IBM punts two racks, a blade, and a hybrid thingy
Enterprise avant garde
Nehalem Day IBM is looking for the new Nehalem EP-based servers to kick start its System x rack server and BladeCenter blade server business, which saw a steepening decline in sales as 2008 wound down.
Big Blue will today announce two new rack servers, a blade server, and some configurable compute and storage nodes for its avant garde iDataplex custom-made server clusters. It is interesting to note that the company will not, as yet, put a tower-style machine - the kind sometimes preferred by small and medium businesses - into the field.
IBM does a lot of its own direct sales to large enterprises, and the Nehalem machines it announces today are clearly aimed at enterprise customers. The SMB shops will get their towers soon enough, and besides, IBM almost certainly prefers that they use the BladeCenter S chassis - which runs on 120-volt power and which is designed for an office environment - instead of towers at this point in the history of computing.
"Lots of folks are going to announce hardware, and we did that too," jokes Alex Yost, vice president of BladeCenter products at IBM. Yost says that given the state of the global economy, IBM, like its competitors, are focusing as much on reducing costs related to administration of machines as well as powering up and cooling down the boxes. IBM is also keen on making it less frustrating to set up and support its x64 machines too.
For instance, the new machines all support the Unified Extensible Firmware Interface (UEFI), a superset of the BIOS means of setting up the iron inside a box. UEFI is backwards compatible with BIOSes. But it has a more modern user interface. It can be configured remotely. And in IBM's case, the "beep codes" that normally are used in BIOSes to tell you something is wrong are now converted into messages (available in lots of languages) that are diplayed on the Lightpath diagnostics screens on the front of the server.
The servers also include a new generation of service processor that IBM is calling the Integrated Management Module, which combines diagnostics and remote and local management of servers. This is where the extended BIOS lives. IBM has also collected more than 42 different System x and BladeCenter sites (used for patching and deploying these x64 servers) into a single tool center with just eight different tools - and all have the same look and feel.
IBM is also rolling out a tweaked Systems Director systems management tool, release 6.1, which now has a virtualization manager integrated into it. This tool can be used to manage power consumption on IBM and non-IBM x64 iron too. In addition to having lots of sensors scattered around the machines to monitor temperature and power consumption, the machines also have altimeters built in, so administration tools can take in the effects of altitude on the running of the machinery.
"This is all about telemetry," says Yost. "We have done a lot of work to simplify the task of getting the most efficiency and productivity out of the machines."
Generally speaking, Yost says that a Nehalem EP-based server will deliver about twice the performance of a two-socket Xeon DP box using "Harpertown" processors and that if you compare it to Xeon boxes from three years ago, you can get about nine times the aggregate performance in the new Nehalem two-socket machines.
Here's the significance of that comparison. Yost says that if you take 137 1U rack-mounted servers from three years ago (presumably using single-core processors), you can cram the same amount of computing capacity into a single BladeCenter chassis with 14 blade servers using Nehalem - and the return on investment for making the acquisition is about seven months just based on power and cooling costs alone.
This is clearly going to be the IBM sales pitch, and one that you will hear from all server makers starting this week. Customers moving from racks to blades tend to be interested in consolidation, and that often means virtualizing the servers. But Yost says there are some customers who replace servers on a one-for-one basis as they upgrade, and they are not interested as much in using virtualization to drive down power and cooling costs and server footprints as they are packing a lot more performance into the same thermal envelope.
Now, let's take a look at the new IBM iron, starting with the two rack servers, then move onto the blade and then finish up with the iDataplex nodes.
Sponsored: Benefits from the lessons learned in HPC