On mySimon: Backpacker Magazine Award Winners
BNET Business Network:
BNET
TechRepublic
ZDNet
TalkBack 1 of 5:
Next »
Is David Berlind the Emperor here?
David's attempts to provide a rational, objective analysis of the value of blade computing compared to 1U servers is akin to the attempts of the head of the U.S. Patent office, following WWII, to convince the government to close the office stating that "Just about everything that could be invented has been invented". He tends to support the idea of 1U servers, being somehow superior to blades. Really? Using RLX in his analysis I can see why he's confused.

A 1U server (stands for 1 rack Unit - a measurement of height in a standard 19" engineering rack) is a self contained computer that fits in a low form factor, and can be mounted into a standard rack. The industry refers to this as "rack and stack". It really came about as a result of companies supporting horizontally scaled computing architectures typical of web environments, as opposed to vertically scaled systems which are a single large computing node managed by a single OS. Vertical scalability means that when you need more power you either add more CPU's to the box, or get a bigger box. Big boxes are expensive and don't lend themselves well to this kind of computing model.

The 1U server took off primarily in response to two factors: 1) Sun had a dominance in the web computing market, yet were expensive ( and there really wasn't much advantage running a horizontally scaled system using Sparc processors and Solaris) and; 2) the .com collapse drove companies to take a much harder look at the bottom line, so the market responded with Wintel based platforms. The "poor man's" approach to computing, but it worked pretty well. To same money, "pretty well" was good enough.

1U servers made good economic sense, and they did reflect America's fascination with the disposable model. If a node failed, there was little impact to the overall computing environment, and due to the low price, servers were discarded and replaced with new ones. Hey, no more maintenance contract!. It was actually more cost effective than running Suns, and meant less SA time and repair costs.

There's just one little problem with the 1U design and shops who went whole hog with this architecture quickly ran into a wall: they are extremely inefficient at heat dissipation, and as such it is not uncommon for servers at the top of the rack to have their face plates melt off, or in some cases, actually catch on fire. For the average data center, this is pretty much a bad thing..

Blades are the next evolution in horizontal computing architectures, particularly in the high performance computing space, and have lead to the parallel processing model and clustering using MPI. (Clustering not in fail over, but in making lots of little computers act like one big one). Blades are a logical extension of the 1U server, and can provide for higher density, better cooling, and even more efficient computing. That translates into lower cost. And despite the Dick Cheney school of economics, cost matters.

What many companies call blades today are nothing new. They're just a 1U server with a different name. The name came about by mounting a computer on a sheet of metal, with a faceplate attached. Whoopee! So, what's the advantage? Well, that depends. Blades from some vendors aren't much different from 1U servers, other than they mount on non-standard racks. Hmm .. not very compelling. They still get hot.

However, this new computing paradigm (sorry) has inspired some good old Yankee ingenuity. Take Verar's Bladerack as an example. These guys looked at the heat issue, mostly due to the use of many small, unreliable and inefficient fans that in some servers counted upwards of 10 or so per 1U server, and they came up with a simpler, more effecient design. mount them vertically, and then above each row of blades, insert a row of 8 5" high CFM fans, and voila. No more heat problem. The end result was the ability to pack 176 processors, each with 4gb of memory, and up to 4 disk drives into approximately 10 sqft. That's 10 sqft of expensive data center floor space. Does that really matter? Did they really achieve anything? Let's take an objective look.

When compared to a 1U server rack, the Verari solution draws considerably less power per unit of compute power by using the larger DC fans, and we all know HVAC translates into AC cooling, which costs bucks. The design also eliminates the vulnerability of the loss of a compute note when a fan fails: the overall vertical wind tunnel effect ensures all blades continue to receive adequate cooling. Fans are hot-swappable (they can be replaced without taking compute nodes down). That translates into much higher levels of availability and reliability, and means lower TCO through less down time, less SA time throwing smoldering 1U servers in the trash, and better service to users (time to market). Add to that less floor space usage, less AC and less HVAC, and the savings are obvious.

David's statements about excess capacity sound so logical, but in fact just aren't factual. Many data centers are faced with two huge issues today: not enough HVAC or AC for the raised floor space they have. Density, combined with inefficient 1U and particularly the inefficient "how row / cool row" syndrome actually made the problem worse. Some shops may have spare floor tiles, but they're getting less computer power for their data center dollar. However you cut it, TCO still matters.

As much as there are many companies out there jumping on the "blade" bandwagon, many offering no addtional value beyond the 1U design other than to offer the ocaissonal weenie roast and halon release, blades continue to evolve into a more cost effective computing model. They represent scalability up to tens of thousands of processors (when using MPI), and a sustainable infrasturcture (add more powerful blades as new processors come out). Sorry Dave, but these things do matter and every company knows that. That's why blade computing is project to grow 8X in the next 4 years.

This also explains why Verari (who? - used to be called Racksaver) ) is the 5th largest seller of X86 CPU's. They redefined computing efficiency, and did so by providing a platform that gives total freedom of choice as to motherboard, processor, and OS. They are vendor agnostic. No wonder they caught you by surprise. What vendor has ever made a computing product with the consumer in mind? You're right though, not all blades are equal. Maybe that's why Verari was the #1 seller of blades last year.
Posted by: wallerich   Posted on: 07/21/04 You are currently: a Guest | Members login | Terms of Use

Alert moderator to an offensive message

Subscribe to this discussion via Email or RSS

Is David Berlind the Emperor here?  wallerich | 07/21/04
Excellent summary: An additional point.  escoles@... | 07/27/04
Will Dell seek to purchase Sun?  Hilton McKenzie | 07/22/04
1st Cabling is a BIG deal, 2nd Space overall is too  dbrichar | 07/22/04
Enterprise Blades are a oxymoron!  todd.smith@... | 07/23/04

What do you think?

SponsoredWhite Papers, Webcasts, and Downloads