Blade servers and convergence pick up steam

FREE

Become An Insider

Sign up now and get free access to hundreds of Insider articles, guides, reviews, interviews, blogs, and other premium content from the best tech brands on the Internet: CIO, CITEworld, CSO, Computerworld, InfoWorld, ITworld and Network World. Learn more.

Blade servers are marching into the data center in increasing numbers -- often dragging converged networking with them

Last week, IDC released the latest version of its Worldwide Quarterly Server Tracker. Above the fold, you'll find news that the server market is continuing to heat up in a very big way. For the first time in more than two years, all sectors of the server market have posted year-over-year growth in the same quarter (Q1 2011), representing aggregate growth of 12.1 percent to $11.9 billion. This indicates a significant loosening of the corporate purse strings as it relates to server infrastructure purchases.

However, there were several interesting observations below the fold that may signal a significant shift in the enterprise storage and network infrastructure spheres. Chief among them is the significant uptick in the x86-based blade market -- nearly doubling the revenue growth figures posted by the server market as a whole. In fact, over 20 percent of investments made in x86 server hardware went to blades.

With more than 50 percent of the blade server market, HP's c-Class blade offerings have remained very popular. However, readers who have always thought of Cisco as a networking company may be shocked to realize that increasing demand for Cisco's UCS blade platform has moved the company to third place with a tad less than 10 percent of the market share -- a fairly amazing feat given how little time UCS has been on the market. However, it should be noted that Cisco's introductory pricing on UCS has been fantastically aggressive, so it will be interesting to see if it can keep up this pace as time goes on.

Part and parcel to all but the smallest blade deployment is heavy use of network convergence. Buying a high-density blade chassis only to then equip it with a multitude of pass-through modules and do the network aggregation somewhere else rarely makes sense, either financially or from an operational management standpoint. The fewer external network ports you end up burning and the less hardware you need, the better. As a result, many organizations' first experience with network convergence, or maybe even 10Gbps networking, also happens to be when they bought their first blade chassis.

HP's solution has generally revolved around its Virtual Connect line of blade interconnect modules. Together with most G6 and G7 c-Class blades, these modules allow administrators to slice and dice dual 10Gbps blade downlinks into discreet virtual FlexNICs -- giving each blade up to 20Gbps of network bandwidth divvied up into as many as eight FlexNICs with their own bandwidth allocations and network configurations. All of that is possible, requiring only a pair of interconnects and no additional mezzanine cards, though more can be implemented if additional bandwidth is required.

Instead of producing a blade offering that could adapt to any potential customer's connectivity needs, Cisco has gone all in with convergence and almost completely commoditized connectivity within the UCS platform. The story starts with the collection of virtual interface cards and converged networking adapters supported by the blades. Unlike HP, Cisco has opted not to include any LAN-on-Motherboard in its blade hardware, instead allowing the customer to choose what kind of connectivity is required for each blade separately.

The most interesting of the available mezzanine cards is the M81KR, which is targeted at virtualization hosts. It allows up to 128 virtual PCIe devices -- be they FCoE HBAs or Ethernet NICs -- to be deployed per card. It also supports Cisco's VN-Link technology, allowing network configuration and policies to dynamically follow individual virtual machines as they move across virtual host blades and chassis. Other more "traditional" CNA options are also available, and it's somewhat telling that Cisco is forcing me to refer to a CNA as "traditional" to set it apart from what it's doing.

All the traffic from the mezzanine cards on the blades is aggregated by a pair of redundant fabric extenders in the back of the blade chassis. Those fabric extenders are essentially one-size-fits-all affairs that support four proprietary 10Gbps uplinks each. These interfaces attach to external fabric interconnect switches based on the same NX-OS software that runs Cisco's Nexus line of converged network switches (the fact that they also run Cisco's UCS Manager software distinguishes them from their Nexus brethren).

Depending upon the fabric interconnect switches chosen, as many as 40 eight-bay blade chassis can be supported from a single redundant switch pair and can support a mix-and-match quantity of standards-based lossless 10Gbps Ethernet and 8Gbps Fibre Channel uplinks into the rest of the data center network, be it Nexus or otherwise.

To be sure, Cisco's offering is targeted at HP's upmarket. Due to the cost of acquiring the required fabric interconnect switches, you aren't likely to see many single-chassis UCS deployments. HP has covered the territory well, even supporting a convertible rack/tabletop eight-slot chassis. That said, Cisco's "bandwidth is bandwidth" approach that extends from the adapter in the blade all the way through to the chassis-aggregation layer and beyond is significantly more "converged" than anyone else's.

To continue reading, please begin the free registration process or sign in to your Insider account by entering your email address:
From CIO: 8 Free Online Courses to Grow Your Tech Skills
Join the discussion
Be the first to comment on this article. Our Commenting Policies