RLX Gets Denser

A reader writes: "There's story about RLX Technologies shrinking their "blades" server on Linuxgram." Knowing how much we pay for our "floor space" at the colo, the notion of having multiple blade machines is pretty cool - and shrinking this to a 1U form factor with 6 blade of the Transmeta Crusoe 5800 line of chip is pretty cool.

Comments

Correct link

[Magila]

http://www.linuxgram.com/article.pl?sid=01/11/07/2 251222&section=newsflash

Crusoe Schmuso (Yeah, I know I'll get flamed...)

[ekrout]

Hemos: ...the Transmeta Crusoe 5800 chip is pretty cool...

No, it really isn't. As much as we all wanted Transmeta to kick some royal ass in the chip market, they haven't, and may throw in the towel very soon. Here's a quote from an article posted here on Slashdot less than a week ago, I believe:

Meanwhile, Transmeta was courting Taiwan Semiconductor Manufacturing Co. to produce its next chip, the Crusoe 5800. IBM had been making the chips, but Transmeta wanted a lower-cost manufacturer. In February, Transmeta struck an exclusive deal with TSMC. But the switch didn't end the delays. Samples of the 5800 chip that Toshiba received had problems, which seemed destined to push the project to November and prompted Toshiba to kill the notebook for the U.S. market. "We'd get products and then find an anomaly. You can put in a workaround but the only way to fix it is through silicon," said Steve Andler, Toshiba's vice president of marketing. Before he was forced out last month, Transmeta CEO Mark Allen said the company was still completing "long-term operating life" tests on the 5800. Sources familiar with the situation said that some of the problems stemmed from the complex design of the chip as well as from Transmeta's testing procedures, which were not weeding out inadequate chips but were giving the company an early, erroneous impression of success. Others, however, blamed TSMC's manufacturing processes. Early on, many of the faulty chips consistently came from the same section of the wafer, which sources said indicated a manufacturing flaw. Normally tight-lipped TSMC blames the 5800's design.

So, Hemos, I'm not sure if you have actual experience with a 5X00 Transmeta chip, but from what I read, they're nothing to brag about.

I'm as pissed-off about this as any of you, but the truth is the truth.

Some RLX caveats

[Wee]

We played with an eval of the RLX deal in April. They were very nice. There were a couple issues with them. Not show-stoppers, but definitely things to think about before deploying them, however, which make the inappropriate in some situations. You'd have to do some work-arounding if you intended to replace "real" servers. Imagine an IBM Thinkpad as a server and you get an idea of what you'd need to do. Things we found:

1. They "slept". I had a server monitoring process which listened on a TCP port. If you didn't connect in a while (seemed like more than a couple hours, but we never timed it), a connection would take like 30 seconds. We figured out that the blade went into a suspend-like mode during periods of inactivity. Since we were looking at these as some sort of low-end round robin-ish thing, that wasn't a desireable feature. Probably could have been adjusted/conf'ed not to happen, but something to think about.
2. And that brings me to the other caveat: they have weasely little IBM microdrives. They're IDE, and slow and completely ill-suited for use in a server. However, you could easily boot off the network (or even the tiny drives; it has slots for two of them, so you could set up failover or something I guess) and then attach a NAS deal to it. But you could probably never get by with just the stock drives in anything but an extremely low-traffic situation.

If you had a shared server web hosting company, could spring for the net storage, and didn't mind a tweak or three, then you could probably host quite a few customers in a quarter rack as opposed to a full rack (the power savings alone would let you pack them in). Throw in another full-powered app/db server and you'd be golden. All in all, the RLX is very interesting in the right applications. Clustering is another possibly appropriate role, now that I think about it (no, I did not say the "B" word).

One more probably uninteresting note: The eval unit we had came with Debian pre-installed. The newer ones have Red Hat (and Win2K, I think). So if you only do Debian, you could probably get the older images and stick 'em on there. Might have to waive support or something though...

-B

Re:Some RLX caveats

[taliver]

They "slept".

We also just worked up a cluster of these, and I dealt with the people there (also very positive experience). We asked about the ability to power cycle and send the blades into and out of sleep mode.

The guy there told me that they had fixed a problem with the sleep state recently (I was talking to him July/August), and that there were some other issues with power cycling they were working with.

What about total power?

[dave-fu]

Nifty that Transmeta's finally (belatedly?) living up to its hype by showing that they can stack mo' CPUs into a smaller space as they run so much cooler... but how much computing power (bus bandwidth and everything) does a Transmeta have when compared to an Intel/Sun/whoever solution? Does it stack (no pun intended) up?

Tandy RLX?

[mmol_6453]

Is this RLX related to my old Tandy RLX 1000 I got from Radio Shack ten years ago?

The RS sales guy made my parents bring it back to prove I had installed Win3.1 on it. They kept saying it was impossible.

Well, it did take up 16MB out of a 20MB hard drive...

That was also the same technician that told my mom that I'd "never, ever need more than a meg of RAM."

RLX is solid gear

[asah]

I work at an ISV building an MPP application, and
we started eval'ing the RLX 324 back in the
summer, and have had 100% success with them:
in a nutshell, each blade is about half the
performance and half the price of our 1U servers.
Overall, the blades are nicerly "balanced" in
terms of performance.

The claims about density, manageability etc. are
all true (divided by 2, ie. comfortable margin).
Beyond sheer density, with bladed servers, you
can deliver scalable apps in a single box, which
removes the big objection that they're "hard to
manage".

We wrote our own cluster admin tools (perl) and run
on redhat 7.1, which they'll pre-install, so it's
been cake-- but this also means that we didn't
try out their management tools.

Having been burned by other vendors on 1U boxes
esp. heat/vibration causing reliability problems,
I've been doubly pleased that the RLX gear hasn't
had any problems-- exactly the sort of stuff
you'd expect from former Compaq execs... but
without all the proprietary crap-- as a test,
we reinstalled redhat from the retail CDs, and
it just worked.

Re:Crusoe Schmuso (Yeah, I know I'll get flamed...

[jandrese]

One thing that is overlooked in rackmount (especially 1U) servers is power drain. Computer centers don't have unlimited power (in fact since you have to both power the equiptment AND power the refrigeration units to draw the heat from your machines out of the CC the power requirements of your servers becomes something of an issue). The ability to run dozens of servers in a rack without creating a power situtation is a big bonus for computing centers.

One of the big trends in computing recently has been for servers to grow smaller and consume lots more power. Just look at your average P4 or Athlon compared to the old Pentiums and Pentium Pros. It's getting so bad that PCs are drawing more power than even energy hogging Alpha based machines in some cases. This problem is compunded by everybody putting servers in the smallest boxes prossible, 1U frequently, so that these energy hogs can be stacked all the way to the top of the rack and draw an enormous amount of power from the circut designated for that tile.