conditions it's more cost-effective to use blade servers over standard rackmount servers. This week's question is:
In what situations would a blade server outperform a standard rack server, and why?
Russell Coker, computer and network security consultant and contractor
Blade servers offer a greater CPU density and a greater computer density per rack. Prior to the common use of virtualization technologies, the benefit of a greater density of computers was more significant. One former client of mine, for example, had a significant number of blade servers running Red Hat Enterprise Linux 4 (RHEL 4) that were lightly loaded. Since the release of RHEL 5, it has been a viable option for them to use a single physical machine with a Xen instance for each server (with significant savings in hardware purchase costs).
The main benefit of blade servers comes from the price-to-performance advantage from a higher number of servers that nonetheless consume less space, power and cooling resources. So the greater the number of blade servers in a deployment or data center, the greater the relative cost savings compared to a rack server approach. This means larger server environments used for clustering or Web hosting are ideal for blades. Space and power-constrained environments are also ideal for a blade approach. Standard rack servers can still be useful for similar savings via virtualization and server consolidation, and may also be a better option depending on the server components and applications to be supported.
Once you get into larger deployments and resource limitations, blades are generally preferable - as evidenced by their continued traction in the market.Click here to see CAOS Theory, the 451 Group's blog on open source in the enterprise. Matt Shields, senior Linux administrator and network engineer
I have found the best use of blade servers is for development and quality assurance networks. The lower total cost of ownership and the amount of space and power saved makes them perfect for the job. In the past, I was asked to build out a data center, and the requirements were to have a full set of production servers plus an equal amount of development and QA servers to replicate the production network setup. Since the development and QA servers were extremely underutilized, we took advantage of the cost benefit of using multiple blade centers, an iSCSI backend storage and VMware Infrastucture 2. We then re-created the entire set of development and QA servers to match the production network at about 40% of the cost. Since 1998, Shields has owned a Web-hosting company which is now called BeanTownHost. Joshua Kramer, middleware developer, Belron U.S.
I'm going to stretch this a bit and consider an ecosystem of servers. By itself, a blade server is not fundamentally different from a rack server aside from that it takes less space. Because of this, I'll concentrate on monetary performance. Blade servers outperform normal rackmount servers in cases where packing additional CPUs in a smaller amount of space provides economic benefit. For example:
- Use less of your space for utility and support functions (the data center) and more space for revenue-generating functions. For example, a call center can halve the size of its data center and add additional phone representatives. This would increase profits by increasing revenue for a given amount of space.
- Using less space to begin with can decrease your rent. This in turn increases profits by increasing savings on the same revenue.
- Being able to run additional servers in a given amount of space can increase your reliability. If you are able to configure a higher number of redundant servers for a given amout of real estate, a failure in any one server is less likely to cause downtime.
In seven units, you might be able to pack 14 servers. One rack might hold 84 servers or perhaps 78 servers and a decent-sized SAN. This allows you to use your real estate more efficiently and increases profit. It also allows you to increase redundancy, which in turn increases reliability. This is where the real performance of blade servers lies.You can find Joshua Kramer's blog here. Tony Iams, vice president and senior analyst of system software research, Ideas International
Blade servers are not optimized for raw performance as much as much as they are for their manageability and form factor. Product cycles for vendors often dictate that faster clock speeds, larger numbers of cores, and greater memory capacity (which are the factors that affect hardware performance) are available in rackmount servers before they become available in blades. But vendors have become more aggressive in driving the highest levels of performance into blade form factors as quickly as possible.
At the same time, users have begun to assign higher-priority to issues such as ease of deployment, availability, power consumption and lower cooling requirements, rather than needing to achieve the absolute highest performance for their workloads. In these situations, blade servers have enough advantages over rack servers that they may justify not having the fastest processors.
The one performance-related metric in servers that is becoming more important, though, is memory capacity, which can be critical to the suitability of a system for virtualization deployment. Since all virtual machines have to share the memory in the VM host, the more memory that a server can hold, the more VMs can be deployed on it. That is why vendors are stepping up their efforts to make blade servers more "virtualization ready" by increasing the memory capacity in their blades.Click here to learn more about Ideas International .
This was first published in November 2008