Slashdot Mirror
Web Servers Getting Naked, For Weight Savings
1sockchuck writes "Cloud computing is causing servers to get naked. HP today announced a 'skinless' server optimized for customers packing thousands of servers into cloud or HPC environments. This follow the lead of SGI/Rackable, which ditched the cover when it introduced bare bones servers for its CloudRack (previously discussed here). HP says the skinless design makes servers far lighter, which is apparently an issue when shipping them by the rackload."
Servers getting naked - IN YOUR EMAIL.
Just sign up for our newsletter, and all of those from our affiliates, co-conspirators, third party hordes, and lawyers...
Sure, you can find naked servers at google, but don't you prefer the personal touch?
If I can slashdot naked, a webserver can host pages naked!
Oh damn, dropped the remote on the floor. Let me bend over and pick it up. BRB.
(enjoy that hairy mental image :P )
Naked doesn't quite equal skinless, unless you're saying that you're naked of skin. Or...?
The new 'blade'; 19" wide and 1.75" tall.
I see discrete Ethernet phys, VGA, USB, etc.; all the horrible stuff blades are supposed to consolidate away. Turns out all the proprietary silicon, software and exotic backplanes necessary to make that real costs too much and is creepy.
And you can quit calling it "cloud" now... they're just hosting providers and you know it.
Lurking at the bottom of the gravity well, getting old
This makes sense, since the dust should already be filtered, which removes a large part of the need for a case. However, I do wonder about the airflow, since an ordinary case helps to direct the airflow through the kit rather than over the top, which might be a problem. On the other hand, without a case, the ventilation will be much better, so what is lost on the swings may be gained on the roundabouts.
This is a nice idea though, and would make sense for rackmount routers/switches, since these usually sit in an enclosed cupboard anyway.
bTW: first?/p
My friend used to run a BBS way back when, and he told me he would just hang the motherboards and other components on a pegboard on the wall. Similar idea, but I think he was doing it to save money on cases and possibly to save space as well.
Company charges more for servers with less steel - film at eleven.
Makes the servers more serviceable, and in a server closet there isn't much that would require a skin to protect against.
Sorry... #28302585 beat you 16 minutes earlier...
<sig> </sig>
... time. It's cooler, faster, lighter, cheaper and better for the environment and it looks a hell of a lot bad ass when you open up a system that's got it's guts exposed and just start hot-swappin' like a mofo. Sad thing is that it's driven by $$$$ and the need for companies to shave even a few pennies off their TCO when I've been doing it to my systems for years now for the above-state reasons.
The eternal struggle of good vs. evil begins within one's self.
turns out the "extra weight" was needed to finish loading his comment.
I would say skinless instead. Probably for servers "naked" should mean without OS installed, what looks specially attractive since you can choose to install on them open clothes.
Nice rack
The product page is here. There are pictures. And specifications. And video. I think we can skip the rest of the blogofrenzy and get our info from the source.
BTW, this looks pretty much like a DL1000 installed backward. It's probably different in some other meaningful way. I didn't look too close.
Help stamp out iliturcy.
Steve Cobert, please go back to your desk and keep quiet on this.
Nevertheless, this is the only naked thing in the world you will get close to.
They added a 12V only power supply and a 12V battery, integrating the UPS as well. All the 12V stepdown can happen on the mainboard!
Totally OK if the battery is an optional replacement for the second hard drive.
Origin 3000 series servers did this a long time ago. The bricks in the system were just fans on the front and a base plate to mount the hardware onto. They were pretty easy to work on in this configuration, you could pull the brick out and replace anything inside within a couple minutes. IRIX on the other hand.........
"Some books contain the machinery required to create and sustain universes."-Tycho
Google designed their own cloud servers as separate components (ie. naked). Old news as its saves a ton of $$$ in cooling and energy costs
"Cloud computing is causing servers to..."
What's with calling everything by near meaningless terns like 'cloud computing' all the time now?. The coverless servers are not due to 'cloud computing', they are just a different technic for server farms. It could be for databases, large analysis, supercomputing, regular network hosting, etc. There is nothing about this that makes it exclusively meant for 'cloud computing' , it's just an idea for large arrays. Unless you a a marketing tool stop saying cloud computing just because it's the hot new phrase. Save it for when it's relevant
Why does a computer have an external skin anyways? It's helpful for desktops to prevent damage from spills, but in the rack mounted environment, unless the skin increases cooling somehow, it's actually worse than useless.
Actually, "cue".
...You can see their bare circuits!
Naked servers have been running for ages, motherboard disk etc. plugged together without any sort of case. Do not forget to correctly ground every component although.
Google even had their first stacks of hard drives running naked, they kept them in rack build of Lego blocks to allow air circulation I would presume. Motherboards were probably naked too although I can't tell for sure.
Everything I write is lies, read between the lines.
Things usually have covers for a reason.
I can see admins trying to install a new server and accidentally dropping one of the mounting screws into the rack. Poof! Admins would probably just try to RMA the unit. I see this costing HP some money because of that.
Here you go.
Except these are designed to arrive with the hoods "pre-lost", saving us the trouble.
I think you underestimate just how much I just dont care.
Running servers naked is likely to make them a larger spewer of RFI (Radio Frequency Interference). All those computers are nice until the FCC comes along and says you have to shut it down because you're interfering with the neighbors.
I see a few confused posts here about "WTF? Cooling?"
Just RTFA, folks. It's a blade server arrangement, not a standalone computer. These "naked" computers are nothing more than a pair of dual-proc computers, in a 1U-ish chassis without a lid, which needs to slide into the appropriate rack-mounted housing in order to work. This housing includes all of the cooling and power supply goodness one would expect, and (of course) includes a top panel to promote useful airflow and limit RFI.
I don't see much "new" about these things at all, since AFAICT most/all "blade servers" were already naked since their inception.
Color me unimpressed.
Kid-proof tablet..
If someone comes by to fix just one server on a live rack, it helps prevent stuff like screwdrivers etc from falling into the other servers. Or cables from tangling with the wrong stuff...
:).
Skinless is fine when you can treat each server/blade as a "card" in the "computer" (rack). Or you're running one of those massive sites that only changes stuff "by the rack". Then you just wheel out the entire rack and replace it with a new one
It's not so good in "messier" and more heterogeneous server rooms - where someone might stack an el-cheapo 8 port gigabit switch on the server, instead of waiting for that new expensive rack-mount switch to arrive.
Que? What pron comments? :P
The blade itself is called an ExSO.
Is the frame it lives in called an ExSO Skeleton?
And you have a deal.
Deleted
Is this thing running on DC?
Starbucks, Harbuckle of Breath.
Back in the good old days we first had computers that were made out of cased modules. Then we made computers out of PC boards that fitted into racks with backplanes. Then (in our case) we combined 3 CPU/memory boards into one chassis, with common fans and PSU, only we were too young and ignorant back in 1981 to know that Marketing had to call it a "blade system".
From scarped cliff or quarried stone she cries "A thousand types are gone, I care for nothing, no not one."
It's a bit silly: for scalable, cheap, replaceable systems, it makes sense, especially by putting the IO in the front for both systems that are in that box. And it avoids some of the single point of failure and high add-on expenses of blade servers, especially the cost of the multiple internal switches and remote KVM capabilities. (Has anyone else run into the "turn off spashimage in grub" problem to get access to the serial port boot console? That is a royal pain.)
But I'm concerned that people will install both of their "high availability" servers inside their skinless server, not realize this is what their upstream site did, and be screwed when it loses power. And power supplies _do_ fail.
But of more importance to trolls: is the server also petrified?
The concept you describe is implemented in some data centres--the rack contains not only shared cooling smaller number of much larger sized fans) but also the power supply, where AC goes in and the DC goes out to all the machines in the rack.
On a smaller scale I've done this at home: At the telephone demarc point where the electrical panel is I mount my DSL modem, a switch and an old FlexATX board to perform router/firewall functions. Instead of a PC power supply and multiple "wall warts" there is one power supply to feed single-rail 12V power to all the devices, thus only requiring one power outlet for the whole works. The whole thing will run off one 12V sealed lead-acid battery as well, for about an hour on the old battery I tried I'd estimate, but I've yet to incorporate it into the setup permanently. The old mainboard uses a CF card plugged into the first IDE connector (CF cards have an IDE-mode operation--the adapters are for the most part just re-arranging the physical connector for the same signals), and to accomodate single-rail power I use the "Pico PSU" that takes in unregulated 12V on 2 wires and puts out all the ATX power required. It plugs into the ATX power socket and isn't all that much bigger than a typical ATX power connector inside your PC, except not attached to all those wires.
If you're starting with new hardware, you can forego the PicoPSU and use MiniITX boards with single-rail power. VIA offers some models that take simple 12VDC which are tailored to mobile and industrial markets (an elegant choice for Automotive PCs).
In a larger application requiring multiple hosts I envisioned a setup that would forego proprietary blade setups or expensive racks and a significant amount of sheet metal:
* very simple metal frame to allow a completely standard ATX board to be mounted vertically into "slide in" tracks of a mini-chassis
* mini-chassis would have a 500W or higher normal ATX power supply and a couple of larger-sized fans to pull air from bottom to top, with room for drive bays, KVMs or whatever along the bottom.
* Such a mini-chassis could be made to fit in 8U of space in a 19" rack and could house 5 standard ATX boards with enough room to allow for low-profile PCI cards to be installed. The depth would be less than 1/2 standard rack depth (perhaps about 30cm) such that two of these units could fit in the same 8U space, for a density of 10 boards per 8U space. Not the most density but still better than one system per rack unit. Many more boards could fit if room for PCI cards is not required. Of course, the chassis need not be rack mounted--it could just sit on a desk.
* boards would be mounted with the "back panel" facing front, and "internal connections" routed out the back--perhaps with connectors mounted for HDD and power fixed to the mini-chassis so that the frames with boards mounted can be simply "plugged in"
* A second mini-chassis could share PSU with the first to make more efficient use of power.
Not rocket science but that sort of "mini-rack" case would use common ATX hardware without proprietary form factors involved, and no more than one PSU per 5 PC boards. This isn't a new concept--it isn't that much different from the SGI Origin systems with multi-node bricks mentioned in another post--the only fundamental difference is that it is based on standard form factors.
Actually, queue. We expect a lot. We're just about to tell them to line up.
The building, however, will need shielding or - depending on scale - birds will fall out of the sky and the electronic blackout might cover a square mile or six (for starters). Carrying chocolate or milk around might not be a fun idea. Actually...
Well then, carbon-boron shielding, perhaps ? And the random standing-wave artifacts ought to be really fun to pinpoint. They might consider using damping-rods, as in traditional reactors. MHD ? Salts ? Plus, they'd look really, er, cool. And they're also really useful, when the plot calls for some suspense, for example.
Reminds me of the Manhattan Project. A gaggle of Nobel-grade scientists manned the axe and the deuterium-filled buckets, just in case the first rod and graphite reactor also became the world's first china syndrome - or just blew up (some weren't to sure). The rock hanging from a rope was a last-ditch 'safety measure'.
It might put a bit of a strain - on the supply of deltas, you know...