I remember reading articles like this about technologies and companies that I was excited about.
In the late 80s the stories were: "The Apple // is still relevant." It wasn't. I remember OS/2 fans pointing to "huge deployments in banking", "huge in Brazil", and "every ATM machine is running OS/2." Whatever truth was there was irrelevant, as IBM had given up on OS/2.
Apple has given up on selling a server operating system. Apple has given up on selling server hardware. Apple failed.
A cluster of minis makes sense if you are running a CI build farm for an OS X application. It probably makes sense if you're building a FCP render farm. In both of these cases, note that it only makes sense because of the licensing issues. In a saner world you would be running on commodity x86 hardware.
If you are an Apple fan and find yourself arguing that your employer should be using piles of minis-running-Linux based on this article, don't.
Apple has not failed with the mini because it never tried for this sort of thing in the first place - it is a happy accident. And one that I believe they noticed and allowed to go on organically with the bare minimum of effort on their part.
The reason collocating minis is intriguing is this: This consumer grade hardware has gotten so reliable and powerful that it makes for a very good server - being a consumer SKU means it has a production run orders of magnitude greater than a Dell server for example.
Thus a Mini is disposable. It is almost like a Raspberry Pi.
A proper server will obviously be more bang for the buck in terms of raw hardware specs, but the TCO of a Mini is favorable because you can resell it and it has a good, none-enterprise-y warranty.
A fully loaded Mini amortized comes out to something like $50/month. It also blows EC2, 'etc, completely out of the water.
In the rather rare case of it failing it is simply swapped out by the facility with exactly equivalent hardware in short order while yours gets handed off to Apple Care, at which point you can simply ebay it for 70%+ of its original price and get a brand new one switched back, all in a matter of a day or two. If you need to "spin up" more, there are always more laying around that can be provided by the facility immediately. Dedicated hardware that is so flexible and convenient is a powerful thing. Depending on your use case you could get away with running on just a couple of minis and save quite a lot of money.
If you buy them in threes for redundancy Apple offers nice discounts and financing. You want to compare how much performance you get out of three minis for $200/mo (total) to what that would cost you on Heroku?
Oh and I'm pretty sure you can boot linux from an SD card or some such.
This is absolutely true. I have a 16GB dedicated mini with ssd and 8 cores for $130/mo. There's no way I could get anything that cheap and powerful from Rackspace or Amazon. I plan to follow suit with more of the same.
Apple failed in the server market. They have had success with the Mini.
A Mini is cheaper than EC2, but that isn't the point of EC2.
You don't run in EC2 because it is cheap or performant, you run there because you can spin up virtual machines quickly. You run there because it is adjacent to S3. You run there becuase it integrates with the likes of RDS, ELB, and Route 53.
It depends on your use case of course. If you spin up dozens of instances all the time then sure - but keep in mind that an email like "put another mini in my rack" gets executed in a few hours and is equivalent to quite a lot of compute units.
That's rather the point, even very large spikes in growth can be swallowed by just a few of these in theory and certainly there are always enough Minis in close proximity such that it makes no difference.
EC2 is inherently unreliable, and so are consumer-grade desktops like Mac Minis.
The strategy in EC2 is to be able to spin up replacements very quickly when your instances fail.
The only strategy for a non-power-redundant, non-RAID, no-IPMI Mac Mini failure is the same, except replacement is measured in minutes or hours instead of seconds. This is huge.
If you're running above-capacity, a single failure of a box in any working set (routers, database shards, app servers, etc.) doesn't matter. Just wait the few hours and let the rest of your boxes silently absorb the load.
If you're running exactly-at-capacity: why?
(This presumes, though, that you're Highly Available enough to begin with that one node lost is a quantitative, not qualitative, failure in your system. If you only have one app server, time-to-reprovision is the least of your worries.)
I think most open source people would say an server os is one without a GUI. People are forgetting Windows Server is still largely the same as Windows on the desktop.
Windows Server has a ton of features not present in the desktop versions. For example IIS running on a Windows 7 desktop is capped at 10 concurrent connections for licensing reasons.
Whereas I could install a desktop distro + apache on my server and serve as many requests as the hardware can physically handle. The only difference would be that default settings would probably not be well optimised for that use case, but some reconfiguration would solve that.
That's not really true. The current version of Windows Server by default has no GUI (and almost no services). It's configured completely from the command line (or if you follow Microsoft's guidance, remotely via powershell remoting).
I certainly wouldn't recommend any shops currently using Linux servers to switch them out, but Windows Server has come a ridiculously long way from being super sucky to being a good server operating system.
Yeah, Gui-less was an option on Server 2008, but not the default. The default for 2012 is what they call "server core" (no GUI, no desktop services).
I was surprised when I took a look at Windows Server recently. I'd sort of mentally written it off a decade ago, but Microsoft has done a commendable amount of work in improving it.
Again, I don't think there's anything about it that I'd tell current Linux admins to swap out their servers for (even ignoring the license costs), but they've improved Windows Server enough that I no longer recommend that existing folks abandon it.
Also, for what it's worth, Powershell is quite possibly the best thing Microsoft has come up with in 20 years. Yes, it's embarrassing that it took that long to get Windows command-line administration to what Unix people have been doing since the 70's, but they've implemented it in a genuinely smart way.
One good test is if they have a decent track record on pushing CVE fixes. Apple tends to hold theirs to about once a quarter unless something is in very common use. They intentionally hold them back so that people don't have to click update too often.
This seems very odd to me. Why restrict one's self to an unsupported, proprietary, unvirtualisable OS when Linux and the BSDs are neither of those things and very similar.
“We thought about applying it to other hardware, but if Dell makes a small little machine, you don’t know that they’ll be making that, in that form factor, six months down the road, or what they’re going to do, or how they’re going to refresh it,”
This is completely upside-down. Apple's future plans are entirely opaque. When Dell releases an Optiplex desktop, Poweredge server, or Latitude laptop they commit to not making changes for some non-trivial amount of time so that corporate images don't need to be fiddled with.
At a previous job, we used http://www.macminicolo.net/ for a couple of years. Awesome (awesome!) service, great pricing. Haven't been with them in about a year, but I haven't heard anything bad about them since I left...
As far as servers go, if you don't need the sophistication of OS X and its great development tools and process handling, why not just use an off-the-shelf and supported desktop (i.e., Lenovo ThinkCentre) with *BSD on it? Paying the premium for monolithic construction and an OS that is likely never going to see the height of its potential in a specified application seems a bit much.
A big part of what's pushing the mini use in the datacenter is that pretty much everybody will give you a cheaper colo deal with less restrictions than a 1u. Because the mini lacks stuff like raid, multiple nics, ipmi etc. the typical customer can be relied on to have a pretty idle server. The mobile parts help, but if you filled one of those high density mini racks and used them as compute servers or cloud hosts or something your rack would melt just as fast as any cpu heavy stack of 1us. But with almost everyone just running a hobby website, directory server or mdm you can fit 10 or 15 of those things into to the power budget of a 1u or 2u cdn edge box.
You can see the artifacts in a lot of their price lists. Dedicated colo? Sure, $35. Oh you want a managed firewall in front of it? that'll be $50 extra. Want some space on our fault tolerant SAN? No problem, $200.
I bet it's a great way for a hosting company to get some use out of cages in the popular dc's that can't sell any more power.
At http://objective-cloud.com we are also using Mac Minis with SSDs and 16 GB of RAM. A single Mac Mini can do so much - computing wise. The only bad thing about Mac Minis is their GPU which may make some use cases less efficient to realize.
"'I’d say the biggest worry is that it runs on 2.5in drives,' Stucki said. 'But, those have been come more reliable lately, and especially in stationary machines like a server.'"
2.5" drives are reputed to be more reliable than 3.5" drives, due to less rotational vibration. The advantage of 3.5" is density.
I have one in a 2-mini shelf in the colo to do OSX Server device management/etc. stuff. It's..interesting. Apple seems to keep changing their remote management tools, and unless you have Kerberos set up, there's no good way to do cut-and-paste long random passwords or anything else through the remote desktop tool (it doesn't work during the secure password input window).
Probably wouldn't recommend this to anyone unless they have a very specific need; a comparable Atom-based Linux server is cheaper, easier to manage, etc. If you need performance, you can get a low-end i7 based server, and if you need ECC, etc., get a low-end Xeon.
I call this concept "Ubiservers" for Ubiquitous Servers. The idea is the consumer demand for highly integrated computing devices, based on SoCs, combined with Moore's law being applied to solid-state storage will lead to very small devices that 'serve' content or computing to others becoming ubiquitous.
Why I do have one and you should consider getting one:
- reliable : besides the 2nd HD giving SMART errors and that I had preventively changed, I never had one hardware problem in over 3 years. It's on 24/7 and on UPS. I have a second one without UPS, also working like a charm.
- small : whenever you need a server, it will fit.
- green : it won't suck too much power, so it won't overheat. I live in a tropical climate- this matters to me.
- versatile : it can be reconverted into a small desktop when you no longer need it.
- standardized : you don't have to worry about the video card or the revision of whatever internal piece changing between batches. If you compile your own linux kernels, that's a godsend.
- easy to find : you need one in a hurry? There are applestores everywhere. now try and get me a dell server in the next hour.
- osx already installed : sure Linux might be better in many cases, but my time is not free. OSX total configuration time tops 20 minutes, including plugging cables.
- quick to configure : if you just need something for php+psql, download osx server package and you're ready to go (That's included in my 20 minutes estimate). It's sucking less than the previous versions (I hated snow leopard server). If you are not in a big hurry, I recommend instead to install your own utilities.
In both cases you can keep .dmg, which is useful in some work environments where for security reasons there is no internet connection.
Oh and here is a reason why I love rudix - prepackaged GNU utilities, with everything included.
No I do not want to recompile stuff. No I do not want to even think about tracking dependancies! If I need say mc, I want a .dmg which contains the whole thing and the libs, and that I can install on another machine immediately without worrying about glib or stuff.
TLDR; the mac mini is great. OSX Server "acceptable", but most of it can be replaced by rudix.
Sure you may be better served by a custom built machine, or a cloud service, but when you want something working ASAP on an intranet, I haven't found anything better yet.
I'm not even talking about the sweet no-questions apple warranty, or the rescue procedure standardization (alt key : will boot on CD or USB) common to all apple hardware. I don't want to remember if F1 or F10 or F12 will get me in the bios, or where I will have to click to let me boot from usb.
I buy mac minis because it's the best hardware I've found for my needs (and I say that while trying some hackintoshes time to time!)
If you haven't tried one yet, it is worth experimenting with.
EDIT: I know that RedHat, Debian, etc installation can be automated. If you have thousands of identical machines using known-to-be-supported hardware, and highly specific needs requiring a gazillion of .debs it's great (+)
But let's say it's 7pm and you need just one machine with apache to serve a php website on the intranet. Do you want to automate an installation and tweak it, or just take a box where a Unix is already installed, along with vi, apache, and the basic tools you need? Some minor configuration file edits will make that box production ready in a matter of minutes. (And BTW OSX software installation can be automated as well)
+ : automated installs do not fully remove hardware variation issues or make sure the drivers work. The best part of apple hardware is that it generally evolves very slowly - your kernel bins, modules, etc. will usually require more tweaking on 5 dells of the same model than on 20 different imacs (!)
In the ends, it's all about costs, and return to scale.
If you have thousands of linux boxes, it's a good idea to use cheaper hardware (and buy spares!) along with install automatization : high fixed cost, low marginal costs.
For lower volumes, a different solution means I have to spend far less time on it, which lets me concentrate on the real work.
> easy to find : you need one in a hurry? There are applestores everywhere.
Maybe in your country.
> osx already installed : sure Linux might be better in many cases, but my time is not free. OSX total configuration time tops 20 minutes, including plugging cables.
I'm not familiar with OSX, but the installation process for Red Hat based distributions can be automated.
I haven't said there's no Apple retailer just that they aren't everywhere. I'm also wondering how many of them have mini-Macs on stock, since iPhones are way more popular.
I live in Romania and I think that our neighbors are in a similar situation.
I would say almost certainly no. Apple has clearly become a consumer/retail oriented company; few organizations ever really bought into the idea of the Xserve line, and to the extent Apple are still interested in "enterprise" sales it's for MacBooks and iOS devices.
I have a feeling that if Apple wants to get back into servers they'll do it with an ARM-based product sometime in the next two or three years. There's not much room for them in x86, and they've clearly abandoned existing customers. But with ARM they can start fresh and leverage a lot of what they've put behind iOS devices.
But surely at this point you are going to be doing everything via SSH at which point the differences are going to be small, although you might be better going with FreeBSD I guess.
As someone who uses Mac Minis for servers the biggest reason is value.
I can fit 2 Mac Minis in a 1RU. And that is a Core i7, 16GB, 2xSSD which is about 20-30x faster than what Amazon, Rackspace offers for the same price. Now throw in the fact that you can (a) send it back to Apple if it breaks or (b) sell it on eBay for nearly the same price it all becomes very compelling.
Given that I use a Java/Hazelcast/Cassandra stack I can simply add more Mac Minis to linearly scale with no single point of failure.
I'm in the market for a couple for installing MySQL Cluster Edition on. I'll probably buy new (for a DOA warranty) and I'm going to skip AppleCare on them - I've had zero failures on the other Minis I've had. I've found that like you said, the used market prices are so strong that unless you're penny-pinching, there's no good reason to buy used.
If there is a discount for buying 3 or more at a time like recuter said above, I may investigate doing that, if it's a steep enough discount. :)
EDIT: Just chatted with Apple sales, and discounts are only for business customers, not individuals. :(
In the late 80s the stories were: "The Apple // is still relevant." It wasn't. I remember OS/2 fans pointing to "huge deployments in banking", "huge in Brazil", and "every ATM machine is running OS/2." Whatever truth was there was irrelevant, as IBM had given up on OS/2.
Apple has given up on selling a server operating system. Apple has given up on selling server hardware. Apple failed.
A cluster of minis makes sense if you are running a CI build farm for an OS X application. It probably makes sense if you're building a FCP render farm. In both of these cases, note that it only makes sense because of the licensing issues. In a saner world you would be running on commodity x86 hardware.
If you are an Apple fan and find yourself arguing that your employer should be using piles of minis-running-Linux based on this article, don't.