Storage revolution in waiting



Building a corporate-wide storage area network (SAN) is no minor undertaking. Take the IT operations of Swedish automotive manufacturer Volvo.

At its Gothenburg site, the organisation has built "one of the largest SANs in operation" that connects 46 terabytes of storage attached to over 1,000 servers – Unix machines, Windows systems, NetWare servers and IBM mainframes.

But that kind of set up does not come cheap. Today's SANs require the installation of a whole array of specialist technology: a dedicated fibre channel (FC) network, storage virtualisation software, and fibre channel-specific hardware including storage network cards, ports and switches. To run that, organisations need to hire staff skilled in those technologies. In short, building a SAN is not something every organisation can afford or justify.

For years, however, an elusive – and much more standard and potentially low-cost – alternative has been on the horizon: storage-over-IP (SoIP). Storage networks that use the Internet Protocol can transport ‘block-level' data – the most widely deployed mechanism for handling business applications data – over Ethernet networks. To date, though, IP storage networks have been almost exclusively used for straightforward data movement such as file transfer.

The allure of SoIP is clear: organisations will be able to make use of their existing IP networks to implement a SAN. "The main benefit of SoIP is simple: it all boils down to cost," says Nick Tabellion, chief technology officer at storage software supplier Fujitsu Softek. This is because nearly all organisations already have an IP network – and the skills to run it – in place.

SANs doubt
But while IP-based storage networks have been hyped for half a decade, products have only just started to appear – and many of those could best be described as ‘proof of concept'. For example, IBM took the plunge in 2000 with its Total Storage Internet Protocol Storage 200i system (see University challenge). Another offering is from US-based niche storage hardware vendor 3ware which has partnered with Ethernet switch supplier Extreme Networks to release a SoIP product set.

Other vendors have been slow to follow. Most of the largest suppliers in the storage market, including EMC, HDS and Hewlett-Packard, stress they are waiting for improvements in SoIP technologies before trying to open up the market.


Geoff Cole, IBM: “By processing data in software iSCSI is nothing like as fast as FC.”


But what are those shortcomings? At present, the main weakness is the relatively poor performance of SoIP compared to fibre channel. "If you look at what IP can deliver today [in terms of transporting block-level data] it is pretty low end," says Trevor Williams, storage network solutions manager for Europe, Middle East and Africa (EMEA) at storage systems supplier Hitachi Data Systems.

A good measure of that performance weakness is governed by how SoIP manages data. The key enabler of SoIP for block level data is an emerging technology called the Internet small computer system interface (iSCSI) protocol. ISCSI is a storage networking standard for transporting SCSI-based information between storage devices over IP networks. The original SCSI protocol itself dates back to the 1960s when systems and software giant IBM developed input/output connectors for its mainframes that allowed them to talk to several different peripheral devices simultaneously.

The crux of the performance issue is how iSCSI processes data over IP. While the FC protocol processes data at the hardware level, with iSCSI "the processing of the IP is done in the software on the client or server that is generating the [storage] transaction," outlines Geoff Cole, a member of IBM's EMEA network storage advisory group.

Jim Rothnie, chief technology officer of EMC, adds: "Right now because all of the [storage] implementations [for IP] are done inside the software, if you try to execute that with a storage input/output load and at gigabit speeds – which is what everyone wants to do – then you practically consume all of the processing cycles that are available on either the server platform or the storage platform. The consequence of doing this work in the server's processor is that there might be only be around 15% of the server cycles left to do other work."


Jim Rothnie, EMC: “Storage-over-IP may not necessarily be cheaper than FC.”


Cole reinforces that point. "FC is more expensive than IP because organisations have to use specialised hardware – host bus adapters – to connect devices into the SAN. But as all of the processing is done in hardware, FC is very, very fast."

Toe hold
To address this specific weakness, a new technology has emerged. US-based specialist hardware vendors including Adaptec and Alacritech are providing intelligent host bus adapters, often referred to as TCP/IP offload engines (TOEs). These adapters aim to offload TCP/IP processing from supported clients, says IBM's Cole. "The issue is that the TCP/IP processing needs to be offloaded from the server to the [offload engine]," echoes Olaf Swantee, director of the enterprise storage group for EMEA at systems vendor Compaq Computer.

"Without these offload engines in a SoIP set up, you cannot do anything that involves a significant I/O load," concurs Rothnie at EMC.

However, while TOEs will help performance, Rothnie stresses that SoIP may not – in the short-term – actually be much cheaper than FC. This is because the price of a storage port for the next generation of high-speed gigabit Ethernet networks is not vastly different to the commonly deployed gigabit FC port. The real cost savings will be for small organisations deploying SoIP over standard Ethernet networks that offer data transfer speeds of 100 megabits per second (mbps). This is because the cost of a 100 mbps Ethernet port is about five times lower than the commonly deployed gigabit FC port.

Rothnie says, it will be only with the arrival of 10-gigabit Ethernet networks that "there could be a big difference [in price] – plus, the cost of installation is also likely to drop faster for Ethernet, which would be a big plus for iSCSI."

Predictions about when it will be possible to run a SAN completely over IP vary, but Swantee at Compaq expects such an alternative in the first half of 2003.

Fibre intake
In the interim, storage vendors are looking to offer products that use IP networks in a way that does not require iSCSI using IP to connect together large numbers of geographically dispersed FC-based SANs.

To link SANs over long distances, IP is critical because SANs today are limited to around 16 to 20 kilometres, or up to 60 kilometres with additional intermediary devices. IP has no such limitations.

Still, analysts remain positive about broader implementations. "I think over the next couple of years, IP storage will [become] a lot more mainstream," says Galen Schreck at Forrester Research. "There is not a lot of point – and so much more added expense – in having two types of storage network [namely, IP and FC]." As that might suggest, IP promises – ultimately – to be the universal storage data transport.


University challenge
Most universities would acknowledge that until recently they did not have a high requirement for a sophisticated storage network. This was certainly the case at the University of Tubingen, one of Germany's largest academic centres, which relied on file-based direct attached storage devices on its local area network.

But the launch of its web site at the end of 2000 and, more specifically, the provision of an online streaming video library service, brought the need to transport larger volumes of data across its whole network. The university considered the option of a separate, dedicated fibre-channel (FC) network but baulked at the cost, says Werner Dilling, the university's systems manager. This would have meant purchasing FC network cards for around 20 additional Linux servers – cards that are around three times more expensive than those required for its Ethernet gigabit network, says Dilling.

The university eventually chose IBM's Total Storage IP Storage 200i because it could provide storage area networking over the existing IP (Internet Protocol) network.

Dilling says the 200i has been able to meet the high bandwidth requirements of transferring large volumes of online video data between machines, so that students and academics can access these teaching aids either as a whole or in chunks of video. That got underway in October 2001 and Tubingen now wants to be able to use the universal availability of IP to provide storage access to other universities. "The main point is we have centralised the storage administration," says Dilling – without having the expense of FC.



Avatar photo

Ben Rossi

Ben was Vitesse Media's editorial director, leading content creation and editorial strategy across all Vitesse products, including its market-leading B2B and consumer magazines, websites, research and...

Related Topics