Sunday, November 7, 2010

Goodbye to XServe

So Apple has decided to discontinue the XServe. Not a surprising move, I'm sure it doesn't sell particularly well and it has not been updated in quite a while. If you compare it to other server offerings it also lacks serious Lights Out Management and Out Of Band management.

The upsetting thing about Apple's move is not that it is discontinuing the product but that it has done it without giving a large number of enterprises any real alternative or any admission that enterprises need one.

Apple's suggestion is that instead of an XServe you can use either a Mac Mini or a Mac Pro, indeed they are now offering a Mac Pro/OS X Server bundle as they do for the Mac Mini.

In the small business market a Mac Mini makes an excellent server. If, however you want something beyond a few users and simple services then you can quickly find yourself needing more processing speed and much more I/O. Providing backup services for a dozen Macs can quickly swamp a Mini.

At this point, yes, a Mac Pro is a good choice. You can install a Fibre Channel card and hang off a RAID for storage and you have two Ethernet ports.

The problem sets in for organisations that go larger than this. Even a large High School or a small University can find itself handling a thousand Macintosh computers. These are the people that start thinking about two big problems that Apple have totally ignored now they have dumped the XServe. Rack space and reliability.

To put a Mac Pro into a server rack will take 12U space, you can fit two next to each other, but that's still 6U per box, compared to 1U for an XServe. Even the XServe is competing with blade systems that allow 16 high performance blades in a 10U blade enclosure.

Then we get to reliability. The XServe has three major advantages over the Mac Pro. The first is the option of dual power supplies, it might not seem like much but it allows you to factor out both a power train failure by attaching to two different power busses and also a failure of the supply itself. The second is that drives and power supplies are hot swappable and easily removed and replaced. Compare that to the many steps required to replace a power supply in a Mac Pro. The third is lights out management. Once again it might not seem important but if I can hit a virtual reset button from my desk when something locks up rather than rely on another member of staff to walk up to the server room, unlock it and hit the button on the front of the Mac Pro it can save vital hours. It also makes it so much easier to get minor changes past a Change Management Committee when it only involves one staff member to make the changes and roll back if a problem occurs.

Now none of these problems are insurmountable for Apple. There are several possible ways to fix the large hole now left for medium to large enterprises who want a well supported Mac infrastructure.

The easiest for Apple would be a simple change to the End User License Agreement for OS X Server and allow it to be virtualized on non-Apple hardware. Parallels already supports bare metal virtualization on Apple hardware, supplying a ROM image so that this could be done on other hardware is easy for Apple and I'm sure Parallels and VMWare would be happy to get it running on a wide variety of hardware.

The second possible way is to do what they did with RAID. They annointed the Promise VTrak E-Class RAID as the 'official' RAID replacing their own. Apple could organise a deal with someone that already makes a range of servers and grant them a license to sell then bundled with OS X Server. IBM, for example, would make an excellent partner.

The third is that some time quite soon they will announce a new product that does the job of the old XServe but better.

There is one final solution, one that doesn't come from Apple. Since all the services on Mac OS X Server are based on open source cores you can host all the services on a Linux box. You won't get all the pretty stuff and not all the GUI front end tools will work but it is possible. I have a feeling that there will be a definite shift forward in the ease of installation and ease of use for this option as the many large enterprise Mac sysadmins start shifting to this solution and sharing their work.

Watch this space. Over the next few months I'm going to start exploring all the options to see what I can achieve.

1 comment:

bingo said...

Hi Tony,

Great post - must heartily agree with your insights.

Seeing a post beginning with 'lights out' and 'out of band' is music to my ears.

It might seem trivial to some but these issues loom large for me when I manage a datacenter in Australia with systems engineers in Europe. When the systems engineeer is a 16-hour flight away distance becomes a very real factor.

If onsite intervention is required a site engineer is a few hours call away at 2am (during European daytime) the overtime is a cost consideration as well as the delays from downtime.

When considering a 'big iron' setup as well 12RU for a server is a bit tricky, especially when your co-lo DCs can charge by the RU. Then there's airflow and standardiation of heating/cooling which becomes another cost factor.

For some organisations such a setup may not even make it past the architecture review, let alone change control - oh yeah and don't even talk to me about budget approval :)

Phew - I'm tired.

Thanks for the post Tony...

Cheers,

Hugh