Slashdot Mirror
Can Open Hardware Transform the Data Center?
1sockchuck writes "Is the data center industry on the verge of a revolution in which open source hardware designs transform the process of designing and building these facilities? This week the Open Compute Project gained momentum and structure, forming a foundation as it touted participation from IT heavyweights Intel, Dell, Amazon, Facebook, Red Hat and Rackspace. That turnout is not an isolated event, but reflects a growing focus on collaborative projects to reduce cost, timelines and inefficiency in data center construction and operation. The Open Compute project is just one of a handful of initiatives to bring standards and repeatable designs to IT infrastructure."
Ironic that Facebook (The walled garden company) is behind this.
Could certainly take away from the potential backdoors and other vulnerabilities that seem to fall in the favour of those countries manufacturing many proprietary chipsets. Nobody can quite explain how China manages to penetrate even the most locked down networks (many theories on exploits appear but few can be simulated). Backdoors built in as artificial network latency and other relatively undetectable methods would be hard to include when the hardware itself is completely open source. A win for everyone and a move away from the seemingly unavoidable TPM movement.
A big difference to Open Source software is in the cost of the manufacturing equipment. Where you can get a nice PC for programming for $1000, including screen, chip manufacturing hardware is a lot more expensive. The Open Graphics project, for instance, is still looking for investors to make an ASIC version possible. See http://en.wikipedia.org/wiki/Open_Graphics_Project
That will exclude those hobbyists who just want to tinker with the design a bit. Only the most determined, who are willing to embark on a multi-year project, have a chance of getting somewhere. And companies with plenty of money, of course.
C - the footgun of programming languages
How about we actually stop the insanity that promulgates the need for the insanely sized data centers? Use smart caching, java applets and just send business logic via the connection instead of the bloated insanity of html. Instead of shoe-horning an intentionally stateless 'square peg' protocol into the 'round hole' actually go with something rational. Then your data servers only need to deal with business logic and you farm out more of your processing requirements to clients. (aka the rational approach). I despair when I see what applets (irrespective of the language - just the general concept) could have provided us and where we are now in 2012.
(P.S. I can spell, honest! It's just UK engrish)
How is this different from using reference designs, blueprints, and best practices?
Or are we only calling it "open source" so software weenies will think they know what's going on?
Best leave the real engineering to real engineers.
Sorry, I just had to get that out there, before some idiot brought it up. While we're at it:
CLOUD COMPUTING!
DISRUPTIVE INNOVATION!
CORRELATION DOES NOT IMPLY CAUSATION!
That should just about do it. Now we can have some intelligent discussion.
Facebook is just making the Working Group sound cool by calling it "open".
This is largely what is already in place, though the web has long since gone mobile and your client systems are mobile phones with limited battery life, limited CPU and display capabilities.
You also need "architects" who are not complete morons.
There are also problems with trust etc, the client cannot be trusted once it's running on the user's computer which means your protocols are opened up to inspection and your servers potentially to abuse.
Deleted
can opener hardware in a data center?
Serenity now, insanity later.
Data Centers care about efficiency and processing density. Can open hardware currently compete?
No, it won't.
As all that changes is the chipset drivers / drivers for other stuff on the MB.
The APP / OS code does not need to be changed to go from let's say corei3 to corei5 or say intel to AMD (Same x86-64 code base)
No.
Nobody that has ever actually worked inside a data center is really so stupid are they? Well, perhaps some of the local staff, but usually not the vendors ... usually. There are always exceptions.
Can it? Yes..
While we might not use a Facebook triple rack built exactly to the drawings. We will be building a prototype out of 80/20 and fully expect to feed technical details back into the project.
At a commodity level it is simply about who has the biggest distribution channel and who can get the stuff made for the lowest cost, probably somewhere in China. Since it is all commodity stuff there really isn't a secret about drivers, firmware or manufacturing.
Move up the scale a little bit to real managed servers with fault-tolerant redundant parts and real diagnostics and you have left the commodity vendors behind. And now there is a considerable value difference between Vendor A's approach and Vendor B's approach. You also have the situation where Vendor A's stuff integrates well with Vendor C but not Vendor B.
Google set a somewhat different standard for building a data center and doing it totally with commodity hardware. Cheap commodity hardware. As far as I know, this example has not been replicated by anyone large. I suspect a significant portion of Google's effort in building a data center this way was dealing with non-fault-tolerant hardware and systems with no management and/or diagnostics. It means stuff is going to go down at random times and you just have to deal with it by pulling the whole unit. I guess it works for them. I suspect most other data center level operations really aren't run as a distributed cluster where the cluster is fault-tolerant but the pieces are not. We are still pretty much at the beginning of clustering and fault-tolerant systems with complete fallover support as far as the mainstream is concerned.
Understand that if a company is supplying nothing but commodity hardware (think the low end of Dell), they can be immediately replaced with any other commodity supplyer. Which is why Dell is getting out of the commodity PC business - there is no value proposition in it. On the other hand, Dell supplying servers which are not commodity hardware but using lots of custom parts and firmware means (a) they can supply much higher value to the data center and (b) they are not easily replaced by competitors that do not have matching parts and firmware. Making that level of hardware "open" is suicide because then you have turned your high value hardware into a commodity with no value at all.
Understand that if a company is supplying nothing but commodity hardware (think the low end of Dell), they can be immediately replaced with any other commodity supplyer. Which is why Dell is getting out of the commodity PC business - there is no value proposition in it. On the other hand, Dell supplying servers which are not commodity hardware but using lots of custom parts and firmware means (a) they can supply much higher value to the data center and (b) they are not easily replaced by competitors that do not have matching parts and firmware. Making that level of hardware "open" is suicide because then you have turned your high value hardware into a commodity with no value at all.
Out of many server suppliers, exactly Dell actually is supplying commodity server hardware and their boxes can easily be replaced by about any kind of vendor.
Dell is taking a few things of what's being sold on the market, do "customize" (brand) its firmware and that's it. And what they're actually replacing usually sucks (e.g. their BIOS) or is somehow outdated and just a little buggy. For example, a colleague of mine did fix a couple of DELL raid controller issues just by downloading official LSI firmware onto those controllers using LSI's linux tools. Of course, we're loosing Dell's support, but in the end - do you prefer "full vendor support" or not loosing your data?
The only thing which isn't completely "commodity" are some spare parts, like power supplies, fans or hard drive trays.
However, Dell's controllers usually don't mind if you replace the dell-branded hard disk by a non-dell-branded hard disk.
Even DELL's kind of out-of-band-management called DRAC isn't that special. If you're not the serial console type of guy and don't like whatever level of IPMI is implemented on e.g. some Supermicro board, you may take a look at AMI's MegaRAC line of products, which coincidentally does have a lot of similarities to DRAC. If I remember correctly, that DELL 2950 I've been evaluating back in 2000 did have a full-length MegaRAC PCI card.
However, even today's DRAC is based upon IPMI, so even in this case it's not that an issue to replace some Dell box by any kind of decent server hardware.