Apple Introduces Xserve Rackmount Servers
2nd Post! writes "MacCentral is reporting the announcement of 1U Apple rackmount hardware. The Xserve, despite its cheesy name, seems quite powerful: dual G4/1GHz with 4MB DDR L3 cache, up to 2GB DDR (yes!) SDRAM, 4 ATA drive bays (up to 480GB), 2 Gb Ethernet ports, 2 64/66 PCI slots (one of which may be taken up by one Gb Ethernet card), and, of course, FireWire. Pricing starts at $2,999 for a single 60GB disk and 256MB RAM." Yahoo! has posted the press release; Doc Searls is writing about Jobs' speech. Update: 05/14 18:14 GMT by M : Apple's page about the Xserve is now live.
They've got a 3U, dual fiber channel, 14 drive RAID Xserver in the works. Keep a lookout for those ;)
This is where I get my recommended daily allowance of "Foot in Mouth."
The original specs are wrong, it's $2999 for the SINGLE 1 GHz G4, $3999 for the dual. Not as sweet a deal, but still not too bad.
Unless I misunderstood, the XServe has two 64bit PCI slots, and only one is used (by an ethernet card). The other ethernet port is onboard. This leaves one slot free, or two if you don't need to ethernet ports.
It wouldn't make sense for them not to...Remote Desktop is a perfect way to deal with any must-be-local issues. I assume that all server management programs can be run remotely, since they ran a server manager that identified all locally-running Xserves.
Probably something similar to their old Mac Manager Server.
And telnet's disabled by default, you have to ssh in
Hire a Linux system administrator, systems engineer,
the $1000 price increase also includes a 2nd 1gHz G4. they failed to mention that the first one is single proc.
A computer without a Microsoft operating system is like a dog without bricks tied to its head
you can access the machine using the Mac OSX gui, remotely. I think this is meant to be headless, but using the admin tools you can bring up a desktop view of it on another machine if you wish.
Only the HFS+ filesystem does.
1000$ for an extra 256megs of Apple blessed DDR
Apple's RAM is always overpriced, just like most OEMs. So you buy extra RAM 3rd party, as usual.
IDE just as fast as SCSI my ass
True, but Ultra3 is an obvious expansion option.
No expansion slots. The second gigabit network card takes up the only PCI slot
I'm not sure where you got that idea. The press release says: "three PCI slots, two of which are 64-bit, 66 MHz". I have no clue how they fit 3 PCIs and 4 bays into a 1U box, but I sure am glad.
- If you want a pure Unix experience at the command line, install OS X on UFS. Trivial. Works. Breaks some third party apps that are Carbon based, but you'll likely not care (I don't).
- porting: Most packages compile out of the bag or with very little in the way of patching (a lot only require a couple of command line arguments. Fink.sourceforge.net currently has 1100 packages 'ported' to OS X, all fully managed by the debian package manager.
Fink has certainly grown in size since your purchase, but not much else has changed.
As James Gosling recently said: "OS X is like Linux, only with Q/A [Quality Assurance] and taste!".
It looks like the release has been changing a bit. The first glance I had of it was when it was first posted (prior to it showing up on Slashdot). At that time it simply stated that a second gigabit card took up the PCI slot with no mention of any other expansion slots. Later it listed two 64bit PCI slots so I assumed that the network card was in one. Now it lists three slots with one being dedicated to the second NIC. Ah well, having two free slots is much better.
"Have you ever thought about just turning off the TV, sitting down with your kids, and hitting them?"
Well the dell *looked* good, but lets see:
Windows 2000 Advanced Server with 25 Client Licenses [add $3295]
VersaRails for Non-Dell 4-Post Rack [add $129]
Dell Remote Assistant Card Version 3 without Modem [add $499]
73GB 10K RPM Ultra 160 SCSI Hard Drive [add $550]
Intel Pro 1000XT Gigabit NIC-Copper [add $189]
Total cost - $6,459.00
But maybe you wanted Linux - $3,323.00
I won't really get into the who SCSI/IDE debate, suffice to say Apple announced a Fibre Raid with 400MB through put, it you really want it. Shipping in Q4 with 1.48 TB of space in a 3U, all hot swappable. The Apple prices are spot on for all the features they bring. IMHO of course.
People think Microsoft is the answer. Microsoft is just the question, "No" is the answer.
The DELL has dual gigabit on board, you dont need the Intel Pro 1000XT card unless you want more then two ports. In which case you would have to add one to the Apple as well.
As for the OS, whats wrong with Linux? Given the market that Apple is targeting it would seem like a more logical choice.
"73GB 10K RPM Ultra 160 SCSI Hard Drive [add $550] " Ok, so how much to add this to the Apple? You seem to want to configure the DELL with a lot of stuff that the Apple dosn't have in order to drive the price up. This is not how you do a fair comparison, but it does seem to be how Apple does things. Which is a pity as I've allways felt that their hardware stands up rather well on its own without resorting to that kind of BS.
"Have you ever thought about just turning off the TV, sitting down with your kids, and hitting them?"
shipping by the end of the year, no price yet...
3U height
*14 drive bays
*14 120GB ATA drives - in same hot-plug format as Xserve
*1.68TB
*Dual 2GB Fibre Channel on system
*400MB/second storage throughput
full info posted at apple.com any time now
OSX supports software RAID, even at the consumer level. Put in 2 or more disks, and you can stripe/mirror all you want. The new servers have 4 independent IDE channels...it's a safe bet that you'll be able to set up a RAID. Maybe not RAID 5, but that's what you buy the forthcoming fiber channel storage device for. In any case, how is built-in RAID a rip-off?
its in the management graphic. i want that too
I want 2D games back.
The ATA drive subsystem has a high-bandwidth I/O bus that minimizes bottlenecks, even when all four drives are engaged at once. That's how Xserve can achieve a theoretical peak performance of up to 266 megabytes per second, compared to a 160MB/s theoretical performance with SCSI Ultra160 disk drives -- at a significantly lower cost, and while generating less heat than SCSI drives.
Read the EFF's Fair Use FAQ
Increasingly Audio/Video production is becoming de-centrelized to the point where editors and producers need to be able to work from a common source that addresses "thier" needs. Not the needs of gamers or SOHO admins.
Since the production work is mostly done on Macs it makes perfect sense to use a Mac server.
Cost of hardware has always been secondary to quality of workflow and consistency in delivering the end product. (meaning: the shit should just work! and it should work the way you'd expect)
Face it, we pay THOUSANDS for audio cards and video equipment. We are not home "tinkerers" and dont want to tinker with our servers.
If these Xservers can also double as workstations 2 birds go down with one stone.
Windows admins and Linux hobbiests will never get it.
Go Apple !
"Corporate rock still sucks. What are you gonna do about it?"
I still think you are crazy. All the sources I have seen say that linux has preemptive multitasking. Perhaps you are talking about having a preemptive kernel (or something like that, I'm a bit hazy on the details), and there is a rather experimental patch for that. But Linux has supported preemptive multitasking for a long time. This differs from the systems used by old MacOSes and Win3.x, but pretty much all modern OSes support preemptive multitasking. Please don't spread misinformation.
They also forgot mention Oracle's high RAM requirements coupled with XServe's low-RAM expandability!
A database with a 2 GB SGA, running on four internal disks? Boy, I hope you weren't expecting to run anything FAST!
2 GB wouldn't even hold my undo segments in memory! heh!
Not as fast connection
Jeremy D. Zawodny /
As a ballpark figure, 1 watt turned on all year costs you $1. Maybe double that if you are in a continuously air conditioned environment like a machine room.
The savings may not be too large. I checked an Athlon system with an ammeter recently. It came in at 120W with one drive in it while doing its server tasks. So, they at least are in the same ballpark. (The measurement techniques are surely different, I would not claim one was higher than the other based on this data. Just that they are near each other.)
Power is one of the reasons I suggest people not use that crappy old 486 or pentium as a NAT/firewall box in their house unless they are doing it for joy. In about a year or so of electricity savings you can pay for one of the new integrated appliances and enjoy increased reliablity and savings in the following years.
*sigh* You are talking about kernel preemption, or preempting the system task, not preemptive multitasking.
In many places in both the Linux (2.4/2.5) and Darwin kernel's (depending on the device drivers), both will fail to preempt themselves for a userland task. (Yes, Virginia, there are chunks of code even Darwin won't preempt) Likewise, in many places (even in extremely old versions of the linux kernel), preemption can happen. You would be correct to say that there is a focus in 2.5 for trying to eliminate or optimize a lot of the non-preemptable code and to say that Darwin experiences marginally lower average latency than Linux 2.4, but to use that as some way to measure system performance is as ridiculous as it is stupid.
Besides, if you want to get super technical, there are two robust and stable implementations of Posix realtime threads for linux (RTAI and RTLinux) that have existed for a number of years. Darwin has no such beast. Now we are talking latencies of 10-15 microseconds vs the low-millisecond ranges of either Darwin or Linux 2.4/2.5
And if you want to get even further into the technical mumbo-jumbo, the ARM processor can rock both the PPC and the X86 in terms of preemption. There are event's called FIQ's (Fast IRQ) on ARM that cause the processor itself to preempt ITSELF and execute some other code! You can call efficient FIQ code on the order of 10MHz and still run your normal stuff on top of the CPU -- and on Linux too -- on top of RTAI Posix RT threads -- or not!
Oh, and Intel makes the best ARM cores, too. Yeah and they have 32 bit registers just like your 64/128bit PPC's.
You don't know what you are talking about.
First, Apple has NEVER produced a machine with a 64-bit CPU. Period.
Second, Linux has supported preemtive-multitasking from the start - something like 10 years ago...
You are confusing it with KERNEL PREEMTION which means the *kernel* process itself can be preemted to support real-time scheduling. Darwin has never supported this, and I've not heard of any rumours that it will in the future either...
As for features, there are plenty of server throughput benchmarks that show Linux to be as fast or faster than Windows. I've never seen any OS X server benchmark that even comes close.
Hint: Photoshop filters are not a server benchmark, and results "up to" X times faster with no details just means it much slower in many other cases...
"Each drive has an independent Ultra ATA/100 bus, an arrangement that allows maximum individual drive performance without choking the throughput of the other drives. SCSI is better than IDE when controlling multiple disks. Apple must have done that math and figured 4 IDE controllers and with IDE drives had a better price/performance than a SCSI-based system."
I also don't think that IDE support hot swapping as well as SCSI. However, It looks like having a controller per disk also allows Apple to get around this (the new servers do support hot-swap drives). They probably just shut down the entire IDE controller for the drive to allow hot swapping.
All in all, I think these new servers look very cool.
The drives are all masters. Xserve has 4 ATA/100 channels, 1 per bay.
Apple's L3 is very fast. It's 500MHz DDR SDRAM (Effictive speed of 1GHz), and there's 2Mb per processor, much better than only 512Kb close coupled Cache.
And your Athlons, nice warantee on them, I stick to systems that come with Service Contracts before I drop $$ on a server for colo or any serious work. Dell can play that game, so can HP, Sun and now Apple, but your Athlon's can't.
Clone boxes are great, until you play in the big leagues, and that's where Apple is palying here.
The Crazy Finn
"You've got an invalid haircut" -Warren Zevon - Life'll Kill Ya
The raw electricity is only part of the equation.
175Wh/hr * 1kWh/1000Wh * 24hr/day * 365days/yr * $0.08/kWh = $122.64/year.
However if you air condition that room a really effecient air conditioner at 25% would cost $490.56/year to run.
Add infastructre cost for a larger air conditioner, extra electrical wiring and equipment, larger UPS, and your system just got very expensive. I don't have any solid numbers here but guessing $1000 per 1U doesn't seem unreasonable.
While we are on cost savings these things pack more into a 1U than I've ever seen. 2 processors and 4 hot swappable drives is usually a 2U feature. So figure you are saving half the floorspace. What does good data center floor space cost? Again I don't have any numbers, but it's certainly expensive. This might even be the overriding cost in a place like San Francisco.
What I'm getting to is that these things really can make economic sense. Let me keep going.
Easier administration than Linux. I administer linux machines, I know it's time consuming. SysAdmins don't come cheap. 1 Admin = $60,000 year.
The G4 Vector processor kicks butt for scientific workloads, really kicks butt. 300% performance improvement is not abnormal. The main reason physics guys don't use macs now is that they don't fit in a rack very well. Cut the number of servers down by 1/3, that's money in the bank.
Finally, these are just too cool whatever the cost. I mean that case looks sweet.
Anyone who cannot cope with mathematics is not fully human.
Actually the ATI vid card is in a 64bit slot, the GigE card is in the AGP/PCI combo slot.
If you spec the Radeon 8500 you lose the second GigE. BTO options for PCI cards are only a U160SCI Card or 1000baseFX(Fibre GigE) card in the top 64bit slot, the bottom 64bit slot is base video or empty and the 32bit slot is 1000baseTX copper GigE or Radeon 8500. Of course you can mix&match somewhat (The base video is a 64Bit card, so you can't put it in the 32bit slot)
The Crazy Finn
"You've got an invalid haircut" -Warren Zevon - Life'll Kill Ya
Wow, umm, where to start? You have all of your information absolutely wrong (as does the post below yours), so I'll just explain what the 280R is and is not:
The 280R is a single- or dual-proc Ultrasparc-III, supports up to 8GB of RAM, and supports up to two FC-AL (yes, fibrechannel, not SCSI) drives internally, along with one external FC-AL connector and I think four PCI slots. It's 4U, not 5. It also has a remote management card which provides LOM-like features (poweron, poweroff, etc.).
And I think it starts at about $12k, and if you want the dual-proc, it's more like $20k. I don't think Apple ever said this would beat a 280r in all categories, but I would say (as someone who has been building and maintaining Sun boxes for years) that this box compares quite favorably with actually competitive offerings: Windows on Intel.
It does, of course, still lose in most areas against the 280R, but only if you are a company who would benefit from the Sun box. If you are a school, or a small creative shop, or even a big creative shop, or any shop which already has lots of OS X and no Solaris, this is the box for you.
That I can see, I guess...
But you returned it (eating the 10% fee) before taking the 30 seconds it would have taken you to find out that the traditional *nix filesystem was an available option?
That's just stupid.
For the record, if you don't like HFS+, you can use UFS. Also if you don't like tcsh, you can install bash (free download from Stallman & co.). If you took a deep breath, calmed down, and did a quick visit to any of thousands of web sites that were chattering about this stuff at the time, you could have found all this out. For that matter, if you had bothered to look into it before buying a $1200 laptop, you would have known all this going in.
Information wants to be anthropomorphized.
Have I installed secure Apache? No, it comes installed. have I set one up? Yes and takes 1 minute to configure a website and 3 minutes to configure a secure one. Right out of the box - that's why it's easier.
Have I installed Tomcat? No, it comes installed - well and older version does. Played with it - not interested. Will install newer version if/when I want to.
Postgres? No, but I played with pre-installed MySQL - 2 minutes to turn on and use. Upgraded to later version (to fix BLOB>255 bug) and continued to run it. That did take more than 5 minutes. Maybe 15? You can downlaod source and install or get binary images and install them if you want the latest version.
I can get an application server up and running for much less than a WebObjects/Oracle solution.
Yes you can. I NEVER said you couldn't. Not everyone is you. Some people want that solution and here's a product that supplies it.
On Mac but preferably on Linux/Intel for hardware cost reasons - hell I provide SCSI RAID
Again, not everyone is you. This is exactly my point: Apple offers a product. People seem offended by it's very existence. When was the last time you heard someone say, "Why does BMW even make or sell cars?" Because people buy them.
I can do cheaper/differently/blah blah blah And now that Apple offers this product you still can. Here's an idea: If you don't want it - don't buy it.
=tkk
PS XServe will do RAID - software RAID as is or add SCSI/RAID with a PCI card. From the Apple BTO store. Go check it out.
Bill Gates - Creationist?!?
http://www.mackido.com/Hardware/G4.html http://www.apple.com/g4/
just my blog and pix
It gets a little more interesting when you take that theoretical peak performance of 630 GFlops for a rack of these babies and look at the most recent Top500 [top500.org] list.
The TOP500 list is sorted based on the performance of double precision LINPACK, so the theoretical peak performance for a rack will be 78 GFlops and not 630 GFlops. (assuming the PowerPC can throughput 1 double precision flop / clock now - it might actually only be 1 / 2 clock, in which case the performance would be 39 GFlop).
If you want to play with the TOP500 gang you'll have to follow their rules!
Apple's tech-specs page for the Xserve lists, under I/O:
Two full-length 64-bit, 66MHz PCI slots
(lower slot filled with PCI graphics card in standard configurations);
supports 3.3V 32-bit or 64-bit PCI cards running at 33MHz or 66MHz
One half-length 32-bit PCI/AGP combo slot with one of the following:
-- Secondary Gigabit Ethernet card in standard configurations
-- AGP 4X graphics card (build-to-order option)
Sounds like a real AGP slot to me. I can't see them hobbling the Xserve with some sort of complicated fake-AGP-over-PCI, when they already design all their other motherboards with video on AGP. Why go to the extra trouble?
I don't know who wouldn't be happy with these remote management/monitoring options:
Server Monitor for remote monitoring of key hardware subsystems: enclosure temperature, processor temperature, blower speed, hard drives (SMART data), Ethernet links, power supply and UPS systems, enclosure security
Server Admin (TCP/IP)
Remote Setup Assistant
Simple Network Management Protocol (SNMP)
Intermapper from Dartware
Secure Shell (SSH2) for secure remote login
Command-line tools for remote configuration and management, including installing software, running Software Update, and setting system and network preferences
Go out and get sailing!
Just recently there was a study saying that ATA was better than SCSI in almost every situation these days ... it's very hard to overcome the huge cheap ATA drives. In this case, Apple is not really just using ATA ... each ATA drive has its own controller all to itself ... so instead of two SCSI drives on one controller giving you 140GB total, you have four ATA drives on four ATA controllers, giving you 480GB total, and the ATA stuff cost a lot less, too. Then you put in 10 1u servers each with four ATA drives and all hooked up with Gigabit Ethernet and you're probably getting some pretty good performance there.
... they have flexibility to make that work where other companies might be using someone else's kernel, or running Windows and just taking whatever they get there.
Also, Apple can make optimizations to Mac OS X Server in order to gain more performance out of Xserve and ATA drives
1GHz CPU, 4 60GB disks (as 2 RAID 1 drives), AppleCare for $3,889.00 (.edu price) Whoo-hoo!
Dontcha love how Apple just trickles this stuff out but when you put it all together, it's just unbelievable?
THIS is what the mac platform is all about. "Apple makes the whole widget," indeed. What Apple has done with Mac OS X and its other apps is build a software environment that is streamlined and effective because it is running on a guaranteed hardware configuration that is known to the developers. It's almost like coding for an appliance... you KNOW exactly what hardware this system will run on, you KNOW what that hardware is capable of, and you know exactly how to interface to it for maximum potential. This leads to a system that's reliable and almost immune to hardware problems or wildly divergent performance.
Apple does not need to maintain different kernels because some motherboards run interrupts in different ways. Apple does not need to have nearly 20 years of backwards hardware compatibility. Apple does not need to stake their reputation on bargain basement Taiwanese programmers writing drivers for the hardware their code will run on. Most importantly, Apple does not need to build an illegal monopoly to build an computing environment that works.
Try being productive under Windows running in Safe Mode. Now try using Mac OS with extensions disabled. Tell me Windows, at its bare core, is better than Mac OS. I dare you.
-Lx?
As a ballpark figure, 1 watt turned on all year costs you $1. Maybe double that if you are in a continuously air conditioned environment like a machine room.
The problem isn't with the power per machine per year, but the power density (watts per volume). Above a certain number of kW per rack (5? 6?) the cost of cooling increases non-linearly because of the need for specialized AC equipment, plumbing, building reinforcement for heavy stuff on the roof, etc.