Post Magazine

February 2010

Issue link:

Contents of this Issue


Page 18 of 47

S ince the beginning of computers, storage (disk and tape) has been separate from networking (Ethernet, wireless, etc).The two technologies use very different chips and cables. SCSI, Fibre Channel and FireWire have dominated storage, whereas Ethernet dominates networking. The goal of converged networking is to bring the two technologies together, allow- ing fast, low-latency storage protocols, like SCSI, to run over Ethernet. This will allow companies to deploy one set of switches, cables and adapter cards, which reduces complexity and cost. There are a few things that true con- verged networking requires: 1. Truly converged hardware. Customers should not have to purchase special cards, cables, switches or devices to use the new infrastructure. 2. Suppor t for all systems. Customers should be able to connect laptops, desk- tops and servers to the new infrastructure. 3. Devices should not require custom chips in order to operate on the new proto- col. Chip development is risky and expen- sive. Additionally, requiring non-standard chips in cards, switches and devices makes things more expensive and usually quirky and less compatible. W H Y D E P L OY C O N V E R G E D H A R DWA R E ? Ethernet and Fibre Channel are good ex- amples to use when answering this question. Ethernet is deployed on all computers. Whether it's a low-end laptop using an inex- pensive USB 10BaseT adapter, or a high-end server with a 10Gb optical adapter, there is some form of Ethernet on every system. As a result, Ethernet chips, switches, ca- bles and por ts are relatively inexpensive compared to similarly performing storage devices like Fibre Channel. Today, new build- ings are plumbed with Ethernet-compatible cabling and fiber as a matter of course. Fur ther, when Ethernet devices age, they don't necessarily become obsolete. Ma- chines are often redeployed as low-band- width servers, and switches can move from the core of the network out to the edge, often times simply to provide additional por ts for a lab or office. The equipment may age, but it generally doesn't turn into a boat anchor. W H AT I S F I B R E C H A N N E L OV E R E T H E R N E T ? Fibre Channel was originally designed to be "better" than Ethernet. The networ k was to be robust and never drop packets. The protocol was to be SCSI, but have an additional layer that would pro- vide for redundancy and scalable perfor- mance over multiple cables. These were all things that presented problems for Ether- net infrastructures. Unfor tunately, Fibre Channel chips never ended up widely deployed on laptops and new motherboards. As a result, those chips remained more expensive and the protocol was largely confined to the data center. This tends to keep pricing high. iSCSI attempted to make do with Ether- net and offer a solution for SCSI over Ether- net that was less expensive. However, the requirement for TCP to act as the error-cor- recting layer made the protocol CPU inten- sive and slow. Latencies tended to be high and customers would need to purchase special TCP Offload adapters if they wanted faster performance. In the past, many com- panies have tried to create offload adapters for IO protocols and the results have gener- ally not been good. Such adapters tend to be expensive, confined to ver y specific de- ployments and incompatible with each other. Additionally, when used with existing infrastructures, vendors had a hard time up- dating their special hardware to fix bugs and add features as required over time. Hard- ware is difficult to modify in the field! FCoE provides a new flow control mechanism to Ethernet to make it "loss- less," much like existing Fibre Channel hard- ware. The result is that FC packets can be easily passed over this network with no ad- ditional loss of performance. This also al- lows TCP to be left out of the mix. If pack- ets aren't going to be dropped regularly, there's no need for a complex error cor- recting mechanism like TCP. FCoE also adds simple quality of service capabilities to be sure that impor tant stor- age data can be prioritized ahead of less impor tant data, like Web browsing. Indeed, the new flow control mechanism even al- lows for unimpor tant traffic to be dropped so that FCoE storage packets are guaran- teed to arrive. There's only one hardware feature that FCoE adapters really need to have — the ability to checksum incoming and outgoing FCoE packets. This same capability exists in existing Ethernet hardware for TCP and UDP packets, and is very simple to add to new chips without spending a great deal of money. This allows for high-end systems to handle lots of FCoE traffic and checksum in hardware, while lesser devices, like laptops, can use the protocol but checksum their packets in software. It's also clear that networking vendors and Fibre Channel vendors are very serious about FCoE as a new direction. Recently, Brocade (a well-known Fibre Channel com- pany with no Ethernet products) purchased Foundry Networks (a well-known Ethernet company with no Fibre Channel products). It doesn't take a great deal of imagination to see where that marriage is going. FCoE will provide customers a fast, eco- nomical and simple infrastructure choice for future storage deployments and allow them to use a single Ethernet Infrastructure Fabric. Information technology will be cheaper, easier to understand and safer as a result. FCoE: next-gen SAN technology T E C H N O L O G Y February 2010 • Post 17 P O S T P O S I T I O N S Intel's X520 is a 10Gb Ethernet adapter that supports the FCoE standard. FCoE helps make Ethernet lossless, similar to Fibre Channel hardware. By STEVE MODICA CTO Small Tree Oakdale, MN

Articles in this issue

Links on this page

Archives of this issue

view archives of Post Magazine - February 2010