You contradict yourself over and over again and have demonstrated a lack of understanding on how to measure performance.
It seems, from you post, that your major concern is performance of a certain microwave antenna simulation program. Everything else should be secondary - the clockspeed, memory latency and throughput, the name on the outside of the box, etc. Now, you may have a price issue... for which then you want the best performance (which is most purchasers outside of governments). Whatever anyone says about performance is useless to you until you intimately understand the performance requirements of your particular application. The easiest way is to run it on various systems yourself as well as use various code analyzers. From the mistakes and misconceptions in your post, I doubt you really understand the performance issues with your application.
What is Intel's fastest CPU for floating point intensive operations? The Itanium2 is definitely their premier performance CPU for floating point operations. Since we don't have the inclination to examine your particular application, even if we had the source and the time to examine it, we'll have to use a rough approximation - SPEC.org's floating point CPU tests - SPECFP2000 results. Again, we're simplifying... everything that goes into a system affects performance, but for now:
HP Integrity Server rx4640 (1500 MHz, Itanium) - 2161
IBM eServer pSeries 655 (1700 MHz Power4) - 1776
AMD Opteron (TM) Model 248 (2.2GHz) - 1691
Intel 3.4 GHz, Pentium 4 Processor Extreme Edition - 1561
Again, the aggregate SPECFP2000 results probably don't correspond to the performance of your application - maybe a specific subtest does. Plus, there are any number of items to tune including the algorithm itself (if you have source), the compiler (from selection to flags), the operating system, hardware options, etc. But the fastest clock rate system (3.4GHz Pentium 4 Extreme Edition) above is the slowest in aggregate SPECFP2000, and the slowest clock clock rate CPU has the fastest result (1.5GHz Itanium2).
We haven't even gotten into whether or not your algorithm can leverage SIMD (as in the G4 or G5's Altivec) for major increases in performance, whether it is really memory bound, etc. Sometimes while you think your need major floating point, your algorithm is either not what you expect or is bounded by an integer operation that may be unnecessary - and therefore you are not leveraging the FP capabilities of certain chips. We haven't looked at whether or not the algorithm can take advantage of parallel systems, and if so, if it can use loosely coupled systems w/o losing too much performance. It goes on and on - and we haven't even tackled the issue of price. But in the end, this statement:
And my conclusion was that raw clock speed IS the most important factor in performance
I didn't want a Palm because I wanted a more functional mini-computer. I own a Zaurus SL-5500 and was set to buy a SL-6000 except for two problems. Price and the lack of Bluetooth. I would have still gotten one at that outrageous price ($700) if it had Bluetooth. Chances are - if I am in range of Wifi, I have my laptop. I need a PDA when I'm not near Wifi and want to access the net over GPRS cell phone - hence the need for Bluetooth.
Further, the relatively lackluster open source-ness of the project and Sharp's seemlingly lack of interest (and the new horrible contacts APIs) seem to seal its fate... I hate it, but only Microsoft would probably have the staying power in the long term.
1) Import video from camera to Dell in DVD formatted MPEG-2
Mistake right there. The question is, why did you do this? Why would you purposefully use a lossy codec for source material and then complain about video quality degradation?
BTW, iMovie takes DV format which is what you get when you grab video over firewire from a firewire equipped camcorder. This makes the workflow much easier for most people - hence that is the way iMovie works. Final Cut Express and Final Cut Pro support more import formats. iMovie expressly does not support taking in anything other than DV format - it's the way 99% of the audience will get their source material. Grabbing it and converting to MPEG-2 before working with the material is just plain idiotic. It's an artifact of the cheap video card you bought in that Dell.
Anyways.. if you insist on doing this, you could have used the free VideoLAN VLC to do the transcoding back to DV. The MPEG-2 playback component you bought enabled you to transcode also and you noticed that it degrades quality. But of course - that's why people don't do it that way.
If your source isn't something that has firewire output, then you can get an analog to firewire bridge or a PCI analog capture card. There exists USB versions, but the quality loss is significant. Seems like you paid for a video capture card in the Dell and didn't for the Mac.
You're doing something that most people won't do... and expressly not supported - it all comes down to iMovie only taking DV format (btw, which the Dell can produce). That's a choice that Apple made with their low cost editing suite and most people don't have a problem with that. The problem of transcoding is the same, regardless of software or platform.
Lots of these laptops have issues with getting stuck at 800MHz when on battery. At 800MHz, the laptop gets decent battery life. You are supposed to be able to force it to 1.8GHz all the time, but with a 80+ watt thermal design power spec, you won't have much battery life. Unfortunately, there have been reports that it doesn't work... 800MHz is all you get on battery.
The firewire port is a 4 pin, unpowered variety. No FW800. No DVI on the monitor output. No gigabit ethernet. No Bluetooth. Radeon 9600 vs. 9700 in the PB. No WPA support on the 802.11g. 1.6" thick, 7.5lbs weight, it's big and heavy. No Mac OS X. It also comes with Windows XP home... not Pro (another $180-200 for the upgrade).
You're not likely to run this machine in 64 bit mode anyways... there is almost no benefit in doing so.
It sure is cheap though. As a luggable, and the fact that eMachines doesn't actually build it, makes it a good value. But it isn't in the same class as an Apple PowerBook 15".
Obviously to you, if you don't perceive a need for it, then nobody needs it.
I don't know ANYONE who has a gigabit ethernet network.
You don't get out much, do you? I know a number of people that have gigabit networks in their homes (including mine), much less at work. There are quite a few inexpensive gigabit ethernet switches on the market, including the NetGear GS108 (8 port) for $150 and the GS105 (5 port) for $80.
With gigabit, I can do AFP or SMB transfers at over 50MB/sec which is a good 5x faster than a 100mb network to/from the network file server, and that is w/o jumbo packets.
Even the slow ATA drive in laptop can push 20-30MB/sec, so network performance would benefit from gigabit over fast ethernet. Try pushing around some video clips and you'll appreciate the speed difference.
I figured that Mail.app just uses OpenSSL for SSL support... and would therefore "pick up" TLS support. I'll have to play with my Mail server to see if I can force TLS only to find out.
An Opteron based system is not necessarily cheaper. Certainly not one that can handle dual procs and the cooling is actually engineered, not merely stuck into a generic case where they hope it will cool properly.
Go ahead and price the rack-mounts from Sun (v20z), IBM (eServer 325) or even Tier 2 vendors (Microway, Boxxtech, etc.) and you'll see that the Xserve G5 is very price competitive.
Add in the cost of Fibre Channel HBA's, and the price diff goes in favor of the Apple solution.
The Apple Store sells at full retail. You can do better from a reseller.
Secondly, the P4 is Intel's best price/performance offering - the Xeon's don't fare as well. Apple's dual offerings are much better price performance than the single 1.6. There are some offering where Apple isn't strong, and others where they are much stronger - of course.
For example - to build the main part of a DV workstation that connects to a shared Fibre Channel array:
Dual 2GHz PowerPC G5 4GB RAM 160GB SATA ATI Radeon 9800 Apple Cinema HD Display SuperDrive Fibre Channel Card Mac OS X $6,819.00
Just try to get an equivalent dual processor Xeon based Dell Precision Workstation under $7k without a FC card. The equivalent FC card (LSI Logic 7202) costs over $1700 list, $1400 at discount. Your Dell would have to come in at under $5600 to beat the price. The Dell 650 I configured hit 7K before I even added a monitor or the FC card.
Apple's Dual 1.8GHz box is also a really good value system: dual 1.8GHz 970's 1GB DDR400 160GB SATA ATI Radeon 9800 SuperDrive Mac OS X $2,970.00, retail
BoxxTech 3DBoxx M4.2: Dual 1.6GHz Opterons 1GB DDR333 NVidia Quadro4 380 120GB SATA (no 160GB offering) 8X DVD-R Windows XP or 2000 $2,818.00
The difference isn't much at all, and BoxxTech isn't a Tier 1 vendor. The price difference is a wash when you throw in the rest of the buying variables.
If you connect that Xserve via FC to a FC array (drives can be ATA, SATA, Parallel SCSI, FC, what have you) then you can easily SAN/LUN mask storage between different "heads." Throw in an APC Masterswitch VM for managing the power outlet, a serial console server from Logical Systems, Cyclades, or Lantronix, and you can have a server setup that can be manage completely off-site. Even do complete build-from-scratch.
Plus there's an entire manual devoted to CLI admin of Mac OS X Server - Server Documentation, look for the Command-Line Administration PDF file.
The Xserve is missing the LOM built into Sun Netras... and the OpenFirmware environment is blown away upon boot, but for most admins, Xserve's serial console access is good enough.
Xserve G5 pricing is competitive, and performance can be dramatically better on the 970 if your code can leverage Altivec/Velocity Engine. For example, a Sun V20z with dual 1.8GHz Opterons, 2GB RAM, and 36GB SCSI HD, and dual Gigabit is $4,445 retail (no OS license included). A dual 2.0GHz Xserve G5 with 2GB Apple RAM, 80GB SATA, dual Gigabit, and Mac OS X Server is $4,399.
First of all, buying an Apple PowerBook does not rule out Linux. Secondly, for someone that is lambasting Apple's small marketshare, it is interesting to me that you do seem to try to keep up with Apple news - even if your interpretation is oddly twisted.
3 x the clock speed of 1.5GHz is 4.5GHz. You really think that even the Pentium 4 will hit 4.5GHz before or at the same time a PowerBook hits 1.5GHz? You think that IBM would ship something like that in a ThinkPad? You should be knowledgable enough to know that clockspeed != performance, and performance != productivity. Are you buying productivity or are you buying clockspeed? Laptops/notebooks are often purchased for productivity (with pure performance as a secondary concern), and small things like wake from sleep in 2 seconds, easy switching across multiple networks and network configurations (including remembering many Wifi names and passwords), firewire target disk mode are conveniences that are definitely worth money. Further, battery life at full speed or equivalent speed tilts the value equation closer to the PowerBook. What model of Dell or IBM would you put up against a 15" or 17" PowerBook?
Microsoft actions are very different from Apple's actions - you haven't been paying attention at Microsoft's first and second anti-trust investigations/suits. The issue isn't being proprietary - that's an issue for the customer to weigh, and most customers these days don't even think twice about issues of proprietary vs. open standard or single source, or all that.
I've purchased commercial Linux (which is mighty expensive), I've purchased Microsoft, IBM, Sun, Digital, Apple, etc. products. The only way to slow Linux's maturity is to convince developers to not develop on Linux. Buying an Apple product does not do that any more than buying a x86 product that comes bundled with Microsoft Windows, even if your intention is to wipe it and put Linux on it. Just try buying a new laptop, or even better, a Tablet PC w/o Windows of any flavor.
Mac OS X ships with all remote access services turned off. You have to run something to open a hole, just like Mac OS = 9. There have been security vulnerabilities in server software for Mac OS = 9 - just like any other operating system.
The whole command line thing is a red herring. Mac OS 9 has no security model at all, so a buffer overflow style of attack can attack _any_ code in the system, including "kernel" routines.
The issue is the firmware on the PCI card. For the Mac, it has to has to work with Apple's flavor of OpenFirmware (which, btw, is less proprietary than common x86 BIOS'es). This is really only an issue with PCI devices that need to support booting. An network card, for example, doesn't need that kind of support.
Some cards have both - x86 and OpenFirmware. For example, most ATTO SCSI cards can work in either. Apple's Fibre Channel cards are LSI Logic FC HBAs and will work in x86's, Sun UltraSPARCs, etc. Most of the stuff you see at CompUSA only has x86 specific firmware - so for booting, they're not going to work.
If you were going to design a system that work with multiple processor architectures, OpenFirmware is about as open, cross platform, and diverse as it gets. It's actually the x86 side which is proprietary.
You can always run Bochs. It's free, and if you are messing around with the machine, it's probably better than real x86 hardware. When real x86 hardware crashes, you have to have messy debuggers or a two machine setup... when a virtual machine "crashes", you can probably step back through and see what happened.
Yeah, I spent all this money on OS/2 1.0, and Microsoft touted it as the future. It was so awful, and Microsoft released a really buggy 1.1 and then basically abandoned it. Microsoft has gotten its act together a bit since then, but I think you'd be better off with an Alpha running OpenVMS.
WTF??!?
"Apple has gotten its act together a bit since then"
Ah... today's Apple isn't the same Apple you know from before. After all, NeXT took over Apple in 1998. You are talking about ancient history.
Care to quantify your remarks here? Extremely poor I/O?
I find Mac OS X Server to be every bit as good as stable commercial Linux offerings on roughly equivalent hardware in real world situations. As far as the scheduler is concerned, I've seen far worse starvation issues with 2.4.x kernels than on Mac OS X. I/O throughput on Fibre Channel is also better (not to mention Mac OS X Server supports probing more then 1 LUN and sparse LUNs out of the box).
Now, there are some performance issues - like the time it takes to fork a process and other things that show up in synthetic benchmarks. But for real world performance? I haven't seen anything that demonstrates that RHEL3 or equivalent is substantially superior from a client's view across the network.
That may very well be true. I now recall FrameMaker running on earlier versions of NeXTstep...
In any case, Adobe could probably make a relatively small investment and have a Cocoa Mac OS X version of FrameMaker. But if FrameMaker isn't in the long term plan, then it probably doesn't make sense - which doesn't bode well for the product as a whole.
Actually, The Omni Group did the Framemaker port to NeXTSTEP. You can still see a reference to it on their jobs page under the "What's Omni Like?" heading. If Adobe wanted to put forth the money, The Omni Group could do the port.
Again, you have to prove that you can find an equivalent cheaper solution. The Apple Xserve G5 compares quite favorably against your common x86 Xeon or Opteron solution, especially if you are going to run Windows of some flavor. Plus, Mac OS X Server is far more approachable for for non-UNIX admins than most Linux distributions as long as what you want to do falls within the GUI. Actually, Mac OS X Server may act as a stepping stone to other UNIX flavors.:)
Have you priced IBM iron? Or Sun iron? Compared the features, performance, and reliability? For the SMB market, Apple's solutions are quite compelling especially if you are looking at centralized storage.
holdendeb didn't mention exactly which HP Smart Array controller his company was getting in those "cheap" DL380's. Considering that a 6404 is over $2k by itself, and the 6402 is over $1100, it's probably not the top of the line. Matter of fact, it is more than likely that it is the basic (discontinued) Smart Array 5i which goes for around $300. That's a dual channel Ultra 160 controller. But let's give him the benefit of the doubt that it is a Smart Array 642 with dual Ultra320 SCSI busses and PCI-X interface which sells for over $700.
While that SCSI controller might be able to do a sustained transfer for 320 megabytes per second on each bus, that's not the total throughput. Remember that with a 3 drive RAID-5 array, you have to write 150% the size of the original data - and the RAID controller has to calculate what to write - it is not as simple as just passing the data blocks onward. The bottleneck is usually the RAID controller - it can't keep up. Further, you then have the overhead of the SCSI bus arbitration - better hope he's using both channels and not all on one. Even if there is only one drive, the controller and the target still have to spend time on arbitration. His I/O's per second will probably be decent and his read performance will be good, but his write throughput will suck and the cost of the drives for the capacity will suck. I've used enough of these things to not buy into the SCSI RAID marketing literature anymore - ATA/SATA on the low end, pure FC on the high end + centralized storage where it makes sense.
Running Classic on a machine running Mac OS X Server as a server just isn't a good idea for several of the reasons you mention and more. We pretty much don't run Classic at all anywhere, but definitely not on a server.
I agree that the GUI coverage can be sparse and you'd have to drop down to the original configuration files sometimes. For some services, like apache, Apple actually helps you with where you can modify things and where you really shouldn't. Some of this is just difficult to do - getting good GUI coverage while preserving the original CLI configuration files and tools. Apple's GUI tools are getting better at this, but I agree this is something Apple has to continue to work on. For the 80-90% coverage, Apple's tools are probably sufficient. Otherwise, it's pretty modular and you can just lock out Apple's admin tools and do it the "old" way without tossing everything. Further, the Server Admin and related tools should have a plug-in architecture for 3rd party add-ons. That would be very cool.
Shared library updates still require reboots on Solaris and Linux. On Solaris, shutting down to single user mode and applying patches, as far as I am concerned, is the same as rebooting. Otherwise, you can't be sure that running programs pick up on the changes - and don't crash when they try to access a shared library.
Driver updates do not, as a whole, require reboots in 10.3. Mac OS X has had the capability to dynamically unload and load drivers and in 10.3, but most installation packages didn't bother (for a number of reasons). With 10.3, you can just send a HUP signal to kextd. For more information, see Technical Q&A QA1319, Installing an I/O Kit KEXT Without Rebooting
There are quite a few 3rd party installation packages that require reboots when one really doesn't need to do so - I suspect that they're just being lazy.
True. But the Xserve + Xserve RAID + Mac OS X Server is a compelling SMB solution which plays "nicer" with much bigger systems (which are typically UNIX) than your typical Windows x86 setup.
Okay... $3k is still more expensive than any iMac - so your statement above is still wrong. Further, $3k for what? Why would I want an ancient RAID 5 array using expensive disks that are bottlenecked by the RAID controller? How much do those StorageWorks FC cards cost at your "discount?"
You're the one that mentioned "proprietary." The question is, did you really understand what you meant by that? Where do you care what is "proprietary" and what isn't?
A current Compaq DL380 G3 with a single 2.8GHz Intel Xeon processor, 1GB RAM, and no hard drive sells for around $2450. That's more than the starting price of any iMac.
Besides, what's with this MAC term that you keep using? Don't you realize that the short name is Mac - short for Macintosh. It's not an acronym.
You've got serious issues if you want to play the "proprietary" card. Exactly how open is the BIOS in that Compaq? Plus, you obviously don't have a grasp on the price issues:
Apple Xserve G5: 2 x 2.0GHz IBM 970's, 1GB RAM, 80GB SATA, 2 x GigE, with Apple Fibre Channel card (LSI 7202, includes copper HSSDC2 to SFP FC cables), Mac OS X Server unlimited and the 3 year premium service and support costs $5,449
Compaq DL360, 2 x 3.06GHz Intel Xeons, 1GB RAM, 18GB 15krpm SCSI drive, 2 x GigE, universal sliding rail kit, Windows 2003 with only 10 CALs costs $6,438
The HP Storageworks FCA2214DC PCI-X HBA costs $2,500 bringing the total to $8,938
You might consider getting your head out of the sand
Actually, you're probably thinking of the 5.25" magneto optical drives that where the primary storage system in the base configuration of the original NeXT cubes. Cubes with magneto optical drives were publically introduced in 1988. I still have a few magneto optical cartridges around somewhere. They were still sold with NeXT cubes up to around 1991/92, upon which they then were then mostly sold with 2.88mb floppies and external CD-ROM drives.
I liked magneto optical drives because of their durability and relatively high storage density at the time, especially compared to Jazz and Zip drives which became all the rage. I used 3.5" magneto opticals for quite some time, but didn't use the 5.25" NeXT version after about 1993.
Slashdot Mirror
You contradict yourself over and over again and have demonstrated a lack of understanding on how to measure performance.
It seems, from you post, that your major concern is performance of a certain microwave antenna simulation program. Everything else should be secondary - the clockspeed, memory latency and throughput, the name on the outside of the box, etc. Now, you may have a price issue... for which then you want the best performance (which is most purchasers outside of governments). Whatever anyone says about performance is useless to you until you intimately understand the performance requirements of your particular application. The easiest way is to run it on various systems yourself as well as use various code analyzers. From the mistakes and misconceptions in your post, I doubt you really understand the performance issues with your application.
What is Intel's fastest CPU for floating point intensive operations? The Itanium2 is definitely their premier performance CPU for floating point operations. Since we don't have the inclination to examine your particular application, even if we had the source and the time to examine it, we'll have to use a rough approximation - SPEC.org's floating point CPU tests - SPECFP2000 results. Again, we're simplifying... everything that goes into a system affects performance, but for now:
HP Integrity Server rx4640 (1500 MHz, Itanium) - 2161
IBM eServer pSeries 655 (1700 MHz Power4) - 1776
AMD Opteron (TM) Model 248 (2.2GHz) - 1691
Intel 3.4 GHz, Pentium 4 Processor Extreme Edition - 1561
Again, the aggregate SPECFP2000 results probably don't correspond to the performance of your application - maybe a specific subtest does. Plus, there are any number of items to tune including the algorithm itself (if you have source), the compiler (from selection to flags), the operating system, hardware options, etc. But the fastest clock rate system (3.4GHz Pentium 4 Extreme Edition) above is the slowest in aggregate SPECFP2000, and the slowest clock clock rate CPU has the fastest result (1.5GHz Itanium2).
We haven't even gotten into whether or not your algorithm can leverage SIMD (as in the G4 or G5's Altivec) for major increases in performance, whether it is really memory bound, etc. Sometimes while you think your need major floating point, your algorithm is either not what you expect or is bounded by an integer operation that may be unnecessary - and therefore you are not leveraging the FP capabilities of certain chips. We haven't looked at whether or not the algorithm can take advantage of parallel systems, and if so, if it can use loosely coupled systems w/o losing too much performance. It goes on and on - and we haven't even tackled the issue of price. But in the end, this statement:
could not be more wrong.
I didn't want a Palm because I wanted a more functional mini-computer. I own a Zaurus SL-5500 and was set to buy a SL-6000 except for two problems. Price and the lack of Bluetooth. I would have still gotten one at that outrageous price ($700) if it had Bluetooth. Chances are - if I am in range of Wifi, I have my laptop. I need a PDA when I'm not near Wifi and want to access the net over GPRS cell phone - hence the need for Bluetooth.
Further, the relatively lackluster open source-ness of the project and Sharp's seemlingly lack of interest (and the new horrible contacts APIs) seem to seal its fate... I hate it, but only Microsoft would probably have the staying power in the long term.
Mistake right there. The question is, why did you do this? Why would you purposefully use a lossy codec for source material and then complain about video quality degradation?
BTW, iMovie takes DV format which is what you get when you grab video over firewire from a firewire equipped camcorder. This makes the workflow much easier for most people - hence that is the way iMovie works. Final Cut Express and Final Cut Pro support more import formats. iMovie expressly does not support taking in anything other than DV format - it's the way 99% of the audience will get their source material. Grabbing it and converting to MPEG-2 before working with the material is just plain idiotic. It's an artifact of the cheap video card you bought in that Dell.
Anyways.. if you insist on doing this, you could have used the free VideoLAN VLC to do the transcoding back to DV. The MPEG-2 playback component you bought enabled you to transcode also and you noticed that it degrades quality. But of course - that's why people don't do it that way.
If your source isn't something that has firewire output, then you can get an analog to firewire bridge or a PCI analog capture card. There exists USB versions, but the quality loss is significant. Seems like you paid for a video capture card in the Dell and didn't for the Mac.
You're doing something that most people won't do... and expressly not supported - it all comes down to iMovie only taking DV format (btw, which the Dell can produce). That's a choice that Apple made with their low cost editing suite and most people don't have a problem with that. The problem of transcoding is the same, regardless of software or platform.
Lots of these laptops have issues with getting stuck at 800MHz when on battery. At 800MHz, the laptop gets decent battery life. You are supposed to be able to force it to 1.8GHz all the time, but with a 80+ watt thermal design power spec, you won't have much battery life. Unfortunately, there have been reports that it doesn't work... 800MHz is all you get on battery.
The firewire port is a 4 pin, unpowered variety. No FW800.
No DVI on the monitor output.
No gigabit ethernet.
No Bluetooth.
Radeon 9600 vs. 9700 in the PB.
No WPA support on the 802.11g.
1.6" thick, 7.5lbs weight, it's big and heavy.
No Mac OS X. It also comes with Windows XP home... not Pro (another $180-200 for the upgrade).
You're not likely to run this machine in 64 bit mode anyways... there is almost no benefit in doing so.
It sure is cheap though. As a luggable, and the fact that eMachines doesn't actually build it, makes it a good value. But it isn't in the same class as an Apple PowerBook 15".
Obviously to you, if you don't perceive a need for it, then nobody needs it.
You don't get out much, do you? I know a number of people that have gigabit networks in their homes (including mine), much less at work. There are quite a few inexpensive gigabit ethernet switches on the market, including the NetGear GS108 (8 port) for $150 and the GS105 (5 port) for $80.
With gigabit, I can do AFP or SMB transfers at over 50MB/sec which is a good 5x faster than a 100mb network to/from the network file server, and that is w/o jumbo packets.
Even the slow ATA drive in laptop can push 20-30MB/sec, so network performance would benefit from gigabit over fast ethernet. Try pushing around some video clips and you'll appreciate the speed difference.
I figured that Mail.app just uses OpenSSL for SSL support... and would therefore "pick up" TLS support. I'll have to play with my Mail server to see if I can force TLS only to find out.
An Opteron based system is not necessarily cheaper. Certainly not one that can handle dual procs and the cooling is actually engineered, not merely stuck into a generic case where they hope it will cool properly.
Go ahead and price the rack-mounts from Sun (v20z), IBM (eServer 325) or even Tier 2 vendors (Microway, Boxxtech, etc.) and you'll see that the Xserve G5 is very price competitive.
Add in the cost of Fibre Channel HBA's, and the price diff goes in favor of the Apple solution.
The Apple Store sells at full retail. You can do better from a reseller.
Secondly, the P4 is Intel's best price/performance offering - the Xeon's don't fare as well. Apple's dual offerings are much better price performance than the single 1.6. There are some offering where Apple isn't strong, and others where they are much stronger - of course.
For example - to build the main part of a DV workstation that connects to a shared Fibre Channel array:
Dual 2GHz PowerPC G5
4GB RAM
160GB SATA
ATI Radeon 9800
Apple Cinema HD Display
SuperDrive
Fibre Channel Card
Mac OS X
$6,819.00
Just try to get an equivalent dual processor Xeon based Dell Precision Workstation under $7k without a FC card. The equivalent FC card (LSI Logic 7202) costs over $1700 list, $1400 at discount. Your Dell would have to come in at under $5600 to beat the price. The Dell 650 I configured hit 7K before I even added a monitor or the FC card.
Apple's Dual 1.8GHz box is also a really good value system:
dual 1.8GHz 970's
1GB DDR400
160GB SATA
ATI Radeon 9800
SuperDrive
Mac OS X
$2,970.00, retail
BoxxTech 3DBoxx M4.2:
Dual 1.6GHz Opterons
1GB DDR333
NVidia Quadro4 380
120GB SATA (no 160GB offering)
8X DVD-R
Windows XP or 2000
$2,818.00
The difference isn't much at all, and BoxxTech isn't a Tier 1 vendor. The price difference is a wash when you throw in the rest of the buying variables.
If you connect that Xserve via FC to a FC array (drives can be ATA, SATA, Parallel SCSI, FC, what have you) then you can easily SAN/LUN mask storage between different "heads." Throw in an APC Masterswitch VM for managing the power outlet, a serial console server from Logical Systems, Cyclades, or Lantronix, and you can have a server setup that can be manage completely off-site. Even do complete build-from-scratch.
Plus there's an entire manual devoted to CLI admin of Mac OS X Server - Server Documentation, look for the Command-Line Administration PDF file.
The Xserve is missing the LOM built into Sun Netras... and the OpenFirmware environment is blown away upon boot, but for most admins, Xserve's serial console access is good enough.
Xserve G5 pricing is competitive, and performance can be dramatically better on the 970 if your code can leverage Altivec/Velocity Engine. For example, a Sun V20z with dual 1.8GHz Opterons, 2GB RAM, and 36GB SCSI HD, and dual Gigabit is $4,445 retail (no OS license included). A dual 2.0GHz Xserve G5 with 2GB Apple RAM, 80GB SATA, dual Gigabit, and Mac OS X Server is $4,399.
First of all, buying an Apple PowerBook does not rule out Linux. Secondly, for someone that is lambasting Apple's small marketshare, it is interesting to me that you do seem to try to keep up with Apple news - even if your interpretation is oddly twisted.
3 x the clock speed of 1.5GHz is 4.5GHz. You really think that even the Pentium 4 will hit 4.5GHz before or at the same time a PowerBook hits 1.5GHz? You think that IBM would ship something like that in a ThinkPad? You should be knowledgable enough to know that clockspeed != performance, and performance != productivity. Are you buying productivity or are you buying clockspeed? Laptops/notebooks are often purchased for productivity (with pure performance as a secondary concern), and small things like wake from sleep in 2 seconds, easy switching across multiple networks and network configurations (including remembering many Wifi names and passwords), firewire target disk mode are conveniences that are definitely worth money. Further, battery life at full speed or equivalent speed tilts the value equation closer to the PowerBook. What model of Dell or IBM would you put up against a 15" or 17" PowerBook?
Microsoft actions are very different from Apple's actions - you haven't been paying attention at Microsoft's first and second anti-trust investigations/suits. The issue isn't being proprietary - that's an issue for the customer to weigh, and most customers these days don't even think twice about issues of proprietary vs. open standard or single source, or all that.
I've purchased commercial Linux (which is mighty expensive), I've purchased Microsoft, IBM, Sun, Digital, Apple, etc. products. The only way to slow Linux's maturity is to convince developers to not develop on Linux. Buying an Apple product does not do that any more than buying a x86 product that comes bundled with Microsoft Windows, even if your intention is to wipe it and put Linux on it. Just try buying a new laptop, or even better, a Tablet PC w/o Windows of any flavor.
Mac OS X ships with all remote access services turned off. You have to run something to open a hole, just like Mac OS = 9. There have been security vulnerabilities in server software for Mac OS = 9 - just like any other operating system.
The whole command line thing is a red herring. Mac OS 9 has no security model at all, so a buffer overflow style of attack can attack _any_ code in the system, including "kernel" routines.
The issue is the firmware on the PCI card. For the Mac, it has to has to work with Apple's flavor of OpenFirmware (which, btw, is less proprietary than common x86 BIOS'es). This is really only an issue with PCI devices that need to support booting. An network card, for example, doesn't need that kind of support.
Some cards have both - x86 and OpenFirmware. For example, most ATTO SCSI cards can work in either. Apple's Fibre Channel cards are LSI Logic FC HBAs and will work in x86's, Sun UltraSPARCs, etc. Most of the stuff you see at CompUSA only has x86 specific firmware - so for booting, they're not going to work.
If you were going to design a system that work with multiple processor architectures, OpenFirmware is about as open, cross platform, and diverse as it gets. It's actually the x86 side which is proprietary.
You can always run Bochs. It's free, and if you are messing around with the machine, it's probably better than real x86 hardware. When real x86 hardware crashes, you have to have messy debuggers or a two machine setup... when a virtual machine "crashes", you can probably step back through and see what happened.
Yeah, I spent all this money on OS/2 1.0, and Microsoft touted it as the future. It was so awful, and Microsoft released a really buggy 1.1 and then basically abandoned it. Microsoft has gotten its act together a bit since then, but I think you'd be better off with an Alpha running OpenVMS.
WTF??!?
"Apple has gotten its act together a bit since then"
Ah... today's Apple isn't the same Apple you know from before. After all, NeXT took over Apple in 1998. You are talking about ancient history.
Care to quantify your remarks here? Extremely poor I/O?
I find Mac OS X Server to be every bit as good as stable commercial Linux offerings on roughly equivalent hardware in real world situations. As far as the scheduler is concerned, I've seen far worse starvation issues with 2.4.x kernels than on Mac OS X. I/O throughput on Fibre Channel is also better (not to mention Mac OS X Server supports probing more then 1 LUN and sparse LUNs out of the box).
Now, there are some performance issues - like the time it takes to fork a process and other things that show up in synthetic benchmarks. But for real world performance? I haven't seen anything that demonstrates that RHEL3 or equivalent is substantially superior from a client's view across the network.
That may very well be true. I now recall FrameMaker running on earlier versions of NeXTstep...
In any case, Adobe could probably make a relatively small investment and have a Cocoa Mac OS X version of FrameMaker. But if FrameMaker isn't in the long term plan, then it probably doesn't make sense - which doesn't bode well for the product as a whole.
Actually, The Omni Group did the Framemaker port to NeXTSTEP. You can still see a reference to it on their jobs page under the "What's Omni Like?" heading. If Adobe wanted to put forth the money, The Omni Group could do the port.
Again, you have to prove that you can find an equivalent cheaper solution. The Apple Xserve G5 compares quite favorably against your common x86 Xeon or Opteron solution, especially if you are going to run Windows of some flavor. Plus, Mac OS X Server is far more approachable for for non-UNIX admins than most Linux distributions as long as what you want to do falls within the GUI. Actually, Mac OS X Server may act as a stepping stone to other UNIX flavors. :)
Have you priced IBM iron? Or Sun iron? Compared the features, performance, and reliability? For the SMB market, Apple's solutions are quite compelling especially if you are looking at centralized storage.
holdendeb didn't mention exactly which HP Smart Array controller his company was getting in those "cheap" DL380's. Considering that a 6404 is over $2k by itself, and the 6402 is over $1100, it's probably not the top of the line. Matter of fact, it is more than likely that it is the basic (discontinued) Smart Array 5i which goes for around $300. That's a dual channel Ultra 160 controller. But let's give him the benefit of the doubt that it is a Smart Array 642 with dual Ultra320 SCSI busses and PCI-X interface which sells for over $700.
While that SCSI controller might be able to do a sustained transfer for 320 megabytes per second on each bus, that's not the total throughput. Remember that with a 3 drive RAID-5 array, you have to write 150% the size of the original data - and the RAID controller has to calculate what to write - it is not as simple as just passing the data blocks onward. The bottleneck is usually the RAID controller - it can't keep up. Further, you then have the overhead of the SCSI bus arbitration - better hope he's using both channels and not all on one. Even if there is only one drive, the controller and the target still have to spend time on arbitration. His I/O's per second will probably be decent and his read performance will be good, but his write throughput will suck and the cost of the drives for the capacity will suck. I've used enough of these things to not buy into the SCSI RAID marketing literature anymore - ATA/SATA on the low end, pure FC on the high end + centralized storage where it makes sense.
Running Classic on a machine running Mac OS X Server as a server just isn't a good idea for several of the reasons you mention and more. We pretty much don't run Classic at all anywhere, but definitely not on a server.
I agree that the GUI coverage can be sparse and you'd have to drop down to the original configuration files sometimes. For some services, like apache, Apple actually helps you with where you can modify things and where you really shouldn't. Some of this is just difficult to do - getting good GUI coverage while preserving the original CLI configuration files and tools. Apple's GUI tools are getting better at this, but I agree this is something Apple has to continue to work on. For the 80-90% coverage, Apple's tools are probably sufficient. Otherwise, it's pretty modular and you can just lock out Apple's admin tools and do it the "old" way without tossing everything. Further, the Server Admin and related tools should have a plug-in architecture for 3rd party add-ons. That would be very cool.
Shared library updates still require reboots on Solaris and Linux. On Solaris, shutting down to single user mode and applying patches, as far as I am concerned, is the same as rebooting. Otherwise, you can't be sure that running programs pick up on the changes - and don't crash when they try to access a shared library.
Driver updates do not, as a whole, require reboots in 10.3. Mac OS X has had the capability to dynamically unload and load drivers and in 10.3, but most installation packages didn't bother (for a number of reasons). With 10.3, you can just send a HUP signal to kextd. For more information, see Technical Q&A QA1319, Installing an I/O Kit KEXT Without Rebooting
There are quite a few 3rd party installation packages that require reboots when one really doesn't need to do so - I suspect that they're just being lazy.
True. But the Xserve + Xserve RAID + Mac OS X Server is a compelling SMB solution which plays "nicer" with much bigger systems (which are typically UNIX) than your typical Windows x86 setup.
Okay... $3k is still more expensive than any iMac - so your statement above is still wrong. Further, $3k for what? Why would I want an ancient RAID 5 array using expensive disks that are bottlenecked by the RAID controller? How much do those StorageWorks FC cards cost at your "discount?"
You're the one that mentioned "proprietary." The question is, did you really understand what you meant by that? Where do you care what is "proprietary" and what isn't?
Apple Xserve G5: 2 x 2.0GHz IBM 970's, 1GB RAM, 80GB SATA, 2 x GigE, with Apple Fibre Channel card (LSI 7202, includes copper HSSDC2 to SFP FC cables), Mac OS X Server unlimited and the 3 year premium service and support costs $5,449
Compaq DL360, 2 x 3.06GHz Intel Xeons, 1GB RAM, 18GB 15krpm SCSI drive, 2 x GigE, universal sliding rail kit, Windows 2003 with only 10 CALs costs $6,438
The HP Storageworks FCA2214DC PCI-X HBA costs $2,500 bringing the total to $8,938
You might consider getting your head out of the sand
Actually, you're probably thinking of the 5.25" magneto optical drives that where the primary storage system in the base configuration of the original NeXT cubes. Cubes with magneto optical drives were publically introduced in 1988. I still have a few magneto optical cartridges around somewhere. They were still sold with NeXT cubes up to around 1991/92, upon which they then were then mostly sold with 2.88mb floppies and external CD-ROM drives.
I liked magneto optical drives because of their durability and relatively high storage density at the time, especially compared to Jazz and Zip drives which became all the rage. I used 3.5" magneto opticals for quite some time, but didn't use the 5.25" NeXT version after about 1993.