Apple’s Xserve was born in the spring of 2002 and is scheduled to die in the winter of 2011, and I now step up before its mourners to speak the eulogy for Apple’s maligned and misunderstood server product.
As a datacenter professional, I’ve been immersed in the world of servers for the past twenty years, and have worked in environments numbering from a handful of servers in a closet, to tens of thousands of servers distributed in multiple large datacenter facilities. I have hands-on experience with every server product that Apple ever shipped (and one that never saw the light of day), as well as servers from other manufacturers, such as Dell, HP, IBM, Sun, and many lesser-known and defunct brands. But this eulogy comes not solely from my experience, since I asked my industry peers via the Macintosh Managers mailing list to share their thoughts on this historic event, along with their collective history with the Xserve and its competitors and predecessors.
The conclusion we come to is that the Xserve never sold as well as it could have. It was hobbled from the start by Apple’s lukewarm support, its basic design, and software that never quite lived up to its promise. Apple says “servers don’t sell,” yet selling servers is exactly what Dell, HP, and others do, in high volume and with good margins. So why did the Xserve have to die?
The Xserve’s appearance in 2002 was a surprise to those of us who manage computer networks. Apple had withdrawn entirely from the server hardware market in 1997 when the company “Steved” the entire server line and laid off the server group. Before that, the Workgroup Server and Network Server product lines, launched in 1993 and 1996 respectively, saw some small success but were a minuscule footnote to Apple’s range of consumer- and professional-focused offerings.
Apple has always had an uncomfortable relationship to what is commonly called “the enterprise”—showing just enough interest to stay involved, but never committing sufficiently to make a real dent in the marketplace, outside of education. Even today, when MacBooks abound in classrooms and corporate offices, hardware emblazoned with the bitten apple trademark is a rare sight inside a datacenter. Apple just doesn’t know how to market and sell to big business, and the Xserve suffered because of this.
Despite Apple’s failings in dealing with the enterprise market, the Xserve started well, because in its early days it had some serious advantages over the competition. The PowerPC G4 processor in the original Xserve had excellent compute power, while consuming very little electricity compared to the Intel chips of the day. In 2003 Apple added the Cluster Node version of the Xserve and a storage array in the form of the Xserve RAID. The Xserve RAID was, at the time, the industry leader in terms of “byte for the buck”—it got you the most storage for the least cost. I saw it being used and deployed by companies that had never considered using Apple hardware before. High-profile compute clusters were being built with Apple hardware and were gaining recognition in the supercomputer realm. Unfortunately for those who purchased these devices, Apple’s support for the Xserve product line peaked right at the start. Things started coming off the rack rails starting in 2004, when the PowerPC G5-based Xserve was introduced.
Hardware Issues -- The Xserve’s basic design was not flexible enough to adapt to changing technology or market needs. Rack-mounted servers are measured primarily in height. Width is fixed by the industry-standard 19-inch rack. (23-inch-wide racks exist, but are rarely deployed outside of telecom facilities.) Racks are divided vertically by multiples of a “rack unit” (abbreviated as RU or, more commonly, U) which is 1.75 inches high. A typical server rack is 42 rack units tall.
Apple chose to design the Xserve to be 1U, meaning 1.75 inches high. While this made sense for compute clusters, it meant that for general purpose needs, the Xserve would always be limited in how many disk drives and features it would have, or how convenient it would be to use and maintain. The main issue with the 1U size was that in order to make the server feature-rich enough to satisfy the majority of customers’ needs, it would have to be outrageously long. The Xserve started out life 28 inches long, and grew 2 more inches over time, leading the industry in overall length of server hardware. Along with limiting features, the long 1U form factor also proved awkward in datacenters, where the Xserve stuck out further than servers from most other manufacturers.
I believe that the Xserve would have been a far more versatile server platform had it been designed from the start with a 2U (3.5-inch-high) chassis for everything other than Cluster Node Xserves. This would have allowed much greater storage and internal RAID options without making the server ridiculously long. Even worse, the PowerPC G5 and Intel CPUs in the later Xserves required more electricity to run and generated more heat, and therefore required more cooling. The Xserve’s original 1U design had to be compromised with large cooling ducts, which required losing a disk drive bay. A 2U chassis design would have enabled Apple to upgrade CPUs, offer far more storage, have redundant power supplies much earlier, maintain cooling, adapt to 2.5-inch drive bays, and provide a full complement of ports on the front of the server.
(Front-mounted ports are vital in today’s datacenters, where the “hot aisles,” meaning the backs of the servers, are usually contained inside an enclosure to partition hot air from the server intakes. People work in the “cold aisles” and avoid the hot ones, hence front-side user ports.)
Other design issues plagued the Xserve:
Disk drive bays that were far too easy to pop out of place accidentally. This was the most commonly cited annoyance of the Xserve in my correspondence with Mac Managers list members, and caused serious unplanned downtime issues for many of them.
A change in disk drive interface design (PATA to SATA) midway through the product line. Every Xserve customer probably bought a “wrong” replacement drive module at least once.
A complicated rack mounting mechanism that changed with almost every iteration of Xserve.
Structural weakness in early Xserves. They tended to bend downward, sagging in the middle, earning the name “Grinning G4.”
Swapped network and power port positions in 2008. This provided the option for dual power supplies, but it threw a huge monkey wrench into cable management in existing racks of Xserves. (Power and network are traditionally managed to be on opposite sides of high-density server racks to minimize potential interference. Changing this causes huge problems for datacenter managers.)
Expensive spare parts that were priced far higher than those from other server manufacturers. Disk drives were noted by many of my peers to be outrageously priced, and hard to find in certain capacities. You can still buy replacement 250 GB drives from HP and Dell, but Apple stopped selling them years ago. This makes managing RAID installations (which need matched drives) very difficult.
Software Issues -- The Xserve shipped with and required Mac OS X Server, and more than anything it was Mac OS X Server’s limitations and frustrations that kept the Xserve behind its rivals over time. However, in the beginning Mac OS X Server had one big advantage: Apple charged no per-user licensing fees. In 2002, that was a disruptive market move and provided ammunition to the switcher movement then underway. When compared to Microsoft’s insanely complex pricing, Apple’s straightforward approach was both refreshing and compelling.
Mac OS X Server does an excellent job of meeting the basic needs of Apple’s traditional customer base in terms of managing user accounts, setting up email, file sharing, and so on. But anyone who tries to extend Mac OS X Server beyond those boundaries quickly discovers its limitations. The problem is that Mac OS X Server’s key benefit was that it provided a graphical interface for the Unix server software running under the hood. Unfortunately, Mac OS X Server’s graphical administration tools have never been fully fleshed out, and system administrators soon learned to bypass them and perform all administration tasks from the command line or with Web-based administration tools. Worse, the graphical and command line approaches were often at odds. Many actions taken in the Server Admin application, for instance, or even upgrades, would overwrite configuration files edited at the command line in ways that weren’t possible from within Server Admin, causing no end of frustration.
This led many to ask, “If I’m better off going into a pure Unix command line administration environment, why should I pay a premium to buy an Xserve and use a slightly strange version of Unix?” Many eventually chose the path of least resistance, switching to standard versions of Linux or FreeBSD on commodity server hardware.
Another big issue is that Mac OS X Server is challenging to integrate into corporate user management systems, especially Microsoft’s Active Directory. Plus, Apple has been slow to allow Mac OS X Server to be virtualized, and places strict limits on how it can be run in a virtualized environment (primarily, that it must run on Apple hardware). These last two items were cited frequently by my peers as huge friction-creating issues when integrating Xserves into large IT environments where directory services and virtualized servers are becoming the norm.
So what began with such promise in May 2002 will finally be laid to rest in January 2011. The Xserve was like that young athlete who blew out his knee before he ever had a chance to compete for an Olympic medal—we’ll never know what it could have become, because it was never allowed to reach its full potential. What was once ground-breaking technology was swiftly outpaced by its competition due to Apple’s neglectful lack of development, along with some initial design flaws in hardware and software that were either slow to be addressed, or continue to nag users to this day.
The Xserve’s demise raises the question of whether Apple is giving up on servers entirely, or just retreating to the Mac Pro (now available in a server configuration) and the Mac mini (which already comes in a server configuration). The latter seems more likely, since while the Mac Pro and Mac mini aren’t appropriate for datacenter use (one is too large and the other too small), they’ll work fine as small office servers. Apple has published a PDF-based Xserve Transition Guide outlining the options. In short, a 12-core Mac Pro is equal to or better than an 8-core Xserve in performance and exceeds it in expandability; the main downsides are the lack of lights-out management and dual redundant power supplies, higher power use, and a vast amount of wasted space in a rack. The Mac mini, on the other hand, lags far behind the Xserve in nearly every way, other than price.
Regardless of what Apple does with servers in the future, for now we can shed a tear for the Xserve—may it rack in peace.