Don't ask how I start down these paths --- last week it was smart antennas --- but today I ended up looking at Storage Area Networks and found an article you might find of interest:
k-net.co.uk
Adding storage to the mix
In this article, Andrew Rowney, Sales and Marketing Director of Cambridge-based, Storage Area Networks Ltd, aims to outline and dispel some popular misunderstandings associated with the relatively new concept of Storage Area Networks.
The recent surge in articles discussing Storage Area Networks (SANs) has been stimulated by strategic relationships and product previews announced by many quarters of the IT industry - from storage companies, server manufacturers and networking suppliers.
The objective of the storage area network is to overcome limitations inherent in the typical point-to-point architecture used to interconnect servers to their storage systems (disk, tape etc), and implement a more open 'any-to-any' architecture. This change has been compared to the revolution that took place when the Local Area Network (LAN) heralded the change from mainframe/terminal connections to today's client/server architecture.
The SAN is typically deployed to:
Improve application availability and performance - Figures released recently by Strategic Research show that 29.8% of downtime is due to disk failures.
Speed up backup and archive activities - Backup (and restore) systems should be able to take place alongside normal applications without impacting network or server performance.
Centralise and consolidate storage - Storage equipment is an expensive resource and must be used efficiently.
Implement high-availability disaster protection - Server and storage resilience.
Scope and geography
The first popular misunderstanding I'd like to address is that where a wide area, metropolitan area or local area network is defined in broad geographical terms, a storage area network is fundamentally defined by the traffic type it carries. A SAN must be considered as a virtual network using a combination of LAN, MAN and WAN technology to make storage systems available throughout the organisation. This might seem to be stating the obvious, but it is key to understanding and addressing the aims of a SAN. Of the four aims listed above, connectivity beyond the local data centre is fundamental to high-availability, archiving and disaster protection - setting up a disaster recovery centre adjacent to the primary data centre clearly makes no sense at all!
Network technology
Although the benefits of SANs have been discussed for a number of years now, it is only with the recent ratification of Fibre Channel standards that storage, server and network vendors have been able to agree on a widely-accepted solution. Fibre Channel is a layer 2 transmission technology which has much in common with FDDI - line encoding, CRC and optical bypass in arbitrated loop are lifted directly from the FDDI standard. In a SAN environment, Fibre Channel simply adds a degree of transparent low level transmission reliability for storage traffic carried in SCSI data format using SCSI protocols.
The second myth to be addressed is the unfortunate way in which the SAN is inextricably linked with Fibre Channel technology. We are all aware that a LAN can (and often does) consist of Ethernet, Token Ring, FDDI and ATM technology and that a WAN is never exclusively X.25, Frame Relay, ISDN or ATM. So why oh why do intelligent and respected members of the networking and storage industries equate SANs with Fibre Channel to the exclusion of anything else?
The installed base of storage devices predominantly uses the SCSI peripheral bus. Recent research by Quantum showed that the Small Computer Systems Interface currently accounts for over 95% of the market and that Fibre Channel will still only account for around 20% by the year 2000. Fibre Channel is, however, regarded as the migration path for high-end storage systems.
By-passing the server
Storage systems are typically accessed via the servers using tools such as FTP or NFS, which involve a significant protocol processing and I/O overhead. Using the server as a gateway to translate storage traffic for transmission at the network layer is very poor utilisation of sophisticated, multi-processor servers and introduces an unnecessary bottleneck. The third issue to address is the integration of storage traffic on the enterprise backbone without impacting network and server performance. The notion that the enterprise network cannot handle storage traffic is merely a reflection of the number of poorly designed, non-scalable backbones currently struggling to cope with existing traffic levels. Many organisations have invested in client and server systems, in application and database software, but have severely neglected the network infrastructure underpinning the enterprise.
With the availability of ATM and high-speed Ethernet technology, multi-gigabit enterprise networks are being deployed with sufficient bandwidth to cope with storage traffic without impacting existing traffic. In most cases there is no need to increase network complexity by building yet another network solely for storage - the key is being able to connect storage systems to the backbone without involving the server.
Integrating storage traffic
Any IT professional, from junior techy to IT director, will quite happily expound the virtues of simplifying their systems by reducing the number of platforms, applications and networks deployed throughout the enterprise. This policy has been widely implemented in the last few years with Microsoft, Intel PCs, Unix servers and TCP/IP dominating their respective market sectors. The cost-savings associated with reducing complexity and standardising on a simplified architecture are well understood.
This brings us on to the fourth and final misunderstanding to be addressed. Network convergence has resulted in the integration of voice and video traffic with conventional data traffic, and perhaps the merging of LAN and WAN or network and telecoms departments. What is needed is a cost-effective architecture where a common scalable LAN/WAN infrastructure has the ability to support a variety of traffic classes as virtual data, voice, video and storage networks. The interfaces to each virtual network should accept data packets, existing voice circuits, video streams and storage transfers.
Delivering storage networks
Where network professionals gather, they agree that strategically, the 'battle for the backbone' has been won by ATM - the only technology capable of converging data, voice and video traffic on a single LAN/MAN/WAN infrastructure. But can ATM also handle storage traffic?
Well ATM can certainly address the distance issue. Native SCSI can only reach 25m unless extenders are used with wide area links and Fibre Channel has a similar dependency on dark-fibre connections that telcos simply no longer supply. ATM is, of course, the only network technology that blurs the boundaries between the LAN, MAN and WAN.
Many organisations have deployed ATM as a backbone technology, building scalable multi-gigabit networks that can easily handle storage traffic. On the other hand, Fibre Channel Arbitrated Loop (FC-AL) is a shared media technology with the implied scalability and performance limitations - a fully loaded 1 Gbps FC-AL loop actually provides only 8 Mbps per device. Switching technology is required to provide the scalability demanded by modern IT systems, and only one transmission technology was designed to provide this - ATM.
ATM is, of course, the only network technology capable of integrating storage traffic with voice, video and data. The profile of archive traffic being transmitted between a high-performance disk system and a streaming tape library has similar characteristics to a Constant Bit Rate (CBR) video or voice stream. An application using disk-mirroring for real-time data resilience creates a bursty traffic profile related to the actual Variable Bit Rate (VBR) data being maintained.
Conclusion
Any organisation considering implementing storage area networks will be a sophisticated user of IT. The storage systems will almost certainly be SCSI-based and the bandwidth required will be in the same order of magnitude as the peripheral bus. The majority of applications driving storage area networks have a requirement to access remote systems over distances beyond the reach of local area network technology. The most efficient way to handle storage traffic is to involve the server as little as possible due to the network protocol overheads while aiming to integrate storage traffic with existing enterprise networks wherever possible.
The storage area network is a relatively new phenomenon currently enjoying a lot of attention, but before deploying the technology ensure you have a clear understanding of your application requirements, existing storage and network infrastructure and consider the issues outlined above.
Andrew Rowney,
<<<<
This guy spoke at the recent broadband conference here in San Diego and while I didn't hear him, a good friend in the industry did and was extremely impressed.
Links to the company and articles on their technology:
san.com
san.com
I'm certainly not pushing the company --- it's private --- but am curious about the technology.
Pat |
