Does Apple Computer Inc. have a role to play in high performance computing for the life sciences?
For years, Apple’s desktop systems have been widely embraced by the scientific community for their ease of use, high performance and affordable price. Yet most serious life science computation today is done on computers from other vendors.
Apple is trying to change that with a one-two punch: the Mac OS X software announced last year (and recently upgraded), and its Xserve hardware systems.
The Cupertino, Calif.-based company is positioning the Xserve to compete with high-end servers from Dell Computer Corp., IBM Corp., and Sun Microsystems Inc. The Xserve is built on Apple’s 1GHz PowerPC G4 processors, and the Mac OS X operating system is based on Unix. Apple asserts that the combination gives users its characteristic ease-of-use features, but with a kick.
The server comes in versions with single or dual processors. Apple says that a dual-processor Xserve can deliver up to 15 gigaflops (or 15 billion floating point operations per second) of processing power.
The Xserve is designed to take one slot in a standard 8-foot data center equipment rack. As many as 42 dual-processor Xserve machines can be stacked in a single rack, delivering up to 630 gigaflops of processing power. In addition, each Xserve can store just under half (0.48) a terabyte of data; a 42-unit rack can support more than 20TB of storage.
Changing Perceptions
Over the years, Apple has not made it easy for its supporters to get their requests for Apple and Mac products taken seriously by corporate IT staffs — even in life science companies where many researchers are Mac aficionados.
With the Xserve and Mac OS X, Apple has gone to great lengths to preserve its reputation for ease-of-use while simultaneously respecting the needs of experienced IT administrators.
An example: For those not familiar with Unix, Apple offers a graphical user interface administration console and wizards to help perform common tasks. This combination lets managers simply click their way through a process such as adding a printer to the network, while the underlying system handles the details of loading the printer drivers and configuring the network.
Such easy-to-use approaches are fine for novices and others not familiar with high-end networking, but for more experienced administrators GUIs are not helpful, and can actually slow them down. “I know the command line entries for most common Unix instructions better than I know my kids’ birthdays,” says one life science network administrator. “The command line entries are intuitive to me. I’ve been using them for years. The graphical interface is not intuitive to me. It’s like when Windows replaced DOS, most people would toggle over to the DOS command line to do simple tasks like moving and copying files rather than going through what at the time seemed like the convoluted way Windows did those tasks.”
Apple, to its credit, recognized that Unix users would not necessarily jump at a GUI, so it has included a command line interface. And the company has tried to win over the Unix purists by duplicating the command line interface exactly — even down to using the same screen color scheme found in Unix systems.
Some experts believe Apple has hit upon the right combination. “It’s an ideal environment where you have access to both Mac and Unix applications through Mac OS X,” says William Van Etten, one of the three founders of Boston-based BioTeam.Net, a consulting group that focuses on creating and delivering vendor-neutral informatics solutions to life science companies.
Apps the Key
One key to the acceptance of the Apple platform is and has been application support. In the past, Apple was criticized for not having as wide a set of applications for its hardware as was available for the traditional PC and servers running at first DOS, then Windows.
So, early on, Apple lined up Xserve support from high-end application vendors that serve the life sciences. For instance, coinciding with the announcement of the Xserve hardware, Apple announced that database vendors Oracle Corp. and Sybase Inc., as well as distributed computing software developer Platform Computing Inc., were supporting the machine.
Oracle announced it would offer a Mac OS X version of its Oracle9i Database and a version of its clustering software, called Oracle9i Real Application Clusters, that would allow users to aggregate Xserve units into a cluster. Platform Computing Inc. said it would give companies a way to use Xserve computers in a distributed computing network through its Platform LSF workload management software.
Interest in Mac OS X applications within the life science community is high. Most of the commonly used informatics analysis programs, such as BLAST and HMMER, have been ported to the operating system and are easily obtained from Apple and other sources.
Other vendors have tapped the Unix aspects of Mac OS X to bring their products to the Xserve and Mac environment.
This summer, for example, distributed computer software vendor TurboGenomics Inc. released a Mac OS X version of its sequence analysis TurboBLAST product — a version of the BLAST program optimized to run in a distributed environment. And the company’s new TurboBench advanced distributed computing software platform, which allows scientists to run applications in a distributed environment while also enabling them to automate workflows, supports Mac OS X nodes.
“We quickly took advantage of the Unix core [of Mac OS X] and put TurboBLAST on it,” said Andrew Sherman, TurboGenomics’ vice president of operations. The process was fairly straightforward. “People who are familiar with Unix should not have any trouble. We found the porting to be very easy.”
TurboGenomics has been interested in the hardware, too. “We came from a scientific computing background and noticed early on that the G4 processor had properties that would be very useful to the life sciences,” said Sherman. “It became clear that many of the things people were doing in bioinformatics were a good fit with the processor and the operating system.”
Earlier this year, Apple and Genentech Inc. took advantage of the G4’s vector processing capabilities and the properties of Mac OS X to develop an optimized version of BLAST.
This summer, the Numerical Algorithms Group Inc. (NAG) released a compiler for Mac OS X. “It’s our first new Mac product in nearly 10 years,” said Rob Meyer, president of the software vendor. “We had compilers for Unix and Linux, but this was the first for Mac OS X.”
Besides its compilers, Meyer’s company sells a variety of applications including visualization and statistical analysis software packages. Most of the applications run on high-end PC, Unix, Alpha and Sun computers.
So why the move into the Mac arena, and why now? Meyer attributes the renewed interest to several factors within Apple itself. For years, Apple looked to three markets — K-12 education, desktop publishing and home users — to keep the company strong. And while many scientists loved their Macs, developers didn’t get a sense that Apple was fully behind them.
“We would hear from Apple insiders that scientific and technical computing were important, but not on the first tier,” said Meyer. Worse for developers, “there would be no help from Apple,” he says. “You would have to buy your own equipment and incur the costs and risks of developing applications.”
But the situation has changed radically in the past year. “We’re getting more support now,” says Meyer. “It’s a ‘meet-you-halfway’ approach, where they lend equipment and help with marketing.”
Developers are certainly paying attention to the Apple market now. “There are about 3,500 native applications available for the Mac OS X today,” said Ken Bereskin, Apple’s director of Mac OS X. “This includes popular Mac applications and many other applications ported over to run on the [operating system].” Nearly every type of application is represented, ranging from office productivity products to high-end computer-aided design programs.
Playing Nice
While Apple is getting its act together on the application development side with Mac OS X, it must still overcome another weak perception before it can fully penetrate the life science market. Specifically, it must convince potential users that its products will fit smoothly into corporate networks.
For many years, while virtually everyone was using Novell Inc.’s NetWare IPX and then TCP/IP for their networking protocols, Apple used a proprietary protocol called AppleTalk.
For those who used only Apple products, AppleTalk was great: Networks based on it were easy to set up and manage. But it was extremely difficult to get computers on an AppleTalk network to communicate with other computers on a corporate network.
However, for several years now Apple has supported standards-based ways to network computers. The company’s computers include an Ethernet jack as standard equipment and full support for TCP/IP, facilitating their inclusion on corporate networks.
Apple recently went a step further: The latest release of Mac OS X Jaguar features many higher-end management functions, including support for the lightweight directory access protocol (LDAP) that makes it easier to tap into existing LDAP servers on other devices. Support for LDAP can reduce the time an IT staff must spend managing a network. For instance, if user access rights and privileges have already been defined for another part of the network or for another piece of networking equipment, and this information is stored in an LDAP directory, the IT staff does not have to re-enter it — Apple systems running Mac OS X can simply reuse the information.
In addition, Jaguar includes a new workgroup manager that makes it easier to configure computers and grant privileges to use them. These new functions, combined with some remote management tools in the earlier release of the operating system, are all critical in order for Apple products to be accepted by IT staffs.
“Functions such as these are essential if the systems are to be placed in an existing network,” says Charles Boatwright, president of Remote Management Consulting, a systems integrator with expertise in Linux and Unix clusters. “Managers need tools that give them information about the state of their servers at any given time and that give them a heads-up to potential problems before they take down a server. They also need tools that help them troubleshoot problems once they happen.”
Xserve seems to satisfy all of these requirements. Managers can use a browser to securely access the Xserve’s management systems over a network. From the browser, the manager can view the status of the server, noting such things as performance (CPU utilization, disk space availability, etc.) and which applications are running.
Additionally, the management tools can head off potential problems by tracking such things as the internal temperature of the server and the performance of the cooling fans, both of which are keys to preventing overheating and resulting crashes.
Such tools from a server vendor are useful, but many managers prefer to use a single, higher-level management system to oversee all the devices on their network. Toward that end, Apple lined up support from Hewlett-Packard Co. before the Xserve was announced so that the machine can be managed via the widely used HP OpenView management software.
This means a manager can use HP OpenView to monitor and manage all the devices on a network, including routers, switches and servers from multiple vendors, Apple’s Xserve among them. “This greatly simplifies things,” says William Perkins, an independent network consultant. “Usually companies use higher-level management systems to keep an eye on their entire network. If a problem arises in a device such as a server, the device passes the information along to a management console, where a systems administrator can identify the problem.”
Remaining Issues
One last challenge Apple will have to overcome is the perception that Apple solutions are expensive.
Managers wanting to tap the features of Mac OS X sometimes, perhaps unfairly, believe its price is high because they must buy Apple or Mac hardware to run it. In contrast, if they already use Windows NT, for example, the cost to move to a different version of Windows, Linux or Unix is merely a software upgrade.
Meanwhile, some researchers have identified another concern: namely, the lack of bioinformatic benchmarks on the system. (At press time, there were no independent benchmark tests available.) Some researchers have run benchmark tests of BLAST on preproduction machines, but are reluctant to share the results.
Remaining issues notwithstanding, the Xserve does seem to be getting the interest of the life science market.
“We’re not likely to see entire grids of [Xserve machines],” says Phil Emer, chief architect at the North Carolina Supercomputing Center. “Rather, we’re likely to see Xserve clusters incorporated into grids.” Emer notes that he’s already had a request from one of his member schools to add a 50-node Xserve system to the existing grid.
Emer, Van Etten and others have traditionally been on the leading edge of deployment. Their acceptance of Apple’s latest efforts suggests that the company may indeed have a role to play in the high-end informatics market.