Slashdot Mirror
Microsoft Joins Open Compute Project, Will Share Server Designs
1sockchuck writes "Microsoft has joined the Open Compute Project and will be contributing specs and designs for the cloud servers that power Bing, Windows Azure and Office 365. "We came to the conclusion that sharing these hardware innovations will help us accelerate the growth of cloud computing," said Kushagra Vaid, Microsoft's General Manager of Cloud Server Engineering. The company is also releasing its Chassis Manager software that manages its servers, fans and power, which which is now available on GitHub. "We would like to help build an open source software community within OCP as well," said Microsoft's Bill Laing. Microsoft's cloud server hardware is built around a 12U chassis that can house up to 24 server and storage blades, offering a different approach from the current Open Compute server and storage designs."
"We came to the conclusion that sharing these hardware innovations will help us accelerate the growth of cloud computing,"
I really don't get this so perhaps someone can explain: What benefit does this have for anybody?
Embrace, extend or extinguish? Which part are we on now?
death rattle
I'm surprised to actually find that they chose Apache License for the project instead of their GPL-incompatible MS-PL. I have no idea what Chassis Manager is actually useful for (and a skeptic inside of me tells me that probably nothing unless you pay for their other products), but it's interesting to see that they actually released 36k lines of code as free software.
So many to choose from. On the hardware side, they didn't like the current hardware design that Facebook liked (with good reason) and so provided an entirely different set of design sensibilities. This isn't about enhanced standardization, this is about a nice-sounding venue by which to deliver requirements to bidders for MS datacenter equipment. I will say at first glance, I like MS's requirements better than Facebook requirements. The points that I'd worry about would be firmware requirements (MS tends to get insane with network protocols) and the unique IO design which limits the market of compliant equipment (basically, the same that can be said of IBM bladecenter, flex, Dell M1000, etc etc).
On the software side, you'll note that it's in .Net. It's very much not in the realm of typical open source datacenter operations projects. MS once again is stuck having to build the infrastructure themselves for lack of a wider community using their tooling for the purposes MS needs. Of course, MS has historically impressed me with how well they manage to do while being a 'lone wolf'.
XML is like violence. If it doesn't solve the problem, use more.
MS sees an overwhelming trend of Linux oriented software for operations management. They don't have the same mountain of people working to do similar stuff but with .Net. They want that mountain of companies and people building an ecosystem around .Net, but it's not happening naturally. MS has only the hope of putting out there and hoping to prime the pump.
The world has moved on. Sorry MS.
It looks like the skeleton of code to put into 'firmware' of the chassis manager. This suggests they believe the chassis manager should be running Windows as the embedded solution... Holy shit what a terrible idea for a standard, the cost of the module would increase to start with (a hardware design for that role runs 40-50 bucks, moving to atom doubles that) and the cost of the OS to run on top of it would be more than the hardware cost total.
Every single time it the same with these guys.
They arrive late to the party, but when they do they bring with them their own music CDs.
Its like the arrival of the vaguely creepy older guy who likes to hang around younger people.
Nobody's got the nerve to turf them out, but at the same time nobody wants to interact with them either.
Later on, everyone says how creep-guy's arrival ruined the party.
While creepy-guy's telling everyone who will listen how awesome it was.
Microsoft + Open + Share ??? Are you serious? Those three words cannot coexist truthfully in any sentence in this Universe. Anyone believing different needs to consult with the kookoo Dr.
That being said, there appears to be a patch of unknown turbulence ahead in Tech Airlines' current flight path. IBM is again "reinventing" itself, having now shed its entire line of low to medium level Intel servers to Lenovo. Oracle can't get its head out of its arse and has no clue how to move forward with all the stuff it plundered from Sun's corpse; namely the Java and the Sparc ecosystems. HP still doesn't convince that it has fully digested the Compac brunch and has left the dark Fiorina days behind. The Chinese upstarts haven't convinced the World that the evil minions of the Communist regime aren't behind their efforts, much like the nefarious "ghost in the shell" presence of the NSA everywhere else. Troubling times indeed.
this isn't your father's Microsoft.
never bring a twinkie to a food fight.
They never seem to quit. Thugs and criminals, look at MS history. It's amazing how the Justice Department in the late 1990's did not break the corporation apart, i guess JD were bought off. This open source thing is probably a trojan horse for future money extortion's just like what they are currently doing with android. Do not get yourselves involved with MS when it comes to open source, nothing good will come out of it.
Prior to Sun open sourcing java, there were also open reimplementations of the runtime. Those reimplementations only kind of worked if the developers targeted them explicitly.
Similarly, GNUStep and OSX can compile the same code... if the developer is careful to target both....
Of course, there is also wine. A project that after decades in the making does an admirable job, but the confidence of a particular application running perfectly is still not a given until you try.
Mono is in the same boat, trying to reimplement MS stuff without a lot of attention/regard from the MS development team.
Meanwhile, the runtime doesn't really offer anything not offered by Java. If under linux, Depending on the application, I'd personally lean toward either C, maybe C++ or python.
Azure is the bluish color #007FFF.
An Azure Cloud is the suffocating bluish smoke belched out by an engine that is reaching the end of it's useful life.
Microsoft is courting Linux workloads to run on their bluish smoke servers. Why would someone who has a business application that runs on Linux want to trust that to a company that has tried to destroy Linux and open source and is actively continuing to do so to this very day?
I'll see your senator, and I'll raise you two judges.
Ah, I remember when Microsoft did the same thing with the OpenGL ARB and more or less poisoned it, leaving OpenGL for dead once they had completed their mission. Also, W3C, Java and pretty much every other standard group they've sabotaged^H joined usually end up the same way.
So forgive me if I see this as yet another attempt at killing off open standards.
More reading
systemd is not an init system. It's a GNU replacement.
Microsoft isn't giving their server designs away out of the goodness of their hearts. They have a huge interest in getting people to move their workloads to Azure. The first step for most places has to be getting them off of VMWare or KVM onto Hyper-V/Windows Server. Next step is convincing enterprises to buy these whitebox server designs to save money on their on-premises stuff. Finally they'll make Azure too good a deal to pass up for the CIO crowd with the usual argument that you can fire most of your IT department. It's already super-easy to publish your applications right from Visual Studio to Azure...again, not an accident.
I actually think the whitebox design method is a good thing...IF...you have a dedicated staff working 24/7 to repair/replace sickly boxes, and the workload is such that a box is a box is a box. This works perfectly for large scale web apps backed by a SAN, or hypervisor hosts. It doesn't work as well for standalone application stacks that have semi-permanent physical server dependencies. Renting 3 servers in the cloud doesn't make as much sense as renting 3,000.
My company does a lot of standalone deployments of applications around the world, in places where network connectivity doesn't permit easy cloud access. It's getting harder to find vendors who aren't trying to steer us to the cloud. Microsoft is making it very difficult to purchase perpetual licenses of software, with the price of a negotiated Software Assurance deal being set less than the equivalent one time license fee [1]. Now that IBM just bailed out of the x86 server market, HP is pretty much the only vendor left making decent hardware for non-cloud applications.
I totally get why AWS, Azure and public clouds make sense. When you're running the back-end for an iPhone app, and need 40,000 web servers all cranking out the same content, it makes sense to rent that. But a lot of companies don't seem to get that it's more expensive to do the cloud thing if the servers are going to be permanent and you're hosting one of those boring line-of-business apps. Hopefully people will realize this before the last decent x86 server vendor quits selling non-cloud-optimized servers.
[1] Licensing SQL Server on multi-socket physical boxes is insanely expensive now compared to VMs. I had to add ESXi to our solution for a recent deployment just to save thousands of dollars on the database license for a low-powered app.
I dont know if anyones used Azure but my boss made me find a reason to shoehorn it into the infrastructure because microsoft swore our 25k in free credit for it was tied neatly to our license discount for Windows. it is a clusterfuck of unworkable web 2.0 line and symbol bullshit that is easily outranked and outclassed by even the most entry-level hosting providers in ease of use. heres a rundown of my experience:
1. signup. microsoft juggles you between 3 different portals, all of which basically mandate internet explorer, and a username with a microsoft TLD. now that im boatanchored to the rest of the redmond world, we can continue to provisioning?
2. no. now you have to apply for a service and confirm the subscription in email. what this means, i mean on a technical level, god only knows. its some ephemeral obfuscation imbued in the product to impart a sense of legitimacy in the process of your virtual cloud experience no doubt.
3. we have a subscription and now we can start provisioning images. you have about 10 different microsoft images and 2 linux images sanctioned by some third party entity no ones ever heard of. Linux VM's require a goofy disclaimer but come with a package selection feature, so i guess thats useful.
4. Windows or Linux, youve made your choice, and youre provisioning nicely but beware: navigating away from the provisioning page will cause the process to stop.
5. Whatever lofty dreams you had about microsofts commitment to cloud and scaleable architecture as a departure from their haggared burro of licensed OS and direct attached storage becomes an afterthought. Microsoft (as they did me) emails you stating they improperly provisioned your VM in the wrong datacenter and that you, not them, are now responsible for the fix. this requires you delete your entire VM and start over again.
6. you stare into the internet, your limit break clearly reached, and observe an ocean of other more capable and well established providers and players in this world of virualized SaaS and PaaS. the interface is clean, the support is in plain fucking english, and if you arent hounded to tie your active directory to it. the thought that anyone, or any group for that matter, would stop to give two shits in an open consortium of existing sucessful and dedicated players to consider an offering from a software company that for its entire existence has sought nothing but ruthless destruction of every other open standard in the world, is bad comedy.
to microsoft: no one cares, and I mean this in all sincerity. its not a troll or a flame its just a sad fact. your designs, your servers and your processes and procedures contextualized historically in their offering to the open anything community have been a complete farce. this isnt your cup of tea and it never has been. You're completely outnumbered, hopelessly outgunned, and the best you can do is peddle lock-in to traditionalist business models sadly manacled in mediocrity. Look at your phones, tablets, zunes, and everything youve fought so hard to make a part of the world thats forsaken you and just stop shoveling time and money into strategies you're laughably unqualified to adopt.
Good people go to bed earlier.
What's next the CIA believes in transparent government?
Are they planning to release their hardware designs with a license that includes an open-ended patent license to actually build equipment using the plans? If not, consider this a Trojan horse.
One thing to start with, OpenCompute is highly dysfunctional. It doesn't do much to build consensus and basically has been just whatever Facebook specifically wants, without regard for any others because Facebook is frankly too full of themselves.
So Microsoft came at this with a number of alterations because their needs ever so slightly differed from Facebook mandating pretty dramatic changes. Of course it's still less helpful to the wider audience, but it does cater to certain situations.
Facebook *currently* doesn't have structural demands that are unmeetable with their weird, 21" wide rack format. The challenge is that there are certain places where trying to make a go of 21" without going over 24" total footprint isn't viable. Structurally, there just isn't enough room to play with for certain earthquake prone situations or ship loaded. So a design for 19" (the *real* standard) opens up options for rack design.
Facebook doesn't see a problem with cable management, and such makes a beast of a thing with unreasonable cable management circumstances. The MS design follows in the footprint of blade servers and pulls IO into tho midplane.
Facebook doesn't need that much memory density, so they advocate a design that frankly can't accomodate much memory per CPU socket.
What Facebook gains in horizontal density, they lose in vertical density. They justify this as improving airflow, but it's pretty much bullshit. There are ways of getting the right airflow without constraining things to fit the fans in the particular mechanical way Facebook selected.
That said, the MS scheme of course has issues. The enclosure manager doesn't seem to be paying much attention to standards either. The 12U enclosure choice seems peculiar and means bigger midplanes to service and less granularity between chassis enclosed and non-chassis enclosed equipment.
Basically, MS is no worse than Facebook, an organization about 'standardization' abusing the term 'Open' as an excuse to make vendors violate accepted industry standards in the way that particular user wants that screws over the vast majority of the market.
Azure must be priced as a loss leader to get companies to sign up. Managers see the dollar difference between Azure and a real hoster that's been around a while, and calculate that into their bonus for the year, and dictate by fiat that Azure WILL BE USED SHUT UP AND STOP ARGUING your technical opinion means nothing because the decision was already made. That sort of thing.
Trust but Verif--
On second thought; fuck that: never trust Microsoft, F/OSS folk!
The 'standard' is in a non-sabotaged state already. 'Open'Compute was mostly Facebook saying 'this is the right way to address the only requirements that matter (Facebook's). Now MS managed to get their word in and it just gets slipped in, without the slightest hint of trying to reach a consensus on anything.
OpenCompute has been full of either things uselessly specific to one user's needs or alternatively useless problem statements that are blatantly obvious without a hint of a proposed solution.
Basically, vendors and purchasers are probably about as dysfunctional as they ever have been when it comes to standards, no matter what the words 'Open' imply.
XML is like violence. If it doesn't solve the problem, use more.
Upon joining, Microsoft's first official contribution was a comprehensive list of over 1100 Anti-patterns that are proven to stifle even the best open computing initiatives.