News

Data center network brawl

Converged platforms that combine servers, storage and networking into a single system with common management tools are the future of enterprise data centers, if vendor claims are to be believed. Proponents argue that these platforms can lower costs, support better utilization of virtualized resources and speed application deployment.

But what really separates the various offerings — from Cisco's Unified Computing System to those out of HP, IBM, Sun and others — from each other? Here's our cheat sheet:

Cisco's Unified Computing approach: Have we got a blade server for you!

The pitch:

A unified data center fabric with a lossless, low latency version of Ethernet, which Cisco calls Data Center Ethernet. FibreChannel and other current and legacy storage and server access protocols would run over this ruggedized Ethernet and resources would be allocated virtually via service profiles, reducing cabling, server adapters, switches, space, power consumption and management disciplines. Cisco says it has 400 customers running FibreChannel-over-Ethernet (FCoE) in production mode. At the systems level, blade servers are optimized for this environment, virtual machine-enabled and tightly interwoven with fabric switches and storage.

Recent news:

Cisco in January unveiled UCS, Cisco's scheme for how a converged data center, including at the blade server level, should evolve. The embodiment of the company's network-centric view of resource interconnection and accessibility. It also entered the blade server market, long the domain of partners IBM and HP.

What it means to the enterprise:

Cisco's unified computing approach offers a way for enterprises looking to cut operational expenses in their data centers while scaling resources. But it introduces a new player to a mature market — blade servers — served for years by more established players. Enterprises might find Cisco's view refreshing or invasive.

Where it falls short:

Enterprises may not be confident acquiring mission-critical servers from a relative newcomer to the market. Plus Cisco's approach might be too all-encompassing for them. Would they be putting too many eggs into the Cisco basket? Cisco is alienating longtime data center partners IBM and HP with UCS, and that could be a red flag to buyers.

HP's BladeSystem Matrix: Jump into the physical and virtual resources pool

Data center convergence pitch: By weaving together numerous hardware and software pieces, HP says it combines physical and virtual resources into the same management system, giving IT pros a self-service portal to quickly design, deploy and optimize applications. Instead of “islands” of IT in which separate teams manage different equipment, HP says the data center should be treated as one large pool of computing, network and storage components.

Recent news:

HP in April unveiled BladeSystem Matrix, a converged software, server, storage and network platform that starts at $150,000.

What it means to the enterprise:

The BladeSystem Matrix starter kit gives customers a full rack with ProLiant blades; a StorageWorks array; HP Virtual Connect Flex-10 Ethernet and 8GB Fibre Channel modules; and Insight Dynamics software to manage and automatically provision resources. Customers have a choice of VMware, Microsoft and Citrix hypervisors. A modular architecture lets customers scale up without limits.

Where it falls short:

The Matrix offers a 15% price reduction compared with buying all the components separately, although there is nothing new about any of the individual hardware components, Gartner analyst Andrew Butler says.

“Existing HP BladeSystem users who want to extend the footprint, but simplify ordering and administration, should consider Matrix,” Butler writes says in a research note. “But users should challenge HP to demonstrate how existing BladeSystem components could be integrated into a Matrix management framework.”

Liquid Computing: Anyone ever heard of us?

Data center convergence pitch: Founded in 2003, Liquid focused on high-performance computing until last year when it unveiled a converged platform utilizing technology from partners such as Microsoft, VMware and NetApp. Like HP, Liquid Computing promotes its ability to natively support virtualized workloads as well as any bare-metal application.

Recent news:

LiquidIQ 3.0, with Intel's Nehalem processors, was announced at the end of March. Pricing starts at about $75,000.

What it means to the enterprise:

Contained in a 26U chassis, LiquidIQ is designed for rapid provisioning of servers and networking, low power use and reduced floor needs. Liquid says “all core application infrastructure including the server, network, storage, operating system, and associated physical and virtual connections are represented as logical components and can be managed from LiquidIQ's Web-based interface or through existing enterprise management systems.”

Where it falls short:

Liquid Computing has the most advanced products in this market, but is struggling to make a name for itself, says Yankee Group analyst Zeus Kerravala.

“They entered this market first, although no one seems to know who they are,” Kerravala says. “They've got the least amount of brand recognition in this space, but the most mature product.”

Amid converging with Oracle, Sun makes its converged data center case

Data center convergence pitch: Sun says it can eliminate the need for a Cisco switch by developing its own switch technology that works with blade servers and reduces the complexity of network infrastructure, cutting the need for cabling by a factor of 10.

Recent news:

The company in April unveiled the key component of this strategy, the Virtual Network Express Module, a blade network switch priced at about $5,000.

What it means to the enterprise:

The NEM is a piece of silicon that slides into a blade chassis and does the job of an aggregation switch, managing traffic between the blades and a central 10-Gigabit Ethernet switch. “Instead of having a managed switch from a third-party vendor, we developed a piece of silicon that makes all the blades think they are talking to their own dedicated interconnect, so you're eliminating Gigabit-to-10 Gigabit Ethernet,” says John Fowler, executive vice president of Sun's systems business.

Where it falls short:

Sun's blade network switch is just a network connectivity product, and it eliminates the need for a Cisco-type switch in only a small portion of the data center, Kerravala says. “It's a very limited part of this market. Sun has a lot of work to do in this space,” he says.

Nathan Brookwood, principal analyst with Insight64, called the NEM “very impressive” but noted that it may provide fewer fine-grained management capabilities than a third-party switch, such as the ability to prioritize traffic from different applications.

The NEM is also designed only for Sun's blade servers, which lag behind IBM and HP in market penetration. With Sun subject to a pending acquisition by Oracle, customers should carefully evaluate the viability of any product under a combined Oracle/Sun company.

Dell embraces partners, rejects one-size-fits-all approach

Data center convergence pitch: Rather than produce a single converged platform as Cisco, HP and Liquid Computing have, Dell touts its ability to partner with numerous vendors to offer customers a range of options for servers, storage and networking, and sell it all in one package.

Recent news:

In January, Dell announced a deal to resell Cisco's Nexus 5020 data center switches, bundled with Dell servers and storage, and let Cisco management tools work seamlessly with Dell equipment. While partnering with Cisco, Dell has also tried to differentiate itself from that company, saying Cisco offers a one-size-fits-all approach that doesn't take the varied needs of customers into account.

What it means to the enterprise:

Customers who want Dell blade servers can work with the vendor's services team to create an integrated system that meets their specific application needs. “We take the best of Cisco, the best of EMC, the best of VMware and the best of Dell to enable our customers to have the right solution,” says Rick Becker, Dell's vice president of software and solutions

Where it falls short:

While Dell has an impressive blade server offering, the vendor has very little in-house network expertise and thus has to rely heavily on other vendors, Butler says. A true “fabric-based architecture goes a little further,” he adds. “It's really converging in one single device the server, storage and networking technology, and I don't see Dell being close to delivering that kind of capability for the simple reason that they can only do it with massive collaboration from one or more vendors.”

IBM beefs up its blades

Data center convergence pitch: Big Blue hasn't offered a converged storage, server and networking platform along the lines of Cisco's UCS, but has focused on integrating many types of network products into its blade servers.

Recent news:

Since the end of April, IBM has announced a deal to resell Brocade's Foundry Networks switches and another deal to integrate QLogic and Emulex network products with IBM blade servers. IBM has also partnered with Blade Network Technologies to provide switches for IBM blade servers and a top-of-rack switch for IBM's iDataPlex system.

What it means to the enterprise:

Because IBM hasn't detailed plans for a single, converged server, storage and network system, this is open to speculation. Kerravala expects IBM will release a product to compete against Cisco and HP's BladeSystem Matrix, but says it might focus on open standards, building integrated products with pluggable components from a list of third-party vendors. IBM also has a large services organization that could help customers weave different network, storage and server products together.

Where it falls short:

IBM is strong on the storage and computing side, but its unified computing story is piecemeal on the whole, Butler notes. “On one hand IBM's strategy is very collaborative, but it makes it extremely dependent on alliances and relationships with customers that have different agendas.”

Alliances with Brocade and other vendors give IBM a good start, but “they haven't put it together yet,” Kerravala says.

Brocade plus Foundry equals…?

Data center convergence pitch: FCoE and Converged Enhanced Ethernet (CEE), eventually, to run legacy storage protocols over a lossless Ethernet switching fabric; but a legacy of FibreChannel storage-area networks (SAN) to migrate. Acquired Ethernet switch vendor Foundry Networks to converge storage networks onto Ethernet. Partnerships with IBM and HP and others to link IT resources to the FC/FCoE/CEE fabric. IBM and HP also resell Brocade LAN and SAN products.

Recent news:

The Brocade 8000 SAN switch debuted in April, a top-of-rack FCoE device with 24 ports for 10GbE CEE and eight ports for Fibre Channel at 8Gbps speed; and single and dual-port Converged Network Adapters for 10Gbps FCoE connectivity from servers to SAN fabrics and LANs. Pricing and availability undisclosed but Brocade doesn't see a mainstream market for FCoE until 2011.

What it means to the enterprise:

Brocade is either serious about migrating its installed FC fabric base to Ethernet or it is stalling that migration. In any event, Brocade apparently has a plan to take customers there, whenever they decide to move. It also supports the CEE specs that many other vendors support, which might make customers more comfortable.

Where it falls short:

Without pricing and availability information on the 8000, and by publicly stating that the FCoE market won't materialize in a meaningful way until 2011, the 8000 might be seen as merely a prop to delay the migration from FC to FCoE and a way for Brocade to preserve its cash cow in FC. And with the market apparently moving to Ethernet where Cisco is the incumbent, enterprises — namely, the people who acquire and are trained on Cisco Ethernet switches — may not want to decommission that in favor of the Brocade/Foundry unified fabric approach.

Juniper's Project Stratus requires a little help from its friends

Data center convergence pitch: Project Stratus is a Juniper-led effort in partnership with server, storage and software companies to develop a converged data center fabric. To date, Juniper has only revealed cloud computing partner IBM as a Stratus collaborator. Stratus comprises six elements: a data center manager, storage, compute, Layer 4-7 switching, appliances and networking. It is intended to be a flat, non-blocking, lossless fabric supporting tens of thousands of Gigabit Ethernet ports, an order of magnitude reduction in latency, no single point of failure, and with security tightly integrated and virtualized. Stratus is expected to support the CEE data center fabric specifications being defined and endorsed by several vendors.

Recent news:

Juniper just began shipping the EX8216, a high-end Ethernet switch that the company is pitching as a cloud switch. The 8216 is a 16-slot platform with a switch fabric capacity of 12.4 terabits, and is optimized for high-density 10 Gigabit Ethernet data center and cloud computing environments, Juniper says. Juniper claims the switch has a per-slot capacity of 320Gbps and delivers up to 2 billion packets per second performance, features the company says will be required for 100G Ethernet in the future. Pricing for the EX8216 starts at $76,000.

What it means to the enterprise:

Juniper has a high-performance switch, credible partnerships with industry heavyweights, and a vision for converged data center operation and cloud computing that enterprises should consider. Indeed, Juniper appears to be getting tighter with data center titan IBM and seems intent on focusing on what it does best — high-performance networking — instead of competing with its partners by launching its own server, storage or software products. Enterprises comfortable with their incumbent IT suppliers but considering alternative data center networking vendors might find Juniper's vision complementary to their own.

Where it falls short:

Because it relies on so many partners, Stratus has a lot of moving parts. Any partner who balks might throw a kink into the works. Also, Stratus is only a year into a multiyear gestation — timelines may not be in synch with user migration to a converged data center. And Juniper is still a largely untested Ethernet switch vendor in the enterprise. The 8216 just began shipping so it is immature. Another chink in Juniper's data center plan is the lack of a JUNOS-based top-of-rack 10G switch. The EX2500 is OEMed, presumably from BLADE Network Technologies, and is inconsistent with Juniper's single operating system consistency and TCO story regarding data center networking. And Juniper has had a lot of operational frustrations with its enterprise business, chiefly in its go-to-market strategy.

Previous ArticleNext Article

Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.

GET TAHAWULTECH.COM IN YOUR INBOX

The free newsletter covering the top industry headlines