Often I hear about the changes iSCSI is going to make to the way enterprises handle their storage area networks (SANs). In one camp (the traditional network vendors) people are very bullish about the cheaper costs of Ethernet vs. Fibre Channel equipment and the more readily available skill set of Ethernet engineers. In the other camp, (mainly traditional SAN vendors), much is spoken of how unsuitable Ethernet is for storage traffic and how initial attempts to bring iSCSI to the enterprise have failed.
At first, the main issues for these arguments escaped me. Ethernet equipment is more cost-effective than Fibre Channel and Ethernet is easier to manage. Also, the basic underlying functionality of modern duplex fiber-connected Ethernet (now common in most data-centers) is almost identical to that offered in Fibre Channel equipment. So why the resistance to adopt?
A good deal of the surrounding confusion is being created by the vendors themselves. Most offerings designed to take vendors into the iSCSI space are all about convergence. For example, offerings from Cisco, Pirus (recently acquired by Sun Microsystems), Nishan and others have been all about offering both Ethernet-based iSCSI and Fibre Channel-based FCP commands in the same switch or device.
On its surface, this makes a lot of sense. Most SANs deployed today are based on Fibre Channel and in order to convert even a small portion of these to Ethernet it's much better if the two protocols can exist seamlessly in the same devices. Surely, this is the perfect way to bring Ethernet into the SAN arena.
However, in recent visits to many enterprise CIOs, it has become apparent to me that the sales pitch has expanded. Many companies are explaining that the real benefit of these consolidated devices comes in the ability not only to merge iSCSI and FCP on your SAN but also to take another step and merge you LAN and SAN traffic into one. Unfortunately, this is the very thing that's delaying the deployment of the very devices the argument was designed to sell.
Adding Ethernet-based iSCSI into your SAN is a relatively straight-forward proposition, given the ability to purchase the right equipment. SANs are kept secure and traffic flows in the main from one point to another with little need for control or quality of service provisioning. Thus, you can easily reap the benefits of reduced management and infrastructure costs.
However, if you actually try to merge the LAN and SAN, things take an order of magnitude step in terms of complexity. Surges in traffic can swamp your SCSI traffic, causing dropped connections and lost data. Lost packets are more common than most would think in these kinds of extreme conditions and SCSI data, unlike regular LAN traffic, is much more time-sensitive and often implements little in the way of guaranteed delivery. Also, SANs make almost no provision for security and SCSI data is for the most part unencrypted. (Although iSCSI does provide for this, few vendors can offer this encryption at any acceptable speed). As a result, merging your LAN and SAN makes illegal data access all the more possible.
My expectation is that Ethernet will replace Fibre Channel in the way it has replaced Token Ring, FDDI and ATM, as it makes life much better for those who have to keep data centers running if they have less to worry about rather than more, and if they can apply the same skills across a wider solution platform. However, those vendors backing this change would be better served to bite off a smaller piece of the networking pie (just SAN for now) rather than trying to bring about a LAN and SAN convergence -- a solution whose time has just not yet come.
Copyright 2002, Blue Arc Corporation.
This was first published in January 2003