Steve Jobs on Xserve: Hardly Anyone Was Buying Them | Hacker News Reader

Steve Jobs on Xserve: Hardly Anyone Was Buying Them

mashable.com

29 pointsGarbage38 comments

Frankly, the Xserve was not an innovative product. When it was PowerPC based, the performance/power ratio was actually quite good, thus the supercomputer built with them. After the Intel switch, that really went away, and their hardware was no longer that much different from something you would buy from HP or Dell.

I'm betting this will go one of a few different ways:

- Apple will allow OS X Server on a small number of validated hardware platforms and/or virtualization environments.

- Apple will focus on HA clustering in it's software. This is already available to some extent with certain portions of their software stack, but this is often dependent on coherent storage (Xsan) which isn't available without a large fibre channel outlay, which rules out the Mac Mini Server.

- I doubt Apple will get rid of OS X Server as some are suggesting. It's implemented as a package that installs over OS X client, and is quite easy/efficient to install and admin. At worst, they'd likely open source more of it and let people roll their own (currently many of OS X Server's individual components are available on http://www.macosforge.org/ )

You're right that it was not that much different - but it was no worse or notably more expensive than HP or Dell solutions. And it ran OS X Server.

Apple has the weirdest communication strategy ever. CEO comments via alleged emails to customers? Bizarre.

I allows Steve to say informal and brash things while maintaining plausible deniability.

Also, it's fun.

Maybe if Apple cared about selling servers they would have. It was obvious to anyone that they never actually tried to build a real server offering. Apple did a good thing by killing this product, preventing a few people from making the mistake of buying these things.

You can't convince people the shinier product is better when they're buying servers based on actual numbers instead of whatever their impressionable peers say they should have.

Steve missed the point with the Xserve. Like all his other projects he should have create a unique solution that integrated with the Mac and iOS platforms.

And if Xserve did provide a unique solution then this is a marketing failure. Regardless Steve didn't follow his standard operating procedure for product development.

The compute density of the XServe was just too low, everyone else had double the number of sockets/U.

It's not about the processing power for Apple customers. It is for most server buyers though. Apple should have targeted the Xserve as a seamless network tool for Mac labs, and later as an enterprise integration tool for iOS. Something only they can offer.

Bottom line they didn't make it special enough. Instead they competed on pure specs. Something Apple has never done will.

It's all about the processing power. Jobs should have known this from his days at Pixar. A compute farm based around GCD and OpenCL could have been a killer product.

The iPhone is not about processing power, the iPad and the MacBook are not about processing power. The iPod neither.

Jobs knew that if was to fight the how fast is it battle he would lose. So he changed the fight to one of ease of use and style.

Changing the positioning of the Xserve is what Jobs should have done, he didn't. So yes in the end it was about the Xserves lack of speed and power. But you miss the point (Jobs too) that it never should have competed on speed and power to begin with.

Well you say that, but 3D on the iPad is very impressive.

The XServe was a server product, its job was to sit quietly in the background doing work that for whatever reason desktops couldn't do, or would be suboptimal for. What to Apple desktop users with significant compute needs (e.g. movie makers, CAD users, scientific users) do? Maybe they can use Mathematica and submit the jobs to Linux boxes running the Mathematica back-end (or whatever) but why not provide the seamless experience?

I am wondering if there is some point where a Mac Pro cluster with top-end CPU and GPU winds up with better compute density, at least for problems that OpenCL can address.

i.e., in 5U you get 40 Nehalem cores with 5 Xserves but only 24 Westermere cores with 2 Mac Pros, yet you can fit the Mac Pros with a Radeon 5870 or a pair of Radeon 5770s each, compared to a GT 120 in each Xserve. Eh, probably not. But maybe.

My guess is that Apple will open up OS X server up to a small number of non-Mac units. They did something like that when they discontinued the Xserve RAID: http://www.apple.com/xserve/raid/

That way they can keep their server software around without having to spend money building immediately uncompetitive hardware.

I think there is a huge difference between a SAN and a Server OS. Specifically, I don't think Promise RAIDs do anything clever to work with XSan. (it's just a SAN with fibre?)

The expense on Xsan is in validating the solution for performance and correct functionality, thus why Apple only recommends using the Promise units purchased from them with specific firmware revisions.

I've never used Mac OS X Server so can't comment on its worth, however... based on the hardware specs and the prices, I can't work out how they ever sold a single one (except to huge lovers of Apple who wanted to look geeky and hip at the same time by having one of these in their home.)

edit: If the price was only a bit higher than what I'd expect to pay for those specs with 5-10tb storage in it, then I'd probably be ok buying it as a home storage device, much the same as I'm ok buying an iPod as an MP3 player. But there's a point where "sexiness markup" becomes "double the price and slap on a bit more".

Having an Xserve with Mac OS X in a company that uses Macs because it all integrates and it would provide email, chat, directory, wiki, and other enterprisey stuff. I've also seen universities with full racks of them to use for distributed computing (I believe using the Xgrid utility.)

Just because it's interesting -- The University of Illinois has a cluster of 768 G5 Xserves: http://www.cse.illinois.edu/turing/. It doesn't appear to actually run OS X though (Fedora instead).

A 1U server generally doesn't make any sense in the home.

Some people used to put shoulder straps on them to carry them to LAN parties.

Can you fit a decent graphics card in 1U? I guess it all depends on the games being played.

Right angle slot adaptor.

Mac OS X Server running Open Directory worked well for us instead of running an Active Directory for integrating Solaris, Linux, Windows and MacOS clients. Unfortunately, Apple pretty much pulled the rug out from under us a while ago, the system administrator was replaced with a Windows zealot and now nothing really works at all.

I administer a Mac Server, and actually I had already made the decision to replace it with a Windows Box. We're using it to serve up content to a grand total of two Mac Minis, and the rest are Windows machines, for whom it's serving as domain controller, file server, DHCP, DNS, and backup. All in all I can't see the use case here.

I'm also generally concerned by the lack of a community around the things. You need a lot of people using them to properly debug and test network stuff, and Mac Server just doesn't have the install base.

I'm surprised that you haven't switched already. Mac Server is generally only useful if your network is above 60% macs that are managed. If it's something small, you could just stack an Airport Extreme on top of a Mac Mini, fire up the mac-centric services, and call it a day.

Running Exchange takes away about half the benefits of a mac server, too.

I started two months ago, and I'm the only guy managing a network of 24 machines that hasn't had anyone managing it full time for about a year and a half. I've been trying to focus on obvious improvements, and avoiding touching things that mostly work. Though it's a little flaky, it's a little less pressing than replacing 6-year-old laptops and upgrading machines running XP SP1.

The biggest problem was that (other than the redundant power supply and 1U size) the Mac Pro was a much better value. It is a lot easier to buy storage for the Mac Pro.

That's the funny thing about the server market, Steve, companies just don't buy them as often.

For example not a year goes by I'm not replacing scores of Apple desktops, iMacs, MacPros, Minis for all our users.

Meanwhile, the Xserves in the server rack are older than most of my friends' kids. There's no business case for replacing them as they do exactly what we need.

I understand all that Steve. The part I'd like you to understand is how creeped out my sysadmins get when we have to stuff our server racks with ill-fitting boxes such as Mini-servers and full-sized towers. I'd expect a company that pays as much attention to aesthetics and design to have at least picked up on that by now.

It would be really nice if Apple allowed OS X Server to be virtualized on bare-metal hypervisors.

They do! (Read the OS X Server license.) The "bare-metal" just has to be Apple-branded metal. :-)

After this, the only servers will be Mac Mini servers. Maybe they'll open up licensing for other servers.

<idle speculation> With Apple about to open a giant data-center in North Carolina, they've clearly put some thought into what kind of servers they will be running in there, and there's little chance that it will be full of Xserves. If Google goes to the trouble of building its own servers, then a hardware company like Apple would almost certainly go to even greater lengths.

My theory? Their data-center runs on low-power iOS servers with custom-designed ARM CPUs and the same small Flash-SSDs that go into the new MacBook Airs.

The Xserve probably looks like an antique to Apple staff who have seen whatever cool tech is appearing in their new data center.

I see some discussion around virtualization. 1-2 years ago I filed a radar and also sent a message to Jobs about allowing for Mac OS X Server in a virtualized environment and it was shot down, to say the least.

We ended up abandoning Mac OS X Server and our xserves. It was easier to deal with Macs and Active Directory than to develop an enterprise strategy blocked by Apple's Iron^W"Brushed Metal" curtain [1].

My sources from the inside mentioned that there were months of some pretty heated engineer-readable-only discussions on my radar #. And things change, but.

[1] Don't get me wrong, I swear by OS X and Apple hardware as a user OS.

Virginia Tech bought 1100 of them, then the world's third fastest super computer:

http://www.vtnews.vt.edu/articles/2004/01/2004-233.html

Except that was one cluster, no one else tried it probably for good reasons. Had that cluster been replicated I'd believe their claims. I don't see what they meant by innovative management software.

More than just VT tried it.

http://www.apple.com/science/profiles/colsa/

The German hoster regfish offers XServe based servers: https://www.regfish.de/cloud/macosx

I wonder whether they'll continue offering OSX servers.

Reply on news.ycombinator.com