There are plenty of "designs" that require fast disk. For example, video production and editing, some of which require sustained data rates in excess of 190 MB/sec.
I can't speak for 3ware RAID... but nothing can be simpler than replacing a failed drive in an Xserve RAID. Pop out old drive, pop in new drive... 2 seconds total.
Though, Xserve RAID as a lot of features that you probably don't need, and hence don't want to pay for, if you're just looking for 1 TB of nearline backup storage. It's an "enterprise" product. And in its class it's about 1/4 the price of anything comparable (say, EMC Clariion CX 300).
What's this "15 devices per channel" crud? I'd certainly expect that a properly designed SATA/ATA RAID enclosure would have but one drive per channel, and have 14 or 16 or whatever channels per rack-mount RAID box.
At least I'd hope that's how it's being done, otherwise performance would suck (in SATA it obviously *must* be a drive per channel, but with PATA any realistic design would do the same).
That's how Xserve RAID can get sustained reads of 340 MB/sec off 14 ATA drives.
Or, buy it off the shelf if you want reliability and performance.
Not to knock you, but software RAID 5 performance sucks ass. If 30 MB/sec is sufficient, go with it. If you want better performance (Xserve RAID can do about 340 MB/sec sustained in RAID 5), other solutions may be better.
As for the drives... sure, go with the Fry's drives. They're not QA'd but that's your perogative.
Xserve RAID may well be overkill for this... I think Promise makes a controller/enclosure that costs about $2K and can take standard IDE/SATA drives in its sleds... that will end up being cheaper than the Xserve RAID but more than a frankenbox running Linux software RAID.
Granted, it is more difficult to program something (from the ground up) that runs distributed, than it is to program something that runs on a giant 2048-way box.
Just like it's more difficult to write multithreaded code than it is to write single-threaded code.
That's where software, and platforms come in. There is a TON of research being done, which uses technologies like Infiniband and Myrinet as interconnects, and can make a cluster "look" like a big monolithic machine. If you as an end user write code that goes down into the TCP stack itself, you're working too hard, and you're going about it the wrong way.
Put it this way: In 5 years the odds are overwhelming that there will be a good software platform that can let you pick 5000 servers and run your app 10,000 threaded, with everything appearing just like a single process, and running "as it would on a Cray." It's easier to solve this stuff with software -- take your problem (distributed computing) and solve the problem with a different set of technologies (high performance/low latency interconnects, shared address space/DMA across machines, etc).
Apple's Xgrid is a step in this direction. It's missing a ton of "Supercomputer" functionality right now, but it's a nice cross-machine GUI scheduler. Right now this type of app can address maybe 20% of what supercomputer apps need... in the future maybe more like 98%.
I had an SN41G2, the noise issue was 100% due to the video card fan.
The issue is, the video card fan points right at the holes in the side of the case, and is about 1/4" from them. So you get loud forced air out the small vent holes, which gets very noisy. I drilled the vent holes out a little larger, but it was still louder than I wanted.
Mine was actually quieter with the case open, because this blown noise didn't exist. But then you got more drive noise.
I ended up selling it and getting a Coolermaster. Though the SFF was very cool.
GF just bought a car... Ford Escape. $22,000 off the lot. $5K, 5 year financing at 2.9% (could get 3 year at 0.9% but hell, 2.9% for 5 years is pretty damn attractive).
About $350 a month. So no $300/month isn't unreasonable.
Have you evern run a "modern game" on a multi CPU machine?
Trust me, I've done it, and I know that Dual CPUs aren't "all that" for gaming. I got a Dell Precision 420 (2x733 P3) for gaming (gasp, $4K, what was I thinking? Whoever says PCs are cheaper than Macs:-P) back when the fastest CPU available was an 800 MHz. The frame rates I got on the dual 733 with my vid card were about 5% less than benchmark sites got on a *single* 800. So... the multiple threads don't pan out in practice.
And I don't think it's gotten any better... games just really aren't optimized for throughput. And "offloading to the GPU in a second thread"... nope.
My strong, experienced advice, says if you want the best gaming experience, go with a single fast CPU, and a fast video card. Dual CPUs are for servers or for desktop tasks when you want enhanced responsiveness. They're not for gaming.
I turned on SMP support with Quake 3 Arena and it crashed all the time. It was unusable. So I turned it off.
Practically, I don't think SMP support is all that "baked" in games. My experience has shown a single, fast cpu is much better than multiple, "less fast" cpus. Significantly.
Though, UT2004 runs quite well on a Mac. As do Call of Duty and BF 1942. Halo is a bit slow.
This at 1920x1200 resolution (23" Cinema Display)... played at WWDC on a 2x2.0 G5 with a Radeon 9800. Frame rates at that res were pretty consistent 70 FPS, never dropping below 40. So it's not ALL bad.
Though BUYING a Mac specifically to be a gaming machine, I might not advise that;-)
An OS X Server (Xserve), is appropriate most anyplace a Linux server is appropriate. The two are very similar -- they run the same open source compoenents, they're both UNIX-y, etc.
So they're both excellent replacements for Windows servers when you're tired of playing CALs. A Mac server is easier for a "novice" to set up and maintain, as the GUI configuration support is superior. It's also a lot easier to integrate Macs with Windows environments than it is with Linux -- OpenDirectory is very easy to integrate with Active Directory via kerberos (3 clicks and you're done, and you can have centralized directory services cross-platform)... with Linux you can install OpenLDAP but good luck getting it integrated with Kerberos in less than a month.
Cost is fairly similar -- out the door the Mac is slightly more, but if you use "Enterprise" Linux like Red Hat Enterprise Linux, then the costs are fairly similar (as the Xserve includes the software license).
So a better question is "where isn't it appropriate?" And the answer to this is usually "when the company is locked into an Exchange server and doesn't want to move off."
How about because anything over 256 kbps is INDISTINGUISHABLE from the original content? People (even the best audiophiles), cannot tell.
So why you'd want to take up 5x the space (or more) for lossless compression, I can't say.
It makes the difference between being able to put 5000 songs or 1000 on a 40 GB iPod. For many people, 1000 songs is not enough. And you're doing this for something you can't even notice a difference on, in fact only your dog can. And he probably doesn't like your music anyway.
Oracle is keenly interested in the TCO of its customers.
Specifically, because given a fixed TCO, if the OS is "less expensive" there is more margin for Oracle in the sale.
For example, 5 years ago, a customer would buy Oracle to run a "big database" on an E10K with Solaris. They may pay $1M for a 3 year contract for the Oracle license, and $1M for the hardware and software maintenance from Sun (these numbers are rough, order of magnitude figures).
These days, the customers can run the Oracle database on cheap, commodity Linux. And use Oracles "RAC" functionality to load balance the DB across the boxes. So the hardware "piece of the pie" changes to $5K (decent dual CPU Opteron with 8 MB of RAM), times 15, so $75K for hardware. They pay Oracle $1.2M for software, RAC support, and Oracle supports their RHEL 3.0 installation (and kicks back some money to Red Hat).
So NOW, Oracle's piece of the pie is better. AND EVEN MORE, the customer is now even more tied in to Oracle's DBMS, because they're using the 10g RAC functionality for load balancing, fault tolerance, failover, scalability, yadda yadda. What happens in 3 years? Yep, license costs go up.
But again, overall TCO has improved in this case... DBMS costs have gone from $2M every 3 years, to $1.5M... a 25% savings. And the sales rep and the buyer can go golf together at Pebble, because they both saved/made more money for their companies.
And Sun can lay off more people, because their bigwig fat cat sales managers can't realize that the last E10K was sold about 18 months ago, and their margins are going straight to/dev/null.
Ummm... so last I checked 100 mbit was what 12 MB/sec? Or 10 MB/sec give or take? So even your 5 year old drive can sustain reads at > 10 MB/sec. So 3-4 times would be maybe 30-40 MB/sec.
Even assuming the drive cannot saturate a gigabit link (even at 50% utilization), which is a point I agree with, the drive certainly CAN beat 100 Mbit. So, again, my point is, given this, why not spend the extra $2 for a GigE adapter?
I mean, how different is this to using a USB interface on a portable HD instead of a USB2 device? In this day and age, that would be stupid now wouldn't it?
I'm fully aware of the differences in FC performance, and still this doesn't affect anything. I have an external, network attached storage device. I have full Gigabit (switches, NICs in my machines -- again, at a CONSUMER price point). So attaching storage via 100 Mbit? I'll look at perf #'s between this device and its competition with GigE, and choose the faster one. Methinks it isn't the 100 Mbit model.
But login window is kerberized. Kerberos is the way authentication is being done, so you'd want to kerberize your services. Another pluggable authentication layer would be superfluous.
Slashdot Mirror
There are plenty of "designs" that require fast disk. For example, video production and editing, some of which require sustained data rates in excess of 190 MB/sec.
I can't speak for 3ware RAID... but nothing can be simpler than replacing a failed drive in an Xserve RAID. Pop out old drive, pop in new drive... 2 seconds total.
Though, Xserve RAID as a lot of features that you probably don't need, and hence don't want to pay for, if you're just looking for 1 TB of nearline backup storage. It's an "enterprise" product. And in its class it's about 1/4 the price of anything comparable (say, EMC Clariion CX 300).
What's this "15 devices per channel" crud? I'd certainly expect that a properly designed SATA/ATA RAID enclosure would have but one drive per channel, and have 14 or 16 or whatever channels per rack-mount RAID box.
At least I'd hope that's how it's being done, otherwise performance would suck (in SATA it obviously *must* be a drive per channel, but with PATA any realistic design would do the same).
That's how Xserve RAID can get sustained reads of 340 MB/sec off 14 ATA drives.
Or, buy it off the shelf if you want reliability and performance.
Not to knock you, but software RAID 5 performance sucks ass. If 30 MB/sec is sufficient, go with it. If you want better performance (Xserve RAID can do about 340 MB/sec sustained in RAID 5), other solutions may be better.
As for the drives... sure, go with the Fry's drives. They're not QA'd but that's your perogative.
Xserve RAID may well be overkill for this... I think Promise makes a controller/enclosure that costs about $2K and can take standard IDE/SATA drives in its sleds... that will end up being cheaper than the Xserve RAID but more than a frankenbox running Linux software RAID.
Granted, it is more difficult to program something (from the ground up) that runs distributed, than it is to program something that runs on a giant 2048-way box.
Just like it's more difficult to write multithreaded code than it is to write single-threaded code.
That's where software, and platforms come in. There is a TON of research being done, which uses technologies like Infiniband and Myrinet as interconnects, and can make a cluster "look" like a big monolithic machine. If you as an end user write code that goes down into the TCP stack itself, you're working too hard, and you're going about it the wrong way.
Put it this way: In 5 years the odds are overwhelming that there will be a good software platform that can let you pick 5000 servers and run your app 10,000 threaded, with everything appearing just like a single process, and running "as it would on a Cray." It's easier to solve this stuff with software -- take your problem (distributed computing) and solve the problem with a different set of technologies (high performance/low latency interconnects, shared address space/DMA across machines, etc).
Apple's Xgrid is a step in this direction. It's missing a ton of "Supercomputer" functionality right now, but it's a nice cross-machine GUI scheduler. Right now this type of app can address maybe 20% of what supercomputer apps need... in the future maybe more like 98%.
I had an SN41G2, the noise issue was 100% due to the video card fan.
The issue is, the video card fan points right at the holes in the side of the case, and is about 1/4" from them. So you get loud forced air out the small vent holes, which gets very noisy. I drilled the vent holes out a little larger, but it was still louder than I wanted.
Mine was actually quieter with the case open, because this blown noise didn't exist. But then you got more drive noise.
I ended up selling it and getting a Coolermaster. Though the SFF was very cool.
GF just bought a car... Ford Escape. $22,000 off the lot. $5K, 5 year financing at 2.9% (could get 3 year at 0.9% but hell, 2.9% for 5 years is pretty damn attractive).
About $350 a month. So no $300/month isn't unreasonable.
Have you evern run a "modern game" on a multi CPU machine?
:-P) back when the fastest CPU available was an 800 MHz. The frame rates I got on the dual 733 with my vid card were about 5% less than benchmark sites got on a *single* 800. So... the multiple threads don't pan out in practice.
Trust me, I've done it, and I know that Dual CPUs aren't "all that" for gaming. I got a Dell Precision 420 (2x733 P3) for gaming (gasp, $4K, what was I thinking? Whoever says PCs are cheaper than Macs
And I don't think it's gotten any better... games just really aren't optimized for throughput. And "offloading to the GPU in a second thread"... nope.
My strong, experienced advice, says if you want the best gaming experience, go with a single fast CPU, and a fast video card. Dual CPUs are for servers or for desktop tasks when you want enhanced responsiveness. They're not for gaming.
I turned on SMP support with Quake 3 Arena and it crashed all the time. It was unusable. So I turned it off.
Practically, I don't think SMP support is all that "baked" in games. My experience has shown a single, fast cpu is much better than multiple, "less fast" cpus. Significantly.
To a Mac near you!
;-)
Though, UT2004 runs quite well on a Mac. As do Call of Duty and BF 1942. Halo is a bit slow.
This at 1920x1200 resolution (23" Cinema Display)... played at WWDC on a 2x2.0 G5 with a Radeon 9800. Frame rates at that res were pretty consistent 70 FPS, never dropping below 40. So it's not ALL bad.
Though BUYING a Mac specifically to be a gaming machine, I might not advise that
Post as "Plain Old Text" rather than HTML, then. No breaks needed ;-)
What do you mean "not backwards compatible?"
DVI doesn't support the Apple resolution.
And VGA? Feh. Nobody's going to want the distortion/poor quality of VGA on a $3300 monitor.
It's gonna run like crap. Dual CPUs don't make any difference in games, and your FX 5200 is hardly better than a GF3, it's a total low-end card.
You could probably run it, but performance would be awful. Why "experiment" with your $60?
Get off the computer and go to an XTC laced rave/rock concert, already!
Well, they all finance the car for $300/month, so who cares about the depreciation?
Plus, parking is cheap on the lawn.
Well, when you compare it to OS X:
;-)
:-P
15 minutes, and 100% tuned to the hardware.
It's still a bit slow
And if I bill out at $150/hour, well your OS install costs a few thousand more than mine does
I'm sure the most significant percentage is still running 7.3 and 8.x. So neither ;-)
Samba is also limited in the 10-client version. You need the unlimited version for > 10 connections.
;-)
I promise I'm right on this
An OS X Server (Xserve), is appropriate most anyplace a Linux server is appropriate. The two are very similar -- they run the same open source compoenents, they're both UNIX-y, etc.
So they're both excellent replacements for Windows servers when you're tired of playing CALs. A Mac server is easier for a "novice" to set up and maintain, as the GUI configuration support is superior. It's also a lot easier to integrate Macs with Windows environments than it is with Linux -- OpenDirectory is very easy to integrate with Active Directory via kerberos (3 clicks and you're done, and you can have centralized directory services cross-platform)... with Linux you can install OpenLDAP but good luck getting it integrated with Kerberos in less than a month.
Cost is fairly similar -- out the door the Mac is slightly more, but if you use "Enterprise" Linux like Red Hat Enterprise Linux, then the costs are fairly similar (as the Xserve includes the software license).
So a better question is "where isn't it appropriate?" And the answer to this is usually "when the company is locked into an Exchange server and doesn't want to move off."
How about because anything over 256 kbps is INDISTINGUISHABLE from the original content? People (even the best audiophiles), cannot tell.
So why you'd want to take up 5x the space (or more) for lossless compression, I can't say.
It makes the difference between being able to put 5000 songs or 1000 on a 40 GB iPod. For many people, 1000 songs is not enough. And you're doing this for something you can't even notice a difference on, in fact only your dog can. And he probably doesn't like your music anyway.
You can't memorize icons, but you can memorize 417 different Emacs control-key combinations?
Sounds fishy.
Oracle is keenly interested in the TCO of its customers.
/dev/null.
Specifically, because given a fixed TCO, if the OS is "less expensive" there is more margin for Oracle in the sale.
For example, 5 years ago, a customer would buy Oracle to run a "big database" on an E10K with Solaris. They may pay $1M for a 3 year contract for the Oracle license, and $1M for the hardware and software maintenance from Sun (these numbers are rough, order of magnitude figures).
These days, the customers can run the Oracle database on cheap, commodity Linux. And use Oracles "RAC" functionality to load balance the DB across the boxes. So the hardware "piece of the pie" changes to $5K (decent dual CPU Opteron with 8 MB of RAM), times 15, so $75K for hardware. They pay Oracle $1.2M for software, RAC support, and Oracle supports their RHEL 3.0 installation (and kicks back some money to Red Hat).
So NOW, Oracle's piece of the pie is better. AND EVEN MORE, the customer is now even more tied in to Oracle's DBMS, because they're using the 10g RAC functionality for load balancing, fault tolerance, failover, scalability, yadda yadda. What happens in 3 years? Yep, license costs go up.
But again, overall TCO has improved in this case... DBMS costs have gone from $2M every 3 years, to $1.5M... a 25% savings. And the sales rep and the buyer can go golf together at Pebble, because they both saved/made more money for their companies.
And Sun can lay off more people, because their bigwig fat cat sales managers can't realize that the last E10K was sold about 18 months ago, and their margins are going straight to
Ummm... no you don't. Only if they want admin privileges.
Ummm... so last I checked 100 mbit was what 12 MB/sec? Or 10 MB/sec give or take? So even your 5 year old drive can sustain reads at > 10 MB/sec. So 3-4 times would be maybe 30-40 MB/sec.
Even assuming the drive cannot saturate a gigabit link (even at 50% utilization), which is a point I agree with, the drive certainly CAN beat 100 Mbit. So, again, my point is, given this, why not spend the extra $2 for a GigE adapter?
I mean, how different is this to using a USB interface on a portable HD instead of a USB2 device? In this day and age, that would be stupid now wouldn't it?
I'm fully aware of the differences in FC performance, and still this doesn't affect anything. I have an external, network attached storage device. I have full Gigabit (switches, NICs in my machines -- again, at a CONSUMER price point). So attaching storage via 100 Mbit? I'll look at perf #'s between this device and its competition with GigE, and choose the faster one. Methinks it isn't the 100 Mbit model.
Don't know abput pam_ified, sorry.
But login window is kerberized. Kerberos is the way authentication is being done, so you'd want to kerberize your services. Another pluggable authentication layer would be superfluous.
Where do you get the 18 months? 10.2 and 10.3 were about 14 months apart.
10.0 to 10.1 was 6 months, 10.1 to 10.2 was 18 months, and 10.2 to 10.3 was 14 months. So where's 18? Pulled from a hat?
I really doubt Steve's going to get into a feature play-up and then the OS won't ship for 12 months.