Analysis

posted on 07 August 2008 17:12

Cisco and QLogic would be very happy if you started evaluating Fibre Channel over Ethernet (FCOE) now, ready for a 2009 implementation. Why this sudden FCOE hustle and bustle? And what does it mean for other suppliers?

FCOE sends Fibre Channel (FC) protocol messages across an Ethernet network and is a means of converging Fibre Channel storage traffic and Ethernet traffic onto a single Ethernet network, thus saving cash and cabling complexity. The longer term idea is that, over time, a Fibre Channel SAN fabric can be superseded by an equivalent Ethernet fabric.

This will need the same kind of FC SAN superstructure, such as zoning and virtual SANs, to be erected on Ethernet-accessed storage as currently exists on FC SAN fabrics. It means having intelligent Ethernet switches to do this. Both Cisco and Brocade are evolving their FC SAN switches into data center switches with both SAN Director capabilities and Ethernet capabilities.

A data center-class Ethernet needs to have a predictable latency and not lose packets, things that Fibre Channel is well-known for. This is why there is a Converged Extended Ethernet (CEE or Data Center Ethernet - DCE) - standards effort. It is expected to deliver a standard in 2009 at the earliest.

Brocade expects servers to start using FCOE interface cards, a combination of FC Host Bus Adapter (HBA) and Ethernet Network Interface Card (NIC) known as a Converged Network Adapter (CNA) to show up servers in 2010 with an implied general adoption starting in 2011.

However Cisco has a much faster time table with server CNA use in 2009 and general adoption starting as early as 2010. Why does it have this much faster FCOE ramp in mind?

There's product to sell of course. QLogic has is 8000 Series Converged Network Adapters CNA) ready to ship now. Cisco has its Nexus Ethernet switch product ready. The two are co-operating in trying to kick-start a FCOE ecosystem of suppliers and approved product. In fact, QLogic and Cisco are functioning as a 2-member FCOE trade association with Emulex co-operating. These suppliers are keen to have their customers dance the FCOE tango, buy product and deliver revenues.

There's market competitors to outpace. Cisco will be very pleased to have its Nexus Ethernet switches in place and being piloted before Brocade's DCX product. Seeing Brocade talk about lack of FCOE product and the need to wait for a CEE (or Data Center Ethernet - DCE) - standard before talking implementation is probably music to Cisco marketeers' ears.

Implications

An implied FCOE marketing message is: "Don't extend your Fibre Channel SAN; it's a dead end. Look at FCOE because Ethernet is the future with Fibre Channel being a cul-de-sac in waiting." 

There are quite strong implications for vendors with products at the heart of the FC SAN fabric world: EMC; HDS; and IBM.

With EMC struggling to have InVista, its SAN storage virtualization and management product that runs in a SAN director, developed and shipped in a reasonable time frame, there being only a couple of hundred of pilot-like installations at present, it faces having to convert the software to running on an intelligent data center switch and it may, let's stress the 'may' here, decide not to undertake the effort.

IBM's SAN Volume Controller is the leading SAN virtualization and management product. For it to transfer to the FCOE world will be quite an effort but one that its product management people are probably already looking at. They will have to talk to Cisco and Brocade as they are aiming to be the two dominant FCOE 'switch' suppliers.

We might potentially think of an SVC Lite product to work with the first FCOE networks if, as is likely, their functionality is less than that of current FC fabrics and their directors.

For a supplier such as Hitachi Data Systems, with its USP virtualizing FC SAN storage controller functioning at the edge of the FC SAN fabric, its development tasks with FCOE is probably simpler than that of IBM with the SVC.

Simply put, add an FCOE target interface to the HDS kit and everything should work as before. It doesn't have the fabric director interface that IBM's SVC has to worry about.

Complication

A complicating factor here could be that native FCOE storage arrays will emerge; NetApp is said to be ready to do this next year. EMC says its CLARiiON CX4 I/O architecture can readily have an FCOE I/O module dropped in when ready.

Such an array would have a kind of reverse CNA interface with FCOE traffic coming in and being converted to the array's internal network messaging.

This leads to an additional convegence possibility.

In the FCOE world we have Fibre Channel traffic travelling over Ethernet to access block-level storage as currently used in FC SANS.

But we currently also have SCSI commands, quite similar to Fibre Channel in some ways, wrapped up in TCP/IP packets and travelling over Ethernet to iSCSI block-level arrays. TCP/IP is used to overcome the limitations of Ethernet concerning, for example, packet loss.

Once we have a non-packet-losing Ethernet than iSCSI becomes a protocol too far; it becomes unnecessary. We may well expect FCOE adoption to be accompanied by iSCSI withering away.

How will iSCSI storage suppliers respond to FCOE? By sticking a native FCOE interface on their arrays of course. Why wouldn't they do that?

Thus FCOE is a threat to TCP/IP offload engine (TOE) card suppliers such as Alacritech and Chelsio.

FCOE promises - threatens - to unify the FC SAN array and iSCSI SAN array markets into a single block-level access market, making storage array product differentiation harder to achieve since the iSCSI vs FC tag goes away, and cutting away the foundations of the iSCSI market.

Where are we now with this line of thought? Fibre Channel SAN fabrics are on the way out. Joining them in the exit room are iSCSI SANs. At such a time Ethernet would carry all block-level storage traffic and be the route to access file storage on NAS (Network-Attached Storage) arrays as well. We're looking at a kind of grand storage networking unification here. Ethernet will rule OK, and then some.

This seems so neat and so logical it will probably never happen.

^{[Chris Mellor.]}