Enterprise Mac | Tom Yager | October 2006

MORE ENTRIES

October 2006

October 31, 2006 | Comments: (0)

Xserve Xeon review: Cooling and noise baselines

It's pretty easy to measure a system's power consumption at varying levels of load, but how do you measure the burdens of cooling and noise?

Luckily, I operate in a contained space. I can control the temperature to within a couple of degrees of a set point because even in the winter, the servers that I run continuously will heat the room to over 90 degrees unless I run the air conditioning. When it's cool outside, running the room AC wastes less energy than opening a window and draining heat from the entire sealed structure.

I can also control ambient noise. I can render the space completely silent by putting my servers to sleep while a test is underway. There is also enough fan and drive noise in here when the servers are powered up to make a realistic baseline for additive noise measurements.

I measure ambient temperature with an infrared thermometer pointed at a piece of cardboard that's out of the draft of the air conditioner. I'll locate hot spots on the outside of the chassis; this heat is transmitted in the thin space between rack-mounted systems. I'll measure the temperature of exhausted air by averaging the highest readings reflected from cardboard that's swept slowly across fan and convective exhausts.

Why cardboard? It doesn't hold heat. It's always the temperature of the air around it, or, if it's in a draft, the air blown at it. Being a dull surface, it gets along nicely with the infrared thermometer, which gets confused by reflections.

To create an example, I pointed a 1600 watt hair dryer at a piece of cardboard while measuring the heat from the cardboard with the IR thermometer. It rose quickly as the dryer heated up. At 30 seconds, it measured 158 degrees F. When I turned the dryer off, the cardboard dropped to room temperature within seconds.

For measuring radiated heat from shiny metal with an infrared thermometer, a large Post-It note with the non-sticky part snipped off works well.

On to noise. No doubt you've seen some treatment of the OSHA sound pressure level chart, but it's not easy to make a connection between the noise level of a subway and the noise generated by a rack of servers.

Dts Osta Otm Noise Images Common Sounds This chart is fun for kids. It answers the question, "how loud is loud?" However, it does a poor job of conveying the relationship between noise and health. This chart is better than most because it draws the danger line in what experts now concur is the right place. Which color marks the danger line, the level at which permanent hearing loss is likely?

Yellow. If your workplace subjects you to at-the-ear noise levels above 80 dBA (A refers to a weighting scale that is patterned after human hearing), you're at risk. That's not a hard mark to hit when you work near powerful computers and high-capacity storage devices. Right now, my office is 55 dBA. at the moment, I'm testing running Tiger server on a Mac Pro.Mac Pro is quiet, but I must forewarn that the Xserve RAID fibre channel array will not do well. Even so, I'm dying to learn.

There's only one way to measure noise, and that's with a gadget called a sound pressure level meter. A few grand will buy you a nice piece of lab equipment for the purpose, but $50-100 will get you a handheld box built for tweaking home theater systems. Amateur equipment cannot be relied upon for absolute measurements. If the same meter is used the same way for all measurements, then test results can be compared with each other, which is all that matters for my application.

I'll take several SPL readings: Office noise, office noise plus music, office noise plus music plus server., with the server driven to controlled levels of stress.

Then I'll break for lunch.

Thanks for being patient, everybody. It's worth doing right.

Posted by Tom Yager on October 31, 2006 09:44 PM

October 31, 2006 | Comments: (0)

Xserve Xeon review: Introduction and rationale for testing power, noise and cooling

System buyers want equipment that's power efficient, quiet and cool, but to date, little to no effort has been invested in conveying measurements of these important criteria in reviews. Systems of a given architecture tend to show insignificant variations in compute performance from vendor to vendor on integer and floating-point computing tests. They're more likely to vary in their environmental impact--that is, the cost that a piece of hardware incurs to offset its effect on human working conditions. That, combined with the cost of electricity, would rise in importance as determining factors in choosing among brands and models of equipment if the data were available and trustworthy.

I decided to use the occasion of the Xserve Xeon review to start building the foundation for a testing regimen that I can use to benchmark architectures (not systems that share an architecture) against one another. The tests I'm designing and executing for Xserve Xeon will be the basis for evaluating architectures to come, primarily those from AMD and Intel.

The benchmark software used for environmental testing doesn't measure anything. It just imposes a controllable amount of demand or stress on the system. As demand rises, the architecture's power conservation technology plays a diminishing role until, at full compute load with all cores running at 100 percent, the CPUs and core logic cannot throttle back voltage or clock speed, unless flaws in the system's design requires throttling the CPU back to prevent the system from overheating. If system design kicks in CPU throttle-down too early in a rising demand curve, that may flatten externally-measured power consumption, but system logic reports each CPU core's clock speed and voltage level. Systems that wimp out can be found out, but I don't intend to crack open CPU performance counters and the like on the first round.

To stress systems for power, noise and cooling tests, I chose SPEC's Java Business Benchmark (SPECjbb). Java Virtual Machine (JVM) implementations are consistent across operating systems by version, and therefore they create comparable and reproducible results. I'll report environmental test results based on the operating system's reported CPU load, that is, the percentage of total available computing resources in use at each stage in the test. On one architecture it may take four instances of SPECjbb to lock the system at 100 percent utilization, and on another it may take three. The number of instances and the statistical results reported by SPECjbb are tangential, although I will pass them along them as items of interest and because SPEC deserves to have its benchmarks treated like benchmarks and not just tools for making systems hot.

I think that SPECjbb is a well designed and realistic full-system benchmark. Whereas synthetic computing benchmarks stress one CPU core with each instance, SPECjbb spreads its load fairly evenly across CPU cores, and it simultaneously stresses memory and disk subsystems as well. Just like real applications do. SPEC's other benchmarks are better suited to the measurement of best-case raw CPU performance, and I intend to choose from among SPEC's suite for that purpose in addition to developing tests of my own.

Posted by Tom Yager on October 31, 2006 07:28 PM

October 31, 2006 | Comments: (0)

Happy Halloween to your ears

Here's a short seasonal sound clip for you, (remixed here with no ambient sound) engineered by yours truly with Apple's awesome Soundtrack Pro. I ask three things: Play this through speakers (If you must play it through headphones, do so at a very low volume the first time); please don't reveal the contents to anyone; and please do pass it along, but pass a link to this blog entry instead of deep-linking to or copying the WAV file.

While this is harmless aural holiday fun, it has many potential uses. If you have a big amplifier--I'm thinking a guitar amp--play this when the street is packed with trick or treaters. At a red light with your windows rolled down is fun, too, provided that you and your passengers react appropriately. Or a friend may appreciate having this as a shutdown sound for their new copy of Vista.

My employer and I are not responsible for any consequences resulting from the downloading or use of the sound clip linked to this blog post.

Details: This project uses seventeen distinct audio samples layered in nine tracks. The engineering and mix-down to stereo were done in Apple Soundtrack Pro v1.1 on a Mac Pro workstation with two 3 GHz Core 2 Duo Xeon CPUs running OS X 10.4.8 (Tiger). The samples are all taken from the Apple Loops for Soundtrack Pro DVD included with the product. Soundtrack Pro is a component of Final Cut Studio ($1,299). A prosumer version, Soundtrack v1.5, is part of Final Cut Express HD ($299).

Posted by Tom Yager on October 31, 2006 12:37 PM

October 26, 2006 | Comments: (0)

Xserve Xeon review: Pause for photos, answering readers about a final concise review, coming attractions

Don't go away; we'll be right back.

Turning my notes into a review, and (as you can tell) questioning and validating my findings as I write, is a serious challenge. InfoWorld and other vendors I've put on hold while this proceeds deserve credit for giving me room to take such an unusually deep approach. I'm keeping my head in the work, not checking traffic or comments. However, If you're reading the series and you have an opinion about the content or my approach, please do file a comment. I may not post comments very often--the volume is enormous, and I'm grateful for that--but I read them all and they help set my course.

Much as I'd like to continue to give this my exclusive attention, I do have to see to other things. I have a column to file, and I've been getting requests from readers for photos to go along with this series. That's probably going to fill the day, but I expect to file some more tonight.

So much about Xserve Xeon's engineering is unique that I can't lean on the standard reviewer's shortcut of comparisons with familiar equipment. And unlike other reviews I've written, I find that some of Apple's more unusual design and engineering choices deserve exploration and explanation rather than reportage. So please hang with me. This is a server worth understanding even if only as a yardstick against which other efforts can be measured.

Just so you know what's coming, the next three segments will be:

Power, cooling and noise: The most detailed look I've ever taken at a system's environmental and ergonomic characteristics. A deep dive on Apple's complete rethink on 1U server cooling, how Apple's temperature monitoring and fan control algorithms work, the impact of ambient temperature on operation, power draw and noise at controlled workload levels

Performance: SPEC Java Business Benchmark (SPECjbb), SPECcpu, STREAM, disk I/O and new multi-threaded encryption and compression benchmarks, compared with Xserve G5 and Xserve Xeon running Windows 2003 Server

Networking and virtualization: NFS and SMB (Windows) file sharing, the impact of high peripheral I/O burden on memory-intensive apps (the Intel shared bus test); running Windows 2003 Server as a guest OS under OS X Server using Parallels

It's a lot, but this box deserves all this attention. I really appreciate yours.

Posted by Tom Yager on October 26, 2006 12:12 PM

October 25, 2006 | Comments: (0)

Xserve Xeon review Part 2: Beyond Intel

Look harder at the hardware
With that subheading, you expect me to drone on here about the intangible value of build quality, great software and other Apple platform goodness that transforms the same-old, same-old $1,500 to $2,500 Intel rack server into something worth $2,999. Buyers of Intel servers are accustomed to comparison shopping based on hardware specifications and price, so let's stick with hardware. The Woodcrest rack server basics that I detailed in Part 1 are a given. Apple went its own way even in its engineering of those features required to qualify Xserve Xeon to compete in its class. But, except to appreciate that Apple hasn't lost its touch for server engineering, Apple-ness doesn't raise the worth of Xserve Xeon. The direct value that Apple adds to Intel's standard Woodcrest system design is easy to see if you open yourself to the possibility that putting Intel chips in a server needn't define the boundaries of a server's capabilities.

SAS, SATA, and DVD-R DL
Server-attached storage has gotten a bad rap for so long that vendors usually assume that 1U rack servers will use networked or external storage. Many types of server deployments benefit from the simplicity and speed of local storage, and to that end, Apple designed Xserve Xeon with three bays that accept removable Apple Drive Modules. Unlike the Mac Pro workstation, which you can expand with raw SATA drives, Xserve Xeon requires that you buy drives from Apple that are pre-mounted in Drive Modules. Fortunately, Apple's markup for Drive Modules is reasonable--lower than other storage and server vendors with which I'm familiar. With three 750 GB SATA drives, 2.25 TB of local storage, an Xserve Xeon might take on roles you wouldn't ordinarily assign to a standalone server. A giant swath of local disk is a real boon for grid applications.

The value twist that Apple put on Xserve Xeon's removable drives is that each of the drive bays will accept either a SATA or a Serial-Attached SCSI (SAS) Drive Module. These two busses share one backplane physically and, it seems, logically: My Xserve Xeon eval unit shipped with three SATA drives, but the system enumerates them in its hardware inventory as nodes on a tree of SAS/SCSI devices. This is a cozy arrangement; Xserve Xeon's internal drives and Fibre Channel-connected Xserve RAID volumes are all connected to logical SCSI adapters.

If you're willing to sacrifice speed for capacity, SATA is for you. If you can't bear the thought of life without 15,000 RPM drives, go SAS. Boot from SAS and put SATA in the other two bays, boot SATA and have a screaming RAID 0 SAS stripe on bays 2 and 3, whatever you like. With that flexibility, it's much easier to plan for the use of hard drives for backup and portable storage. That strategy is advanced by OS X Server's standard GUI and command-line tools for doing file and folder-level archiving with compression, creating virtual disk images that have compression and/or encryption applied, and creating formatted disk images that are ready to burn to DVD. OS X Server Tiger reduces full-volume imaging and restoration to single operations.

When OS X Leopard ships early next year, Time Machine will make backup continuous and automatic. I'm not sure how Time Machine will operate with removable drives, but given that they're built into Mac Pro and Xserve Xeon, I'm full of ideas.

Every so often, Apple rolls out a feature that could have been just for me. Apple's operating system install discs are now shipped as DVDs, but the Xserve G5 sitting behind me has only a CD reader. Apple not only made a DVD-ROM drive standard, but it now offers a DVD +/- RW, dual layer drive as an option. Who needs a DVD burner on a server? I've got an external drive connected to my Xserve G5 now, and I build my PC servers with burners as a matter of course. Every server should have as many modes of data transport as possible. There are times when sneakernet is the best or only way to get from point A to point B.

BIOS is dead, long live EFI and LOM
Every Intel-based Mac has shipped without a feature that's a core design necessity in nearly all Intel x86 client and server PCs: The IBM PC BIOS (basic input/output system). That doesn't make sense. After all, the BIOS is part of the PC standard. Why break that? Let me count the reasons. No, let's save time with a brief rundown on what EFI, the Extensible Firmware Interface, is.

EFI is an embedded programmable firmware subsystem developed by Intel. Intel publishes detailed specifications and reference code on its site, which makes the word "extensible" plausible enough. You could look at EFI as a little system and OS that boots before your big system's OS boots. EFI uses device drivers, written by vendors or contributed from the community, to locate, initialize and even interact with hardware through scripts. With drivers that give it the ability to access a system's hard drive and understand its file system structure, EFI can load and execute scripts from that drive. It can negotiate booting from the LAN, and Intel's EFI reference code includes an interactive EFI command shell that lets you do some rather unexciting exploring before Xserve Xeon (or any Intel Mac) boots. Apple took advantage of EFI's flexibility to make the Boot Camp Windows compatibility hack work, and in the run-up to Vista, Apple appears to be working on EFI to ease direct booting and installation of non-EFI-aware OSes. In short, BIOS bad, EFI good, and Apple made the right choice.

Elsewhere in firmware, Apple has come around to a new idea: If Xserve requires a successful full boot of OS X before Xserve can be managed remotely, then diagnosing and correcting a boot failure is a tough job. Enter lights-out management, which Apple abbreviates LOM. Lights-out management is enabled by an autonomous embedded system that responds to management queries and commands without the OS's involvement or users' knowledge. Apple's LOM implementation conforms to Intel's Intelligent Platform Management Interface (IPMI 2.0) specification, with the exception of serial over IP, which is the rerouting of the serial port through a LAN connection. As with other approaches to lights-out management, Xserve Xeon keeps an embedded processor running even when the main system is frozen, sleeping, powered down or unable to boot. That processor listens for IPMI traffic on both of Xserve Xeon's on-board Ethernet ports. It does this without interfering with the server's normal TCP/IP operation. No matter what's going on with your server or networks, if you can work out a physical path to one of Xserve Xeon's on-board Ethernet ports, you can interact with LOM.

Theoretically. My Xserve Xeon review unit arrived loaded with a private post-release build of OS X Tiger Server (10.4.8). In that release, the Setup Assistant that runs on the system's first boot asks for IP addresses to assign to lights-out management. This, I learned, requires some planning because I fiddled with it long past sunrise and never got it right for my LAN. I did manage to get Server Monitor running on Xserve Xeon to talk to its local LOM, and located there the option for reassigning the LOM's IP addresses. I'll have to come back to the LOM; miles to go and all that.

Impressive connections
Earlier, I said that a system should have as wide a variety of data transfer modes as possible. Apple nodded at this with Xserve Xeon's versatile removable hard drives, but under the urging of throughput-hungry customers, Apple kept going. Its new server has a total of three FireWire ports: two that operate at 400 or 800 Mbps, and one that operates at 400 Mbps. There are two USB 2.0 ports as well, but external storage and digital media devices are at home on FireWire.

Servers are just at home on FireWire. A seldom-recognized benefit of Mac FireWire ports is their usefulness as dedicated point-to-point TCP/IP links. The FireWire port on Xserve Xeon's front panel is certainly there for external media devices, but it's also a means for a FireWire-equipped Mac or Windows PC to move data to and from the server, establish a speedy Remote Desktop link, string a dedicated line between servers for replication and fail-over, or do anything else that one can do over Ethernet.

Xserve Xeon has risers for two expansion cards, and Apple puts a curious spin on this, too. Both of the bus slots are 8-lane PCI Express, with room for one 9-inch card and one 6.6-inch card. However, the 6.6-inch slot can also accommodate a 133 MHz PCI-X card. PCI-X was the Xserve G5's bus, and Apple didn't want existing customers to add the cost of PCI Express replacements for PCI-X cards in perfect working order.

Lastly, Apple didn't cheap out with the on-board graphics. Yes, Xserve Xeon is a server, but it's also a Mac. The ATI X1300 Radeon GPU (graphics processing unit) is fast and it outputs to either DVI (LCD) or VGA. To save back panel space, Apple uses a nearly-invisible mini-DVI plug. Thoughtfully, Apple includes mini-DVI to SCSI and mini-DVI to DVI pigtails. It plugged straight into a 30-inch Cinema Display and displayed a maximum resolution of 1280x1024 at 32 bits. You wouldn't play video games on Xserve Xeon, but it is Quartz Extreme compatible, meaning that all of Apple's complex graphics functions, such as native PDF rendering, can be done directly on the GPU.

Posted by Tom Yager on October 25, 2006 11:46 AM

October 25, 2006 | Comments: (0)

Xserve Xeon review Part 1: Introduction

[I have elected to split this review into parts so that readers can view only those sections that matter to them. Note, too, that this has not passed through InfoWorld's Copy department, and it shows.]

At the August 2006 Worldwide Developer Conference in San Francisco, Apple's Steve Jobs announced that Xserve Xeon (my name--Apple refers to it as Xserve), its latest rack-mount server, would ship in October. It turned out that Apple didn't get Xserve Xeon into full production in October, but I was honored by the surprise delivery of one of the first Xserve Xeon units produced. I've seen and handled Xserve Xeon several times, but never powered up. The box arrived five days ago. I plugged it in and at that point, I unplugged my phone, my inbox, my BlackBerry and my clocks. Once I learned that Apple would start accepting pre-orders on October 25, I was determined to put a full review in the hands of prospective buyers, as well as those inclined to see Xserve Xeon as a two socket, four core Woodcrest rack server, indiscernible from the imagination-smothering overgrowth that is the Intel server market. For existing Xserve G5 owners and others running OS X Server, Xserve Xeon raises questions about whether it's worth a switch to Intel and what the risks and benefits will be.

Apple's new server is a story, not because I'm Mac-aligned, but because if Xserve Xeon were stripped of the logo, the aluminum and the trappings of Apple culture, it'd be hailed as a server design breakthrough. Xserve Xeon is an entry-priced Intel rack server engineered to satisfy mid-level server buyers' expectations. Compared with competing Intel rack servers based solely on performance, total hardware design, features, build quality, expandability, adherence to standards, manageability and serviceability, Xserve has no better among two-socket Core microarchitecture servers in the sub-$5,000 category. But to put Xserve Xeon in a fair lineup against other Woodcrest rack servers, you would either need to subtract $999 from Xserve's $2,999 base price or add the cost of a preconfigured, unlimited seat license of either Windows 2003 Small Business Server or SUSE Linux Enterprise Server to the other contenders' price tags.

Unlike other Intel OEMs' boxes, Xserve Xeon ships from Apple as a server platform, not as a server. The difference? A server platform is a whole--hardware, OS, standardized services, GUI management, dev tools, server application frameworks, documentation and so on--that emerges from its shipping carton with functionality that fully satisfies the needs of the majority of buyers. When I plugged Xserve Xeon in for the first time, it did what I expected: It gave me my choice of services including Web, database, J2EE, e-mail, blog, IM, Windows and UNIX file/'print sharing, gateway, proxy, firewall, and on and on. It took about three minutes, with no reboot, to put all of those services on the air and hook them into a single GUI management interface. At the other extreme, Xserve Xeon will boot to a UNIX text console or to the server's serial port, it can be administered and configured entirely from the command line and nearly all of the non-user-facing pieces of OS X Server are available as Apple-supported open source. A platform is a whole, but it can be an infinitely malleable whole, at least as Apple approaches the Xeon Server platform. (you could just as well call it Apple's OS X Server platform; the hardware and OS deserve equal billing)

Xserve Xeon is not flawless--there are design challenges presented by compact rack servers than even Apple hasn't overcome--and the above is nothing near representative of the case that Xserve Xeon makes in its favor.

Nothing to see here
One Apple advantage that Xserve Xeon does not ply is a novel appearance. If you've seen one Xserve, you've seen them all. Truly, if you got an eyeball high from your first peek at Xserve's puss years ago, you won't be reliving it soon. Xserve Xeon is the spitting image of the PowerPC Xserve models that preceded it. I'm glad for that; Xserve's face has matured from art to function by its continuity across Xserve's generations. As a practical note, matching faceplates also make people with big racks (so to speak) happy because unlike other first-tier vendors' products, Xserve doesn't show its age. It seems a low priority, I know, but I shouldn't be able to estimate your yearly server budget from one glance at your server cabinets.

Faceplate aside, next to nothing was brought forward from prior generations of Xserve. Xserve Xeon is not Xserve as you've known it, and if you're well familiar with x86 1U (one rack unit = 1.75 inches tall by 19 inches wide) servers, you'll find that Xserve Xeon is unlike any x86 server you've seen or used. That's what I discovered, and the more I explored Xserve Xeon, the more pleased and impressed I was to find that Xserve Xeon shares no genetic link with Xserve G5 or with other vendors' Intel rack servers. It can't be said that Apple started with a clean slate; this is, after all, an Intel-based server whose most basic technical specifications are easily matched. Here's what Xserve Xeon's base configuration has in common with the whole world of Woodcrest rack servers:

Dual 2 GHz Xeon 5100 Series CPUs, two logical CPU cores each
One 1.33 GHz front-side bus per socket (each core pair gets one FSB)
1 GB of 667 MHz DDR2 fully buffered DIMM (FBDIMM) memory
Dual on-board gigabit Ethernet ports
Nine-pin serial port, two USB ports
Three removable drive bays, one 80 GB SATA drive standard
On-board display adapter
PCI Express expansion riser

Any vendor can do that. For that matter, give me a screwdriver and $1,500 and I can do that. If you look at Xserve Xeon with Intel OEM tract server expectations, you'll only see what I listed above: Qualities that Xserve Xeon has in common with common Intel servers. You'll be hard-pressed to see how Apple makes this add up to $2,999 or how I can call Xserve Xeon an entry-priced, mid-grade server.

Posted by Tom Yager on October 25, 2006 05:51 AM

October 25, 2006 | Comments: (0)

Apple to miss October Xserve delivery, will start accepting pre-orders today

I'm camped out on Apple's Xserve page, breathlessly awaiting the emergence of Xserve Xeon from its "notify me when available" state. It's not going to leap from that to "buy now" as those in Apple's server program had hoped and planned. Apple can't make its October deadline, so it's taking the intermediate step of creating a pre-order queue. I know that practice strikes some as cheesy, but I'm glad Apple's doing it. Slipping a ship date creates a panic at Apple, probably a legacy of the bad old days when Freescale and IBM decided when Apple would ship a new Mac.

I see Xserve Xeon's slip this way: With IBM, HP and Dell homogenization of Core microarchitecture Intel rack servers, Apple's got to make its first shot at an x86 server count. In servers, the Apple brand carries no weight (yet). It'll all on the machine and OS X.

That's worth taking a little time to get right.

As for whether it's worth your while to get in line for your very own Xserve Xeon, I can help you with that. Keep reading.

Posted by Tom Yager on October 25, 2006 01:44 AM

October 24, 2006 | Comments: (0)

Core 2 Duo MacBook Pros arrive

Woke up this morning and what'd I see? A mobile landing pad for Leopard in plenty of time for Macworld Expo.

Apple is now shipping its new Core 2 Duo-equipped 15-inch MacBook Pro, with the 17-inch MacBook Pro notebook going out in about a week. These 64-bit notebooks offer a few, yet important advantages over their Core Duo predecessors:

First 64-bit Mac notebooks
First Apple notebooks based on Intel's 64-bit Core microarchitecture; Core Duo was dual core, 32-bit Pentium-M
Apple claims performance improvements of 39 percent over Core Duo MacBook Pros
Dual-layer SuperDrive DVD burner in all models (6X on 15-inch, 8X in 17-inch model)
400 and 800 Mbps FireWire ports throughout lineup
Expandable up to 3 GB of memory (see below)
Hard drive capacity of up to 200 GB
Build-to-order options now include MagSafe airline power adapter (see below)

Apple's approach to expanding MacBook Pro's RAM is to factor in availability of 2 GB DDR2 SODIMM memory. These are still 2-socket systems. To get to 3 GB, you put a 2 GB DIMM in one slot and a 1 GB DIMM in the other.

The MagSafe airline power adapter hooks directly from the power jack under your seat to the charging connector on your notebook. I'm watching this one closely. At present, I can't run my MacBook Pro 17 (Core Duo) on under-seat power on American Airlines flights. American uses 12V "cigarette lighter" jacks. Previously, that required plugging the full-sized charger into a DC-to-AC inverter, and from experience, MacBook Pro exceeds the 75 Watt limit of the underseat outlets. Apple tells me that with the airline adapter, power draw stays within American's imposed (and enforced) limit.

Until I get my hands on one, that's as far as I can take it, save whatever analysis occurs to me as I start digging around in earnest. In any case, it's an important day. Now if you want a 32-bit Mac client, you'll have to go low, down to the MacBook or Mac mini.

Posted by Tom Yager on October 24, 2006 05:46 AM

October 16, 2006 | Comments: (0)

Will Leopard get buried by Vista buzz?

If you were counting on Leopard gaining an advantage from shipping before Vista, it looks like you're out of luck. Vista is in its last test release before shipment, and historically, the final release candidate becomes the supported RTM (ready to market) product. OEMs and volume licensees are expecting November delivery of the finished Vista, and the onesie twosie, shrink-wrap buyers will see Vista early next year. Subscribers to MSDN (Microsoft Developer Network, Microsoft's counterpart to the paid flavor of Apple Developer Connection) will probably get the RTM code sometime between those two dates.

I'm not being too cynical, am I, to suppose that despite all the long faces and equivocal mumbles by MS execs when asked about Vista availability, November was never really in doubt? I mean, it is nice to have your bases covered; no one would have been surprised or disappointed if Vista's initial deliveries to VIPs slipped into '07. But Microsoft has been telegraphing impending availability to partners, developers and even end-buyers with marketing campaigns that talk up Vista as though it's already gone platinum. It might as well have. Microsoft has pressurized the pipeline by teasing buyers and developers with seemingly uncertain availability. The media's shameless mockery of Microsoft for having blown Longhorn/Vista targets time and again has helped generate pent-up interest in Vista by turning mere delivery of the OS into page one news. (watch the headlines; the word "finally" will work its way into a lot of story titles)

To the dismay of so many observers, when Vista goes RTM, it'll explode. It'll shatter every standing record for software units and revenue in a given period of time. It'll bury everything else in a mudslide of press coverage, and it'll even make story #4 or #5 in non-technical publications and network news. I predict a BusinessWeek cover.

One could say that Vista's got a greased track to record sales: PC models that currently ship with Windows XP will begin shipping with Vista. Perhaps Microsoft will offer an install-time choice between OSes the way it initially did with Windows XP and Windows 2000 Professional. In any case, every PC with Vista pre-installed, even if unused, counts as a sale. I'm not suggesting there's anything dishonest about this. I'm just preparing you for mind-boggling Vista numbers in case you might find such statistics disturbing.

So Winter will be all about Vista. Where, oh where does this leave Apple? When Vista reviews start running, we'll see the professional and lay media pile on with the predictions of doom for OS X and the "too little, too late" Leopard that journalists have never seen outside apple.com. Vista is it, and Macs are at a disadvantage for not shipping with Vista. Journalists of that ilk will null the value of OS X and insist that the true cost of a Mac is the machine plus the copy of Vista that really makes it work. There will be plenty of reasons for the Mac faithful to sound a call to arms.

Journalists love to proclaim the ends of eras and the dawning of new ones. For example, the fact that Intel shipped Core microarchitecture CPUs meant that the sun has set on AMD.

The media wags and flame-baiters will have a nice run through early January until MacWorld shuts them up. Steve Jobs had a wink in his voice when he projected Leopard's Spring delivery during his WWDC keynote.

For Apple, there is no Spring (the tune, "in heaven there is no beer" inexplicably just came to mind). Apple has exactly two opportunities per year to draw mass press and public attention: January and August. I suppose that Apple could view some combination of 64-bit MacBook Pros, eight-core Mac Pros and probably the announcement, but not delivery of SAS/SATA Xserve RAID as a full marquee for Macworld Expo. But I think it would be a strategic problem if a full round of 64-bit gear went out with Tiger, with Leopard held out almost as a Vista-like tease (and a $$ upgrade for very recent buyers).

Apple's Senior VP of Software Engineering, Bertrand Serlet, who I think is the sharpest knife in Apple's executive kitchen, laid the foundation for a Vista vs Tiger battle with a funny and convincing comparison of Vista's GUI and its few bundled apps with the Tiger GUI and app features from which Vista's great ideas were derived. Serlet's message was that even if Vista lands before Leopard does, Tiger will still show Vista to be a derivative work. Seen another way, Serlet's presentation was meant to show that Microsoft had plagiarized elements of Tiger while Apple had already gone several laps past its own best OS/application environment.

It's interesting that although Microsoft borrowed heavily from Tiger's look and feel, Microsoft didn't capture the human-factored behavior that spawned Tiger's visual elements. At its heart, Vista is Windows, plus a collection of modernized UI widgets for developers and a bucket for the marketable ideas that emerged from Microsoft Research. Tiger and Leopard bake consistent behavior, look and feel and integration into everything from its dev tools to its Web browser. Vista can't go there: Windows will always be an operating system. Don't get me wrong; Vista is a huge step for Windows, a real godsend for those stuck in the Windows XP rut. But Apple will retain the state-of-the-art title, and applications will still rank it #1 in their compendium of best places to live.

Oh, and did you catch the quip in Steve Jobs' WWDC speech about Leopard not requiring on-line activation? I wonder if journalists bring this up. Perhaps the best of them will go searching for Apple Genuine Advantage or its like and marvel at its absence. I expect, though, that we'll read the supposition that Apple's IP protection for its client platform is covert.

Posted by Tom Yager on October 16, 2006 02:32 PM

October 16, 2006 | Comments: (0)

Xserve Xeon: Upgrade your own CPUs, but AYOR

Apple's laudable and full embrace of the x86 system architecture standard creates an interesting possibility that prior Macs (other than Mac Pro) haven't offered: CPU upgrades. Xserve's Intel Woodcrest CPUs are seated in simple, x86-standard ZIF (zero insertion force) sockets. You just remove four ordinary screws holding down the very pricey full copper heat sink, lift one lever, and out pops the CPU. The only inconvenience is that you'll have to acquire a tube of thermal goop.

When competing first-tier vendors are found to be locking standardized components down, please take them to task and get an on-the-record explanation regarding their efforts to make off-the-shelf parts non-replaceable. I do ask, and I've heard: "Our advanced heat sink design requires riveting it to the bottom of the chassis;" "We run our [marked-up] CPUs through a more rigorous validation process than Intel does;" "We find that few of our commercial customers wish to service or upgrade their own systems;" or, as you've heard from me before, "The minimum field replaceable unit is the system." Of these, only the last holds any water. If you've ever had what passes for a PC vendor's on-site service call, you know why (I've had no experience with Apple's on-site service).

Intel's next CPU, Clovertown, is on deck for early 2007. This four-core uses the same pinout as dual-core Woodcrest, and the die is exactly the same size. Intel built Woodcrest to so it could swap out with Clovertown. That drop-in replacement design (which AMD pioneered; Intel will not keep Server CPUs socket-compatible for three years) is primarily a build-time cost saving for OEMs. Woodcrest and Clovertown machines at a range of clock speeds can be built using the one motherboard and a single assembly line. But that also means that you can yank a Woodcrest and drop in a Clovertown, or a faster Woodcrest, as long as the system's firmware doesn't preclude it.

I wouldn't try it with any system in production that's under warranty.

So, why should I celebrate a capability that I advise readers not to exercise? It's simple, and I say it time and time again: How vendors build their systems speaks volumes about their attitudes toward their customers. It's cheaper to nail down components than to make them easily replaceable, and nailing them down improves the chances that buyers will buy new servers when they need faster ones. Apple could have gone that way with Xserve. Instead, Apple has embraced the Intel x86 standard system architecture in a way that would pass muster with purists.

Although Apple took the off-the-shelf route, it didn't get there by buying a tract chassis or punting logic and layout to Intel's reference design.

Now, if you asked me whether I thought yanking and upgrading your CPUs was a good idea, I'd spend so much time grilling you about your qualifications that Apple would be shipping eight-core Xserve before I let you out of the room. The short version of the cross-examination amounts to this: How many rack servers have you assembled entirely from components? How many of your homebuilts has an employer of yours trusted to run in production? List at least four common mistakes things that can a self-installed CPU. And finally, in the unhappily likely event that you fry your Xserve while you're modding it, can you afford another one? I hope so. Apple doesn't put forth a specific policy, saying only "We don't support upgrading the CPUs." Apple is careful to avoid saying explicitly that swapping CPUs voids your warranty, but I expect that dispensation from AppleCare on a CPU upgrade is contingent on the number of Macs you buy per year. Virginia Tech would probably get away with it. You probably couldn't.

Posted by Tom Yager on October 16, 2006 01:51 PM

October 16, 2006 | Comments: (0)

Apple's Xserve Xeon: Built to fall apart

[Please note: Contrary to the link on Apple's home page, this is not my review of Xserve Xeon. That is coming shortly; watch this space. I understand the reason for the minor mix-up and I always appreciate links to my content. However, I have not tested Xserve Xeon. I have only been allowed a guided, semi-hands-on encounter with the box at Apple's Cupertino campus. This entry addresses build quality, one of very few attributes I'm willing to address prior to one-on-one production testing.

Don't touch that dial. I promise you that when I hit Xserve Xeon you'll be the first to know. My whole life's been leading up to this.]

A couple of weeks ago, Apple invited me to its campus to get a close-up look at Apple's Xserve Xeon. It is a marvel of physical design, so much so that I find that it implausible that Xserve Xeon and Xserve G5 could have been designed by the same company. Xserve G5 was pretty tight, but Xserve Xeon makes its predecessor, not to mention every PC 1U rack server I've seen, look slapped together.

I was struck by perfectly Xserve Xeon was designed, and in particular by how easily it comes apart. I have high standards in this regard. I told a friend that I will only buy or recommend servers that I can install, remove, disassemble and repair with one hand, a TSA-approved butter knife and no instructions. While others are marveling over the blue LEDs on the front panel (which, yes, now reflect the load of four cores), I'm looking for blobby spot welds and for paper-thin steel that's been bent into a U to grip a cooling fan (once). I look for hacks done to make a server interior "tool-less," such as levers hooked to long rods hooked to blades that work as embedded pry bars. In that well thought-out design, the tool-lessly removable part is exactly where it would be in, and uses exactly the same type of connector as, a tool-fully serviceable server. A sexy server assembled with a welder, a metal brake and a torque wrench is like a sports car that drives perfectly well, but which requires a pneumatic nut driver to open the door.

Apple's Xserve Xeon falls apart with the slightest touch, and I like that. I don't mean that it's fragile. Like most of what Apple has made, Xserve Xeon is obviously ruggedized to survive a fall. Not that any server would ever slip out of your rack, and no reader of this blog has ever tipped a rack over by sliding out that 3U RAID rack you (I mean, that ham-handed assistant of yours) mounted too high. Apple troubled to make connectors that friction fit very tightly, yet don't need to be wiggled loose. As with other rack servers, Xserve Xeon's fan bank is removable. But in Xserve Xeon, it lifts right out with almost no effort. Likewise, the logic board is not rectangular. It snakes around the chassis like a road laid on a riverbank; by my recollection, it never ducks under other components. It was not demonstrated to me, but it appeared that the motherboard could be unscrewed (it uses fat, grippy thumbscrews) and lifted straight out. There are very few cable connections from the motherboard to the chassis and to peripherals, and no opportunity to get a cable in the wrong spot. I should also note that there are no long ribbon cables stretching across the chassis to connect a drive that's planted up front to an on-board SCSI connector near the back. The chassis was designed first, and the logic boards were designed to fit it. What a concept.

You can see that I am a connoisseur of server design, and that I have no tolerance for rack servers that aren't built to micrometric tolerances. I'll test this theory when I get an evaluation unit, but it looked to me like any piece of Xserve Xeon, from the logic to the power supplies to the fan bank, and even the CPUs, can be removed and replaced in a five-minute operation while eating pizza. Apple even redesigned the rack attachment to--get this--eliminate the need for cage nuts. Yes, someone at Apple stood up in a design meeting and said, "look, this two motors per fan idea is nice and all, but can't we do something about those darned cage nuts?"

Xserve Xeon is, as all commercial servers should be, built to be user-serviceable. You can buy packaged spare parts kits and get AppleCare walkthroughs of parts replacement. If you have multiple Xserve Xeons, you can scavenge and swap parts, just as God intended.

Now, having found myself in this situation before, I feel compelled to ask readers: Is my obsession with operator-centered design mine alone, or do people in the real world make buying decisions based on a server's build quality? Do you have a 1U x86 box in your rack that you could disassemble and put back together, sans diagrammatic assistance, more easily than a kid's bike? I know that's desirable. I just don't know whether anyone thinks that's worth $2,999 when a perfectly functional, but horribly built two-socket x86 rack server can be had for half that. It is typical now that the minimum unit of field replacement is the entire system. Is that just the way of things?

Posted by Tom Yager on October 16, 2006 07:05 AM

October 06, 2006 | Comments: (0)

Professionals' ten biggest objections to buying Macs

Some readers have e-mailed me to ask if I'm still breathing. Thanks, folks, for the concern. Yes, I remain above ground in body at least, and my brain is about to join me there.

The fact is that I've been too busy writing and learning about the Mac to blog about it. My biggest recent print project, InfoWorld's Special Report on Apple in the Enterprise, finally saw daylight on September 22, along with my ten word review of the richly-appointed quad core Xeon Mac Pro. I also managed to get a few words in on Leopard.

Coming up, I'm blogging the fruits of my detailed briefings with Apple on Xserve (Xeon), due this month, and Leopard Server.

Thanks for staying tuned.

Posted by Tom Yager on October 6, 2006 03:53 PM

Technology White Papers

InfoWorld Technology Marketplace

» Technology White Papers Library

Technology White Papers by Topic

Technology White Papers E-mail Alert