Essential guide to datacentre convergence

Datacentre convergence may take a great leap forward with FCoE, but few organisations will need this sort of data capacity right now

Datacentre convergence may take a great leap forward with FCoE, but few organisations will need this sort of data capacity right now

The convergence of storage area network (SAN) and local area network (LAN) architecture onto a single architecture will offer companies at least the potential to reduce costs and increase storage and speed. Potential, because previous LAN/SAN convergence attempts based on iSCSI and Fibre Channel over IP (FCIP) have achieved, at best, only modest success. However, Fibre Channel over Ethernet (FCoE) may tip the balance by helping reduce maintenance requirements.

LAN/SAN convergence and FCoE is only one element in a wider trend towards converged datacentre architecture, where the IP LAN and Fibre Channel SAN are converged together with server backplanes to minimise the amount of equipment in the datacentre, thus saving space, power and management resources.

"The way we see it is that this is going to reduce end-user costs from a capex and opex point of view, because if you start putting data and networking into a single cable, less hardware and maintenance is required," says Eugene Berger, solution architect at HP UK & Ireland.

"With the adoption of 10 Gigabit Ethernet (10GbE), we've reduced the number of physical cables from multiple one Gigabit links to one 10GbE link, but there is still an FC cable for the SAN. So the natural adoption is to have FC and Ethernet and look to converge those two."

Emir Halilovic, programme manager for networking and infrastructure at research company IDC, agrees: "There are often latency and application performance problems [in datacentres], and then the cost overhead of keeping three different teams managing three different parts of the datacentre. Adding additional capacity to all three is not going to make you a happy camper, with the demands being placed on datacentres [getting] higher by the day."

The extent to which any one organisation has, or will, hit those pain points is anyone's guess. For the moment, it seems that only telcos and service providers involved in the hosting business, which are moving large virtual server workloads between servers and storage resources, are even close. But Halilovic suggests that nobody can afford to be complacent and that an upgrade requirement is inevitable in the long run.

Essential guide to datacentre convergence

Datacentre convergence may take a great leap forward with FCoE, but few organisations will need this sort of data capacity right now

“Some larger operators already have datacentres facing these issues. You either follow the current path with the notion that you are going to hit the wall in a short while or take a capital expenditure hit with a complete overhaul,” says Halilovic.
“We are not talking about a software solution, or repurposing or building on top of current infrastructure. In most cases, it is a serious rip-and-replace.”

However, Trevor Dearing, head of enterprise marketing at Juniper Networks, believes that inadequacies within existing legacy datacentre architecture is only the initial driver. “Phase one of that adoption is where they see the benefits of convergence in the server rack and the cabling, and the fact they have to go to 10GbE to support virtual machines, so they may as well go to network adapters as well, for power and space savings,” he says.

“Phase two will be application-driven, where apps such as Skype or Fusion IO running on solid state disks demand a lot of movement. This will change the way we build datacentres: less of a centralised and more of a distributed storage model.”

Converging in a crisis

As ever, cost is an issue, particularly in a market where many of the countries you would normally expect to lead technology adoption face a continuing economic crisis.

Full convergence requires new switches, cabling and server adapters, if not new servers with converged network adapters (CNAs) built in. The top-of-rack switches offering dual-purpose Ethernet and FC ports that can be plugged into existing datacentre networks are nothing more than a low-cost starting point. They offer no benefit other than a component that can be re-used when the convergence revamp takes place.

“There is no cost premium between a SAN-enabled switch and a 10GbE switch as we are trying to push it [LAN/SAN convergence] forwards, so we would like to think our solutions are relatively inexpensive and the total cost of ownership [TCO] is good,” says HP’s Berger.

Uncertainty about standards is another hurdle, says Halilovic, as is entrenched support and expertise for FC SAN technology. An FC roadmap that aims to deliver transmission speeds to 32Gbit/s in 2014 and 64Gbit/s equivalents when the market demands is also a problem (FCoE is following the Ethernet roadmap and is currently delivering 10Gbit/s, but with 40Gbit/s a few years away and 100Gbit/s further down the line).

“First and foremost, there are barriers within organisations in terms of how things have worked historically with separate networks, separate storage and separate servers, and that is hard to break,” he says.

So can FCoE succeed where FCIP, iSCSI and a host of other technologies have previously failed? Can it persuade potential customers that it is worth building a single network to connect all the resources that separate LANs and SANs now?
“FCIP in effect morphed into what FCoE is now. When iSCSI came out, there was not the same extent of adoption for 10GbE, but now we have 10Gbit/s Ethernet and we are moving towards 40Gbit/s IP pipes, so you have more bandwidth to play with and a big enough pipe to converge the two,” explains Berger.

“FCoE has more support from vendors, including Cisco and Brocade, than FCIP, and converged network adapters are built into the fabric,” says Dearing. “People think FCoE will be as simple as Ethernet, even if the reality is that they are all iSCSI over something. It feels like there is more acceptance across the board for Ethernet, whereas iSCSI was driven by mid-tier storage vendors. If you are going down the convergence route, it may be easier to end up with non-storage people building non-Ethernet storage networks.”

Vendor race to the FCoE line

Fibre Channel over Ethernet (FCoE) may not be the only technology component in the converged datacentre architecture being put forwards by major networking, server and storage companies, including Cisco, Juniper Networks, Brocade, HP, Oracle and IBM. But it forms the majority constituent of its foundation: local area network (LAN) and storage area network (SAN) convergence.

FCoE is a storage protocol that enables data traffic from Fibre Channel (FC) devices to run directly over Ethernet networks, converging storage and IP protocols into a single cable and interface used by servers, storage arrays, networks and other devices within the datacentre.

The technology was originally defined in 2009 by the Institute of Electrical and Electronics Engineers (IEEE) in its FC-BB5 specification. It requires that FCoE-enabled switches support datacentre bridging (DCB) traffic management standards.

Proposals for its successor – FC-BB6 – are subject to the usual vendor squabbling about what it should and should not include. HP solution architect Eugene Berger says that, like its competitors, HP is busy “investigating” the proposals and sitting on the relevant IEEE boards. But the delay is seen as something of a problem for all interested vendors.

“We need to all use the same FCoE protocols. Otherwise, we risk customer lock-in, especially converging the network and the SAN,” he says.

Essential guide to datacentre convergence

Datacentre convergence may take a great leap forward with FCoE, but few organisations will need this sort of data capacity right now

Similar arguments rage about the necessity of Transparent Interconnection of Lots of Links (TRILL), a proposed standard by the Internet Engineering Task Force (IETF). It is designed to support Layer 2 multi-hop routing across the converged architecture as an alternative to the bandwidth-limiting spanning tree protocol (STP) when scaling out to larger datacentre networks.

“Not having defined standards is detrimental to this technology [FCoE] gaining traction – that is very much a given,” says Emir Halilovic, program manager for networking and infrastructure at research company IDC. “There are different approaches for TRILL and back bridging, and we are in the early stages of development for those technologies.”

In other respects, the converged platforms on offer are similar. Cisco’s UCS is built on a device manager, interconnects (20 and 40 port FCoE switches), a fabric manager to handle storage networking across the Cisco SAN and unified fabrics, as well as extenders that connect the UCS blade chassis to Ethernet switches, and Emulex and QLogic converged network adapters (CNAs) for servers.

“Cisco’s UCS is basically an all-in-one device that has gained lots of traction, though we do not have the ability to directly measure sales because the company’s reporting is somewhat obscure,” explains Halilovic.

“It [Cisco UCS] does not really fit well with big datacentre environments, but it is a rather good general solution – perhaps better for mid-size or smaller datacentres.
“There are also approaches to flattening the architecture, with the fabric type of solutions from Juniper, HP and FC specialists such as Brocade, which are bulking up on the Ethernet part of the bus.”

Juniper’s QFabric is built around a single ‘logical’ switch that connects network, storage and server resources. The first component was a top-of-rack 10GbE switch offering 64 ports, including 12 dual-purpose Ethernet and FC ports, which acts as the QF/Node component. It was first launched in February 2011.

Juniper followed it up with the QF/Interconnect, which provides the
high-speed chassis, and the QF/Director server-based common management and control software, in September.

The company is partnering IBM, NetApp, CA Technologies and VMware for the QFabric technology. And while some of the connections within QFabric are proprietary, outfacing connectivity at its perimeter use standard interfaces, meaning non-Juniper equipment can be plugged in.

Juniper says it constructs its QFabric differently from the other vendors “by making it look as though you are always on the same switch, whether you are doing provider backbone bridging, or the protocol is FCoE or iSCSI, providing all the necessary end-to-end flow control and lowering latency,” says Trevor Dearing, Juniper head of enterprise marketing.

The technology itself has been enough for Cisco to present a spoiler campaign aimed at pointing out its deficiencies, a sure sign that it is worried.

However, it is not just about the hardware: all the vendors in the converged datacentre architecture space are pushing the benefits of the reduced overhead associated with merging two previously separate networks. But that benefit is only as good as the management software that supposedly cuts down on the administration burden.

“Giving visibility over separate networks from a single management console is crucial to making this convergence a living thing,” says Halilovic.

“If you look at what HP, Juniper or Cisco are offering, the focus is on moving from hardware devices into software that links entities to provide a unified architecture that makes everyone’s life easier.”

Essential guide to datacentre convergence

Datacentre convergence may take a great leap forward with FCoE, but few organisations will need this sort of data capacity right now

Brocade UK manager Simon Pamplin agrees. “We provide a single pane of glass management tool through to the Ethernet network and the SAN, with plug-ins and APIs, which connect to other umbrella management tools that the organisation may already have,” he says.

“One of the biggest headaches is moving a virtual instance from one physical device to another physical device, because if you do not get the ACL [access control list] right on the receiving port, the migration fails and you have to start the whole process over again. What this does is automate what was previously a manual process.”

Convergence for reluctant buyers

Fibre Channel over Ethernet (FCoE) and datacentre convergence would look like a marginal technology struggling to make its mark on a largely disinterested audience if the success – or otherwise – of any one technology could be measured simply by examining current real-world usage within customer deployments.

None of the vendors to which Computing spoke was able to put forward a reference customer willing to speak about their FCoE deployment. However, Cisco claims thousands have installed its Unified Computing System (UCS) converged datacentre platform, and rival vendors insist many firms have taken the first steps by purchasing top-of-rack Ethernet switches that can be added or reconfigured to support FCoE at a later date.

This slow-burn adoption mirrors analyst forecasts. In its bi-annual SAN five-year forecast report, networking market research firm Dell’Oro Group predicted 2011 would be the year when revenue from FCoE storage equipment would outgrow that from FC for the first time. But the underlying figures show how healthy the market for FC still is, and how small FCoE remains by comparison: FCoE will grow by $300m (£194m) – from $274m in 2010 to $583m in 2011 for FCoE, versus growth of $180m from $2.68bn in 2010 to $2.86bn in 2011 for FC.

“IDC forecasts do not see tectonic changes in terms of shipments or revenue for FCoE components, but more steady growth from a very low starting point,” confirms Emir Halilovic, program manager for network and infrastructure at research firm IDC.

Having been first to market with its FCoE-based UCS in 2008, it can come as no surprise that Cisco is leading the way when it comes to shipments. An official blog written by Cisco’s Soni Jiandani in May 2011 claimed it had 5,400 customers and an annual sales run rate of $900m for product orders at that time.

In 2010, Loughborough University adopted Logicalis’ Cooperative Cloud platform as the basis for a JANET-connected hosted cloud service that it plans to deliver to its own staff and other universities. The architecture will be based on a mixture of Cisco UCS and Catalyst switches.

Loughborough University IT director Phil Richards describes the cloud service as “a key milestone in validating our decision not to spend money regenerating our old datacentre”, saying UCS formed the basis of a new-build project.

Meanwhile, five customers have Juniper’s QFabric in beta test, the company told analysts in September. Canadian telco Bell Canada was the first to go on record saying it intends to deploy Juniper Networks’ QFabric single-tier datacentre architecture as the basis for the hosted cloud services it will be delivering to its corporate customers.

Gradual implementation

There are no details as to how many servers, storage arrays or network nodes Juniper’s QFabric will connect, what it replaces (if anything, the release hints at a new-build rather than rip-and-replace project), how much network capacity it will provide, or even when it will be installed.

Bell Canada has said only that it will “unify the server, storage, appliance and network elements of its new hosting environment” and that it chose Juniper over rival networking vendors because QFabric “demonstrated low latency and the ability to scale gradually over time without disruption to daily business operations”. Read through the lines and this is precisely the sort of gradual implementation that appears to be the only way in which vendors can actually sell FCoE platforms.

“Whether customers are implementing it now or building it in for the future is up for discussion. Some people are trialling it but others are saying they are not adopting it because they already have very large FC implementations,” claims Trevor Dearing, Juniper head of enterprise marketing.

UK telco Cable & Wireless is using Cisco UCS as the foundation for its cloud service hosting platform as part of its next-generation network (NGN). However, it is unclear if the infrastructure is being put into existing or new facilities.

While Brocade is unable to name any reference customers of the technology, the company’s UK and Ireland systems engineering manager, Simon Pamplin, remains adamant that many have signed up – if only purchasing the top-of-rack component in anticipation of a later migration to a fully converged network.

“We have had take-up from media companies to those running HPC, higher education and financial trading companies as they prepare to move over. They are looking ahead for future proofing,” he says.

HP tells much the same story. “Our customers are not ready for adoption now, but we want to be ready when LAN/SAN convergence and FCoE becomes widely adopted,” adds Eugene Berger, solution architect for HP UK & Ireland.