn't lie with the processing power of the server any more, but with the nature of the data itself. Imagine the bandwidth of your network as a motorway. In an ideal world, you would be able to drive at the maximum speed from A to B. However, what if you had to negotiate winding country lanes before reaching the start of the motorway, only to face a similar situation at the other end? And what if the motorway was clogged up with juggernauts? This is what happens on a network. The juggernauts are the large amounts of data being ferried around the network, such as backups, using up bandwidth. And the country lanes are the interfaces with the storage equipment.

Most storage equipment still uses SCSI as the interface technology, but SCSI was never intended for use with a network ( it was originally designed to connect standalone PCs to peripherals ( and it's fairly slow. Even though SCSI has effectively doubled its bandwidth every year, it is still struggling to keep pace with network requirements. So, many of the problems are inherent in network design and implementation.

Unfortunately, with data being held all over the network, and with a lack of management in the majority of networks to ensure that access and storage are carried out in any sort of logical or efficient way, this situation doesn't look like improving in the immediate future. Indeed, it looks like it will only get worse, with the introduction of Gigabit Ethernet and the increasing popularity of ATM - both of which provide data traffic an order of magnitude above traditional network technology.

Of course, there are some measures which can be taken to reduce network traffic: an obvious one being to schedule backups to run overnight. However, even this has drawbacks. With the amount of data that needs to be backed-up, the backup can actually intrude into the working day. Secondly, we are moving towards a 24X7 business environment, where there is no downtime. It is clear that something needs to be done before we have complete information gridlock.

Ironically, the solution appears to be the reconsolidation of data. It is virtually impossible to go back to the old method of storing all the data on one huge storage device ( the time and effort involved are prohibitive. But it is possible to accomplish this logically.

One possible solution involves going back to basics, and considering the collection of storage devices as a network in its own right. Rather than having a mishmash of storage devices and file servers connected across the PCs and the network, they are effectively removed from the network and configured into a network of their own - a storage area network (SAN). The SAN connects with the PC network via a fast single point of contact.

The benefits of this method are immediate. Firstly, the single point of access can be made much faster by replacing SCSI with Fibre Channel, a more recent technology which is far more efficient at transferring larger and more complex amounts of data. Fibre Channel is more expensive than SCSI, but becomes financially viable if there is only one high-speed connection.

Secondly ( and more importantly ( the SAN becomes a black box of storage. To the PC network, it becomes irrelevant where or how the data is being held; it is sufficient to know that the data is there and available. This gives the data administrator considerable freedom in storing the data in the most efficient format, rather than in one designed to allow easy access from the PC network. The SAN can be upgraded or reconfigured without users even noticing. For example, storage can be added or removed, and RAID arrays can be dynamically altered without having to take down the PC network ( which is what happens in a traditional network. Furthermore, backups all take place within the black box of the SAN, so this traffic is removed from the network.

There are cost considerations as well. Storage isn't as dependent on data integrity, so the error-checking that is provided with an IP network isn't actually needed. That means that you don't need such an intelligent - and therefore expensive - network. A SAN switch is much cheaper than an equivalent Gigabit Ethernet or ATM switch.

One other important benefit is that SANs are ideally suited for heterogeneous networks. The days when every single product in your network, from server to switch to desktop PC, came from the same manufacturer are long gone. Today's networks comprise best of breed from various manufacturers. Although this means that each component is potentially the best there is in the market, it also means that there are compatibility problems, either with the hardware or the software. With a SAN, part of this problem is removed, thanks to the black box model; the PCs, servers and peripherals on the PC network don't need to know where the data is coming from.

There are two main hardware methodologies that comprise SANs. The first is specifically aimed at the LAN, and it can be considered as an "overlaid network". Although still a black box, the SAN connects with the PC network at every point, with fast interfaces - either fast SCSI or Fibre Channel - ensuring that all users can access all the data that they need. For this to work, the SAN has to be geographically close to the LAN.

This is the method favoured by storage stalwart StorageTek. StorageTek has been marketing its StorageNet product range for some time now, building on its existing product ranges and considerable expertise. Although it has some cost implications - for optimum use, you need a high-speed interface to the SAN on every network device that needs access - it offers the same degree of data access without the overhead of data traffic on the network.

The other SAN method is ideal for WANs. With this method, there is a single point of access between the PC network and the SAN, and all data is passed through it. This has a number of advantages: only one interface is required, which makes it cheaper. Also, your data can be geographically remote; this is ideal if you need to store your data away from the WAN, for example, if you are backing up to a disaster recovery site.

San Ltd is one of the leading vendors in the wide-area SAN market. The product of ATM leader K-Net's expertise and backing from developers, San Ltd's DiskLink product range offers a solution to wide area storage. This method is cheaper in terms of the number of high-speed interfaces required - there is only one point of contact between the PC network and the SAN - but the wide area link has to be an ATM channel for it to work at optimum.

It is interesting to note that these two methodologies are far from incompatible. The overlaid network, such as StorageTek's, ends where the remote SAN, such as the solution from San Ltd, begins, with the technologies interfacing at the network edge. As the technologies become more robust and cheaper, many companies will choose to go for the integrated solution that offers them the benefit of local and wide area storage.

Quite apart from the hardware considerations, there is also the software. Vendors such as Veritas are making headway in ensuring that, within the black box, data is correctly balanced across the SAN, taking account of the different types of storage that will comprise the SAN, such as tape drives, magnetic storage and magneto-optical arrays.

However, despite the enthusiasm for SANS, there are still problems that need to be overcome. Unfortunately, SANs are an idea which is too advanced for the current technology. Many of the hardware solutions currently on offer have to make do with existing product ranges, which have effectively been forced into working as a SAN. However, perhaps the most pressing problem hindering uptake is education. Most companies haven't even heard of SANs, let alone the benefits that they can bring. Vendors and resellers are currently embarking on a series of education and training programmes to let the industry know that SANs are out there.

Within the next couple of years, both StorageTek and San Ltd will see the technology finally catching up with the methodology. Alongside that, vendors and resellers will have completed their training and education programmes. Who knows, in five years' time, companies will be wondering how they every managed without storage area networks.

Craig Hinton

This was last published in July 1999



Enjoy the benefits of CW+ membership, learn more and join.

Read more on Networking hardware

Start the conversation

Send me notifications when other members comment.

By submitting you agree to receive email from TechTarget and its partners. If you reside outside of the United States, you consent to having your personal data transferred to and processed in the United States. Privacy

Please create a username to comment.