There have been marked improvements in blade server technologies in the past year, ranging from increases in horsepower...
By submitting your personal information, you agree that TechTarget and its partners may contact you regarding relevant content, products and special offers.
(faster processors, quad-core, and more memory) to improvements in I/O capability (more NICs, HBAs and higher aggregate bandwidth) in the latest generation of blade systems. These improvements, combined with the benefits that blades offer in manageability, modularity and ease of provisioning are starting to cause users in the know to take a closer look at blades for virtualization. (Many are already using blades for consolidation, before moving to server virtualization, and vice versa.)
In an interview last week with Jan Stafford, senior site editor here at SearchServerVirtualization.com, I talked about some of the reasons why blades are/will be increasingly chosen as a hardware platform of choice, in partnership with server virtualization as the software platform. I figure that the eve of the Server Blade Summit -- Blades and Virtualization: The Perfect Marriage, in Anaheim, Calif., which I am chairing this year, is a good time to prep you on other tidbits to think about, as you consider whether blades and virtualization might be a good fit for you, as you move down the consolidation path and beyond, with virtualization.
Server market share
Blades are still a relatively small percentage of servers in the world (IDC reported blade revenue at $788 million in Q4 2006, or 5.2 % of the server market), they are growing disproportionately at roughly 30% year over year (while the overall server market grew at roughly 5% year over year). Interestingly, one of the biggest impacts on the growth of the server market overall will likely be the growth of server virtualization. As users consolidate multiple virtual servers onto fewer physical servers, the number of new physical servers required is decreasing. So while overall server revenue will not grow as fast as previous predictions, the blade share of the server market is definitely going up.
The fourth quarter of 2006 held another interesting data point: HP claimed the number one spot in blade revenue from IBM for the first time, with 41.9 percent revenue share as compared with IBM's 37% share. The new HP c-Class chassis has clearly contributed to this growth, offering a new generation of the HP BladeSystem, with enhancements in networking capabilities, energy efficiency, and virtual I/O. (Note that for the full year 2006, IBM still held the lead overall, with 40.0 percent revenue share, followed by HP with 37.4 percent.)
Virtual Infrastructure 3
Research is also showing that adopters of VMware Virtual Infrastructure 3 seem most interested in using blade servers. This may be related to the profile of VI3 users as the high-end users, most interested in advanced features like HA (high availability), Dynamic Resource Scheduler (DRS) and VMotion, which are very compatible with a blade system approach. In addition, improvements in VI3 related to I/O, FC HBAs, failover, and boot from SAN, may be related as well. (For example, in ESX Server 3.x, all HBAs are handled by the VM kernel, with multipathing occurring within the VM kernel). SAN issues are significant because there is a higher percentage of SAN attach in blades vs. rack servers, and a high correlation of SAN usage, blades, and VMotion.
In addition, as users move towards a major commitment to the future with VI3, and go beyond basic virtualization and consolidation (e.g., more focus on availability) some of the additional benefits of blades become even more compelling. With more virtual servers running on a single physical server, availability, redundancy, and failover become more important, so as not to lose that physical server and all virtual servers running on it. (With more eggs in one basket, you need to be more careful with the basket.) The redundant, hot swappable, out of band management aspects of blades offer a high availability hardware platform to support VI3.
Virtualization and blades versus virtualization in blades
In conversations with many users as part of the research we did on our book, "Blade Servers and Virtualization: Transforming Enterprise Computing While Cutting Costs", we talked to many users who were focused on consolidation as a high priority. As they evaluated how to consolidate, there were often two related issues under consideration - software and hardware. The hardware decision was to blade or not to blade. The software decision - to virtualize or not to virtualize. Many users chose to implement the two technologies together in one implementation project, with server virtualization solutions running on blade systems. In general, when we talk about virtualization and blades, that's what we mean.
However, because blade server systems offer a new form factor with a group of server blades sharing components within a chassis, they also have been leading the charge in incorporating the use of virtualization technology within the blade chassis as part of a server, network and storage system. This has taken various forms from different vendors, but involves the concept of virtualized I/O pooling.
Egenera, as an example, introduced shared virtual I/O pooling inside their BladeFrame system from the outset, roughly 5 years ago. Their concept is to separate the processors and memory from all other I/O and create what they call a Processing Area Network (PAN) similar in concept to a SAN. Their blade system has processor blades with the processors and memory, which then share a virtual pool of I/O through their control blades, which contain the NICs and HBAs for communicating with networking and storage devices. This virtual I/O approach reduces the number of physical NICs and HBAs down to only what is required by workload (balanced across all NICs/HBAs), rather than requiring 2 or more NICs/HBAs per blade as a minimum. (It also automatically provides redundancy with N+1 instead of N+N redundancy).
A variation on this virtual I/O concept has been introduced recently by HP with their Virtual Connect architecture for the HP BladeSystem (in partnership with QLogic). Virtual Connect still utilizes the physical NICs and HBAs on each individual blade, but creates a pool of virtual MAC addresses and Worldwide Names to share across all the blades in the chassis. In this way, it virtualizes at the software level rather than the hardware level (like Egenera). This virtualization significantly simplifies maintaining the I/O configuration for servers in the chassis, allowing changes to be made within the chassis without having to reconfigure everything outside the chassis. IBM is also active in this area with NPIV (N_Port Identifier Virtualization) which IBM created and then offered as a standard, to enable the sharing of host adapters in IBM servers and FC fabrics. NPIV allows an HBA to be shared by assigning a virtual adapter to a virtual server, which is separately identifiable within the fabric.
For more information on where these capabilities are going and where they might help you, check out the Virtual I/O session at the Summit on Wednesday. If you're looking for information on where virtualization in general is going, come to the Future of Virtualization panel on Thursday and hear thoughts from VMware, Intel, IBM and XenSource.
Do you have experiences with virtualization and blades that you'd like to share? Email me via firstname.lastname@example.org, and we can keep the conversation going.
About the author: Barb Goldworm is president and chief analyst of Focus Consulting a research, analyst and consulting firm focused on systems, software, and storage. She chairs the 2007 Server Blade Summit conference and has just released a book entitled "Blade Servers and Virtualization: Transforming Enterprise Computing While Cutting Costs" published by Wiley. She has spent 30 years in the computer industry, in various technical, marketing, sales, senior management and industry analyst positions with IBM, Novell, StorageTek, Enterprise Management Associates, and multiple successful startups.