Slashdot Mirror
A Quarter-Million Dollar Box For A Free OS
popeyethesailor writes: "According to a CNET story, the server startup Egenera will be debut its high end Linux servers for financial services customers, running Red Hat Linux.
An earlier CNETstory details their design." That's a hefty pricetag, but the companies they hope to sell to ("market--financial-services companies and service providers") aren't shy about investing in tools. Of course, an S/390 isn't cheap either, no matter how many GNU/Linux images it's running ;)
Could somebody please explain to me where the $250,000 value is? Is this just another case of bad allocation of venture capital? The $250,000 is the BASE price of a system that can hold up to 24 cpu boards that CAN be connected to a network or CAN be connected to a drive array. The stated purpose in the article is to provide redundancy for failover. The only cool thing I can see is that if a cpu fails, another cpu will assume its name, characteristics and storage space. What wasn't clear was whether or not all 24 CPU boards were redundant, or whether you could have several redundant machines within the same "cabinet." But there wasn't anything really magical going on here. These boards would contain either 2 or 4 high-end processors (just over 1 GHz). I can see a price tag of maybe $40,000 or something, but certainly nowhere near the order they are asking. Anybody have any insight on this?
GreyPoopon
--
Why is it I can write insightful comments but can't come up with a clever signature?
Looks like they bought a copy of the Redhat High Availability server for about $2000 and loaded it into a rack of CPU's.
Pretty much any competant tech could do it. I've had customers running systems like this for Geophysical 3D Migrations for over a year now. No big deal really.
It sure took me forever to find a "product" in their website. Mostly just organisational and marketing bullshit.
Here's why:
n g-multimedia stuff. (I'm not bitter-I'm jealous)
1. The dot-com boom has pretty much evaporated, leaving the realm of "professional computer work" to geeky types with college degrees and bad hair (I'm one of them). The work that is done is now more mundane and laborious(billing, insurance, reporting, etc) than $20K-bonus-scooter-riding-dot-com-hipster-streami
2. Computers are now getting bigger and more mainframe-y (See comment above). More and more enterprises are centralizing mission-critical functions, primarily for ease of management as well as power and security. Proof:. We've already got Linux/390, the Solaris E10K, there's some newbigandexciting Intel box out there I keep hearing about that has 64-way SMP and now this.
Anyone have the newest Creative Computing?
"Beware of bugs in the above code; I have only proved it correct, not tried it." -- Donald Knuth
What we have seen 10 yrs ago? Last-generation hardware being used for servers. Now we see newer and better software running on older hardware designs (e. g. S/390). Do the math. Next generation of even more powerful software will run on even older (yet refurbished) hardware designs: expect Linux 4.x run on 8192 processor UNIVAC, with 5.0 kernel for 50GHz ENIAC in the works.
Computers make very fast, very accurate mistakes
Actually it's more like a really dense server farm.
Possible markets? The only thing that really comes to mind for me is ISPs. This could replace racks of essentially standalone machines quite nicely. But of course that's not the best market to go after right now.
I'm wondering if more conventional companies would go for this. There are lots of companies that have a ridiculous number of little servers floating around. If they had been deployed on a beast like this from the start then they could have saved a lot of money. But now that they have all these little servers, it's hard to imagine them throwing them all out and replacing them with one box.
It is tempting, if the only tool you have is a hammer, to treat everything as if it were a nail. - Abraham Maslow
This is a verry good trend when you stop to think about it.
One of the key issue technical column writers have been b!tching about is that Linux lacks enterprise server credibility.
With Linux driving mainframes and massive Credit Card / insurance company type machines who could complain about Linux's capabilities to handle their buisness demands. (if it can balance the budget for a fortune 500, it can host your stupid ASP/Intranet/fileserver/DB)
Think about the (Ugh! I'm gonna be sick) marketing angle... the average small buisness, or even home user, can have access to the same toys as multi-billion dollar corporations and goverments. (barring the obvious memmory and other hardware limits, this is about perception after all)
And it's not about a free OS. It's about the ability to develop the app on a PC and recompile it to run on a computer that makes Deep Thought look like Rain Man. And on top of all that the big system will work just like any other linux box running X. So it's easy to administer (wow! Who would have thought to say that about Linux!!)
I would rather be ashes than dust!
This just goes to show that the total cost of ownership for Linux/Unix/NT/2k has very little to do with the license for the OS at all. Hardware, admin, the software running on the box and so on more than make up the the trivial price differences between most server operating systems. Just because a Linux CD might be free doesn't mean running it on an enterprise box is going to save you a single penny.
http://twitter.com/onion2k
This must be what microsoft is talking about when they say that Linux has a high total cost of ownership ;)
Bryguy
microsoftword.mp3 - it doesn't care that they're not words...
Check out www.rlxtechnologies.com. They have had the same technology available for almost a year now. The 'blade plane' for reducing the number of cables needed... etc... etc... And you can get three blades in a 3U case for $5k.
Then there's render farms (for movies) and various scientific applications.
But it still seems like a pretty small market. Of course, a small company doesn't need to sell that many boxes to survive, so maybe that's ok.
It is tempting, if the only tool you have is a hammer, to treat everything as if it were a nail. - Abraham Maslow
The labor component of TCO (the biggest) is inversely proportional to the population of people who know about and can support the system. As more and more programmers/sysadmin get "on board" with Linux, TCO goes down.
This is also called "lock-in", the primary value of a software product is not intrinsic, it's how many people know about and use your system. It works very much like rock music... the more well known it is, the more popular it becomes (even if it is god awful). Of course, in software it's double powerful beacuse people familar with the software make other software that is dependent on the base software, thus creating a multiplier to this effect which is so very powerful.
NT and Office have a "low" TCO, since one can *hire* people off the streets to administer and use these products without additional raining. Hopefully Linux will be the TCO leader by saturating the sysadmin market from the bottom up. If sysadmins perfer Linux over NT, then Linux will eventually have the lower labor component of the TCO.
- A number of front-end or presentation servers, often web-based
- A set of middleware/application servers
- A number of back-end servers, normally running a database of some sort
In addition you might see a load balancer in there as well for more complex systems. This box allows you to put all of these things in a single physical unit, with a nice high speed interconnection between them, along with the ability to add servers as required.A single server with many CPUs like the Sun E10K is great but very complex and really expensive. It doesn't give you the freedom to separate out components. That's why people moved away from monolithic boxes and on to the distributed model. This machine is trying to combine the best of both worlds, with modularity of servers but a much better sense of locality for a single application spread across multiple systems.
Sharing interfaces to the real world makes sense, too, as most of the traffic can stay internal. Think of the cost of $2-3K per fibre channel interface and $1k per GigE interface, not to mention the relevant switches, and suddenly this box doesn't seem to be too expensive after all.
I imagine that this will ultimately stand or fall on the TCO, the biggest part of which is bound to be management...
It's a cluster, not a server. It's hard to tell how this is that different from a rack full of 1U servers, but I didn't read their Web site carefully.
Besides the fact that Linux can access more than 4GB of RAM, this thing isn't a single server; it's a cluster. If each of the nodes has 4GB of RAM, that adds up to a lot.
The actual benchmark machine for 'Charlie' was a rather low end machine, probably 1 million total cost. With 40,000 images that's 25 bucks a server. Let's say that in practice that's off by a factor of 20. That's right let's say the benchmark understates the actual cost by 95%. That ends up $500 bucks a server. Still too much?
That's the only way most corporations will ever accept the use of (Free || Open Source) Software. I work as an IT consultant to @BIG_OIL_COMPANIES, and you wouldn't believe how hard it is to get them to accept things like perl. Hell, I think the only reason they did eventually let us use perl is because ActiveState is around so an actual company is out there that we can point to. Sad? yes. But that's the way it is out in the trenches.
News for Geeks in Austin, TX
At (workplace - 2) I was the PFY at a place that used a horde of PCs in a compute cluster. Horde as in north of 150. Probably half of our time was spent simply running around fixing dead or dying machines. I think we had an average of one total machine failure a week, with lots of lesser events
thrown in to make life interesting. The most common failure mode was just a power supply crapping out (not unsuprising becuase these guys were running at 90+% system load 24x7x365).
News for Geeks in Austin, TX
Vern Brownell was CTO at GS. He's CEO, not CTO, at Egenera.
Don't confuse a large number of 'logical' machines with physical ones. If a Pentium III had the ability in hardware to subdivide itself into thousands of functionally identical logical processors you would be able to run thousands of Linux instances on that one CPU. You probably see the problem that you would immediately encounter: each Linux instance would have only a tiny fraction of a percent of the PIII's processing power. Yes, you'll have thousands of distinct running instances of Linux, but they will be very slow when several of them tries to do something cpu-intensive at the same time.
A mainframe CPU is not dramatically faster than (any other) microprocessor anymore. In recent years I've only been able to indirectly compare the benchmarks; it seems that IBM isn't interested in submitting it's mainframes for industry standard benchmarking these days. Bottom line: a 12-CPU mainframe is still a 12-CPU box, even if running 1,000 or 10,000 instances of Linux.
The mainframe's value is no longer in being a honker of a computer. Reliability, the ability to run existing OLTP workloads, and manageability are the big reasons people still buy mainframes.
Move along now; there's no magic going on here.
Geeky modern art T-shirts
But if the part that the computer thought was defective turns out to be O.K. will it go on a murderous rampage?
Even though the system in question was not a mainframe (intel based blade plane type system), I do want to say a few things about S/390...or whatever IBM is calling them now. Everyone thinks when MS or Linux adds support for a new fangled thing (say the new buses on the PC that are supposed to be mainframe channel like....), well, the mainframe has been doing it for years! When the PC folk added virtual ram via paging stuff out to disk, that came first on the mainframe. Almost every type of PC technology that comes down the pike has it's roots in the mainframe world. PC's need better I/O buses....in comes channels and so on and so on. Our mainframe support consultant that I work with used to call PC's pretend computers because they didn't have half of what the manframe did. Now servers are starting to get these I/O things and we are supposed to gasp because it's new. Well, it isn't new and it's been around for 15 years on the mainframe. Mainframes are solid so long as your network stays up and you don't have students hammering on the thing! :)
Gorkman
This thing has got nothing on my cluster of 3 12MB 486DX266's hooked up with fat 10Mbps ethernet to a screaming 16MB P-75 controller running Slackware with IPVS kernel patches and giant 800MB IDE disk.
I gots ta ding a ding dang my dang a long ling long
Nor does it have ESCON adapters or OSA's or virtual routers or WLS or RACF all of which if they were functionally implemented on PC would tend to eat it alive. The point is that while the CEC itself maynot execute more 'ticks' than a Pentium the system architecture is designed to provide efficient performance.
Well, it sounded good, anyway. I think you get the idea.
What we call folk wisdom is often no more than a kind of expedient stupidity.-Edward Abbey