For the record, with a good controller (or software), a 4-drive RAID 1+0 is 4x read, 2x write, not simply 2x.
Theoretically that's true... but overly optimistic. The reality is more that with a 4-drive RAID10 you'll see somewhere between 2.0x and 3.0x read speeds over a single drive. (And with 2.6.17 Linux, I have yet to see anything better then the low-end of that scale for 4-disk or 6-disk RAID10 arrays.)
Although I still plan on doing some more testing using 3-disk RAID1 arrays (3 disks, mirrored at the same time). Which is a nice trick that mdadm lets you accomplish.
Buy your hard drives OEM from your local beige-box builder instead of from a retailer like Future Shit or Worst Buy, and you get a 3 t 5 year OEM warranty instead of the stupid 1-year warranty, and you pay a bit less.
Ever had any issues with RMAs? Also, what do you ship the drives back in for the RMA? Seems like all of the OEM drives I've seen lately are only packed inside either a plastic tub or a static bag wrapped in bubble wrap.
How picky are the RMA folks about how you pack the drive for shipment back to them?
I'd guess that they died early due to either heat or poor power.
But in general... I agree that you don't screw around with a working RAID array. Leave the disks alone until it's time to retire the entire array. Keep a hot-spare (or two) hooked up to the array and run something sensible (like RAID6 or RAID10).
Replacing disks on a schedule is a fool's errand. Plan for failure instead (by having good backups and hot-spares).
I use RAID 5 and there is no way I'm ever going back to a non-RAID setup, you save loads of time in not having to back stuff up to CD or DVD.
Rules of RAID: (repeat after me)
1. RAID is not a substitute for backups
2. Your RAID array *will* scramble your data at some point in time. That or your file system will decide to perform the honors instead. And that's only if your fingers don't do the walking before that time.
3. RAID is not a substitute for backups
4. While your RAID array is rebuilding, you *will* have a 2nd drive failure at some point. This occurs especially often with RAID5 arrays where the disks suddenly become extremely busy while attempting to do the rebuild, which can lead to a premature death of a 2nd disk.
More recently I've not had a HD go for some considerable time but I've been loosing a PSU a year - they used never to die.
Sounds like it's time for a better UPS, one that filters the incoming voltages to protect against sags and surges. (I've seen power companies push >135V across the wires. My UPSs were complaining but handling it properly.)
RAID 1 + 0 is faster than RAID 5though, because no parity bit needs calculating. There are trade-offs in all the RAID configurations. Pick the one that suits your needs best.
The other advantage is rebuild time. When a drive dies in a RAID10 array, rebuild time is related to the size of individual disks in the array rather then the overall size of the array. I'd hate to see what the rebuild time is for a RAID5 across 6 750GB SATA drives, but I know that RAID10 rebuild time is only 180 minutes (the time for a single RAID1 pair to rebuild itself).
With mdadm, you can even do fancy things like:
Multiple RAID1 sets using 3 disks in each RAID1 set with RAID0 layered over the top. Now you can withstand even more disk failures in the array before you lose data. The downside is net capacity of only 33%.
On one of our servers, the boot and root partitions are RAID1 across 6 disks (yes... 6 copies of the partition) plus a hot-spare partition on the 7th disk. The nice part about that is you never have to worry about GRUB/LILO booting an outdated boot partition since the boot partition across all 6 disks are kept in perfect sync. Which is a scenario that I ran into when I had a simple RAID1 across 2 partitions with 5 hot-spare partition slices on the other disks.
I haven't tested yet to see if a 6-disk RAID1 offers increased read speeds over a 2-disk RAID1 array...
The main downside of Software RAID is if you are CPU constrained or if the Software RAID traffic will bottleneck on your I/O bus between the CPU and the disks. Hardware RAID keeps the additional disk traffic off the CPU and off the I/O bus.
Re:So what is the oldest continuous content?
on
The Web Is 16 Today
·
· Score: 1
I don't know when Kevin and Kell made the jump from CompuServe to the web, but he's been putting out strips since the early 90s.
For servers, you should look at NIC bonding where you use 2+ cables between the server and the switch. Dual-NIC Intel cards are only $170 or so and there are even quad-NIC cards (or you could just install two dual-NIC cards). That will give you 2-4 gigabits per server of bandwidth to share among the workstations.
You may even want to consider NIC bonding for the workstations, but that will require extra wire runs. Or maybe you get an inexpensive 16 port switch (~$240) with trunking and put it into the middle of a cluster of 4-8 workstations. Swipe a few of the links back to the wiring closet and trunk them together with the workstations connecting to the new gigabit switch.
Heck, our network is still only a gigabit core switch with 10/100 hubs hung off of it. I'm still trying to get away from hubs and make to the move to gigabit switches. (But the gigabit switch at the core was a good investment.)
I've seen unmanaged (but with web interfaces, a.k.a. "smart") 48-port gigabit switches for around $1500. I plan on using one of those for the non-critical desktops and it will allow us to retire a pair of 24-port 10/100 hubs.
My business class DSL with Verizon in Pennsylvania is 1.5Mbit down, 368Kbit up and I'm paying around $140/mo.
But that's a completely unfiltered connection, static IP, no caps, priority support, no hassles about how much bandwidth I use. Which is good, because I'm a digital packrat and I always have something downloading in the background.
RAID5 won't help. It's good at reads, but generally sucks for writes and rebuild times are often horrid. It's major claim to fame is net capacity.
If you want increased throughput (and maybe slightly faster seek times) go with 2 or 3 disk RAID0. If you want peace of mind, go with a 4 or 6 disk RAID10 setup which will be 2x to 3x faster (throughput) then a single disk or a single RAID1 set. Net capacity is the same as RAID1, half the gross capacity.
I have a few RAID10 arrays now and I keep wanting to build even larger ones. (We either setup a 4-disk + hot-spare or a 6-disk + hot-spare configuration.) The nice part about RAID10 is that even when you get the array thrashing with random seeks, there's a decent performance floor that you can rely on. A decent RAID card will balance reads between the two disks in each RAID1 set.
And, of course, higher RPM results in better seek times which translates into better performance. (Although a 6-disk RAID10 array of 7200rpm 750GB SATA drives is nothing to sneeze at...)
Consumers are happy now. computers have stagnated hard for the past 3-4 years and the performance gains offered by this new stuff is only marginal for them.
That's pretty true, but inexpensive multi-core might change that.
The big issue since 2001/2002 is that processor power for a single-core CPU has only barely doubled over a period of about 4 years. Which means that a machine from 2002 feels slow, but not unbearably slow compared to a brand new single-core machine. (I should know, my laptop is almost 5 years old now and I use it daily...). That's a far cry from the rapid doubling of processor power every 12-15 months in the mid-late 90s when a 3-4 year old machine was 8x to 12x slower then a brand new machine.
However, now that inexpensive ($150 or less, the $300 price from last year was way too expensive) dual-core chips are here, I think we're going to see a large shift take place. Savvy consumers who have used a multi-core machine for more then a few minutes are going to notice the responsiveness compared to their old single-core machine.
Will they bury a dual-core CPU in work? No more then they buried their old single-core machine. But at least with the multi-core unit, they're more likely to have a free core available to respond to their input without lag or stutter.
(We switched over to buying dual-core CPUs this summer for all of our new machines at work. Even the low-end administrative people who only shuffle documents. A conservative guess is that we'll easily get 8 years out of these machines before they seem too slow to be useful. Maybe as long as 12 years if none of the parts fail.)
Bah, genie & compuserve. I laughed about you posers on usenet, using my university dialup account. Of course, I had to pay for internet after I graduated and moved away. Everyone thought I was insane. Paying for internet?!? What's an internet!?!
Back in the 1990 timeframe (give or take a few years), it seemed like almost every major software and hardware vendor had a message area on CompuServe. Some vendors only had a single forum, others had multiple forms (and their own "go" word).
It was an interesting time. CIS definitely didn't adapt fast enough to the advent of ISPs, SMTP e-mail and HTTP/WWW. I think I finally bailed out of CIS sometime in the late 90s after getting a dial-up account through IBM (or some other local ISP).
Heh, I remember the prevalence of "under construction" signs in the mid-90s on the web. It was definitely something that the advent of money interests shoved aside. There was a certain innocence and naivety at that point about putting up not-quite-complete pages and fixing them up later.
I don't think I've seen an "under construction" sign in quite a while. But then, the majority of web browsing that I do is either forums, blogs, news, or shopping sites.
To build your own system you have to know exactly which CPU will work together with which motherboard
Nonsense... go with a motherboard bundle from a shop like MWave.com where they'll put together the bundle (and test it) for only $9. Well worth it (and I build half a dozen systems per year) because they do the research leg work for me.
Cases come with bags of screws (Antec, Lian Li).
You can do a dual-core, 2GB RAM machine for about $600 assuming all new parts. Office Pro and WinXP Pro will add another $430 to the price. The key components will have a 3 year warranty and you can shop around for decent warranties on the rest. If you want a good video card, you can drop one in for around $150.
Even tough technology is still improving, just how often are computer manufacturers expecting consumers to purchase a brand new PC? It sounds to me that the slowing growth is more in part due to market saturation than anything else. Computer sales have enjoyed double digit growth through more difficult times than these (Windows ME fiasco, I.LOVE.YOU viruses, massive job loss in the bubble burst, terror attacks and wars, various US and foreign stock upsets, etc).
The bigger issue is that PC speed increases have slowed dramatically over the past 6 years. Back in the mid-late 90s, performance doubled every 12-15 months and that 3 year old machine was in drastic need of replacement. (Since it performed close to an order of magnitude slower then a brand new machine.)
Now look at the rate of increase from 2002-2006 (excluding multi-core). A 2002 machine is still about half as fast as a 2006 machine. So there's no huge incentive to upgrade if your machine was made in 2002 or later. Only gamers and other CPU-hungry users are upgrading more often then every 5 years. A lot of companies are moving from replacing machines every 3 years to only replacing them every 5-7 years.
Multi-core is changing the landscape slightly. With the lower-prices on dual-core, and if you don't want to deal with Vista until 2008-2009 (or after the 1st or 2nd service pack comes out), it's a good time to build a dual-core WinXP machine. You'll get a machine that will last 5-8 years for only $1000 (including MS Office software) with 2GB of RAM and dual-core. With luck you can probably even stretch that lifespan out to 10-12 years.
All I have to say is RAID10... go play with a 4, 6 or 8 disk RAID10 array sometime. Net utilization is lower then RAID5, but makes up for it in speed and predictable performance (and predictable recovery times).
Because once they are, the duals I want will become the cheap alternative.
I hear that. We postponed and put off and postponed our desktop upgrade schedule for almost 2 years while waiting for dual-core CPUs to drop in price.
The price cuts by AMD in late-July this year were extremely welcome. All of the new machines (and I mean ALL, even the people who will never max it out) are now coming in the door as dual-core. I figure we'll easily squeeze 8-12 years of lifespan out of those machines before we have to replace them.
(We still have a bunch of Win98 machines from 1998-2000... long past their prime.)
And what I mean by that is that if you want to run the latest whiz-bang OS or Office suite from MS, all I can say is good luck trying. If you want a machine to last five years and you are a Windows user, you MUST pay the $2000+ pricetag for mid-level computing. If you buy an $800 box now with Windows XP and it runs OK, I can guarantee you that you'll be spending a lot more money trying to upgrade it to run Vista or whatever comes after vista than you would have had you just plunked down the extra $1200.
That was true in the late 90s, but ceased to be true past about 2001. In the mid-90s, my rule was you should spend $1000 per year of lifespan that you wanted. So a $3k machine would last you about 3 years before it was hopelessly outclassed by a brand new machine.
But time has marched on, and the rate of progress has slowed dramatically. Now you can build a machine for about $950 (dual-core, 2GB RAM, quality parts) that will easily last 5 years for a moderate user or as long as 10 years for a lightweight user. A power user machine where they need lots of CPU power can be done for $1500.
And that power user's machine can be handed off after 4-5 years to a less demanding user who can use it for another 5 years.
Dual and quad core drastically change the landscape for how long a machine will seem "fast enough" for a user. By having multiple cores in the system, you gain a lot of responsiveness in the UI. Which means the user doesn't necessarily care that task X takes 30 seconds instead of 10, because it isn't slowing them down for doing the other 3 things that they're working on.
Well, I'm in the process of paring down from about half a dozen machines to just one. And that's going to be just a lowly dual-core X2 w/ 4GB of RAM and lots of hard disk space. Unless you're running the CPUs flat-out across all 16 machines, you could probably already start making the move to dual-core or twin-CPU dual-core boxes.
(Long live Xen...)
Being able to setup scratch servers in a virtual environment without having to put hardware together is rather addictive. Instead of giving every developer their own physical "play" box, I can set them up with virtual guest servers instead. Which is a huge savings over the old method of individual boxes.
(And yes, we'll be buying the AM3 quad-cores when they come out next year and upgrading, hopefully, our existing X2s.)
Well, that's only $28/GB which isn't too far off once you get into high-level hardware, 20% net utilization (dual RAID-10 arrays plus hot spares), SCSI drives that were probably built a few years ago (36/72GB disks with a higher $/GB), backups, the SAN hardware, etc.
Even a low-end SATA storage unit runs about $2-$3/GB. And with SCSI drives you're typically looking at $6-$8/GB at the low-end with it being pretty easy to get up into the $15/GB range once you include stuff like backups.
3. In a laptop, HDD speed makes a huge difference in the overall performance. The jump in performance on a 5 year old laptop between a 5400 and 7200 RPM drive is not only noticible, but amazing. So anything that further bottlenecks one of the worst bottlenecks for a laptop in the first place seems like a bad idea to me.
Encryption speed is not likely to be a bottleneck, as long as you're using a CPU speed over 1.5GHz against a laptop drive (probably even 7200 RPM). Go look at the TrueCrypt forums and you'll find a benchmark thread that lists the potential encryption / decryption speeds of the various algorithms on different hardware. Or you can download and install TrueCrypt and run the built-in benchmark tool.
On a newer dual-core CPU, it's even less likely to be CPU-bound.
Would Seagate really attempt to market a drive that was going to protect pedophiles and terrorists? (Not to mention us ordinary citizens who don't wholly and utterly trust the organs of the state to act systematically in our best interests.)
Wrong target market... think Digital Restrictions Management (DRM) in things like PVRs.
Slashdot Mirror
For the record, with a good controller (or software), a 4-drive RAID 1+0 is 4x read, 2x write, not simply 2x.
Theoretically that's true... but overly optimistic. The reality is more that with a 4-drive RAID10 you'll see somewhere between 2.0x and 3.0x read speeds over a single drive. (And with 2.6.17 Linux, I have yet to see anything better then the low-end of that scale for 4-disk or 6-disk RAID10 arrays.)
Although I still plan on doing some more testing using 3-disk RAID1 arrays (3 disks, mirrored at the same time). Which is a nice trick that mdadm lets you accomplish.
Buy your hard drives OEM from your local beige-box builder instead of from a retailer like Future Shit or Worst Buy, and you get a 3 t 5 year OEM warranty instead of the stupid 1-year warranty, and you pay a bit less.
Ever had any issues with RMAs? Also, what do you ship the drives back in for the RMA? Seems like all of the OEM drives I've seen lately are only packed inside either a plastic tub or a static bag wrapped in bubble wrap.
How picky are the RMA folks about how you pack the drive for shipment back to them?
The Maxtors lasted about 6 months each.
I'd guess that they died early due to either heat or poor power.
But in general... I agree that you don't screw around with a working RAID array. Leave the disks alone until it's time to retire the entire array. Keep a hot-spare (or two) hooked up to the array and run something sensible (like RAID6 or RAID10).
Replacing disks on a schedule is a fool's errand. Plan for failure instead (by having good backups and hot-spares).
I use RAID 5 and there is no way I'm ever going back to a non-RAID setup, you save loads of time in not having to back stuff up to CD or DVD.
Rules of RAID: (repeat after me)
1. RAID is not a substitute for backups
2. Your RAID array *will* scramble your data at some point in time. That or your file system will decide to perform the honors instead. And that's only if your fingers don't do the walking before that time.
3. RAID is not a substitute for backups
4. While your RAID array is rebuilding, you *will* have a 2nd drive failure at some point. This occurs especially often with RAID5 arrays where the disks suddenly become extremely busy while attempting to do the rebuild, which can lead to a premature death of a 2nd disk.
5. RAID is not a substitute for backups
More recently I've not had a HD go for some considerable time but I've been loosing a PSU a year - they used never to die.
Sounds like it's time for a better UPS, one that filters the incoming voltages to protect against sags and surges. (I've seen power companies push >135V across the wires. My UPSs were complaining but handling it properly.)
RAID 1 + 0 is faster than RAID 5though, because no parity bit needs calculating. There are trade-offs in all the RAID configurations. Pick the one that suits your needs best.
The other advantage is rebuild time. When a drive dies in a RAID10 array, rebuild time is related to the size of individual disks in the array rather then the overall size of the array. I'd hate to see what the rebuild time is for a RAID5 across 6 750GB SATA drives, but I know that RAID10 rebuild time is only 180 minutes (the time for a single RAID1 pair to rebuild itself).
With mdadm, you can even do fancy things like:
Multiple RAID1 sets using 3 disks in each RAID1 set with RAID0 layered over the top. Now you can withstand even more disk failures in the array before you lose data. The downside is net capacity of only 33%.
On one of our servers, the boot and root partitions are RAID1 across 6 disks (yes... 6 copies of the partition) plus a hot-spare partition on the 7th disk. The nice part about that is you never have to worry about GRUB/LILO booting an outdated boot partition since the boot partition across all 6 disks are kept in perfect sync. Which is a scenario that I ran into when I had a simple RAID1 across 2 partitions with 5 hot-spare partition slices on the other disks.
I haven't tested yet to see if a 6-disk RAID1 offers increased read speeds over a 2-disk RAID1 array...
The main downside of Software RAID is if you are CPU constrained or if the Software RAID traffic will bottleneck on your I/O bus between the CPU and the disks. Hardware RAID keeps the additional disk traffic off the CPU and off the I/O bus.
I don't know when Kevin and Kell made the jump from CompuServe to the web, but he's been putting out strips since the early 90s.
(Strips were originally posted on CompuServe.)
Aye, consider your network topology.
For servers, you should look at NIC bonding where you use 2+ cables between the server and the switch. Dual-NIC Intel cards are only $170 or so and there are even quad-NIC cards (or you could just install two dual-NIC cards). That will give you 2-4 gigabits per server of bandwidth to share among the workstations.
You may even want to consider NIC bonding for the workstations, but that will require extra wire runs. Or maybe you get an inexpensive 16 port switch (~$240) with trunking and put it into the middle of a cluster of 4-8 workstations. Swipe a few of the links back to the wiring closet and trunk them together with the workstations connecting to the new gigabit switch.
Heck, our network is still only a gigabit core switch with 10/100 hubs hung off of it. I'm still trying to get away from hubs and make to the move to gigabit switches. (But the gigabit switch at the core was a good investment.)
I've seen unmanaged (but with web interfaces, a.k.a. "smart") 48-port gigabit switches for around $1500. I plan on using one of those for the non-critical desktops and it will allow us to retire a pair of 24-port 10/100 hubs.
My business class DSL with Verizon in Pennsylvania is 1.5Mbit down, 368Kbit up and I'm paying around $140/mo.
But that's a completely unfiltered connection, static IP, no caps, priority support, no hassles about how much bandwidth I use. Which is good, because I'm a digital packrat and I always have something downloading in the background.
RAID5 won't help. It's good at reads, but generally sucks for writes and rebuild times are often horrid. It's major claim to fame is net capacity.
If you want increased throughput (and maybe slightly faster seek times) go with 2 or 3 disk RAID0. If you want peace of mind, go with a 4 or 6 disk RAID10 setup which will be 2x to 3x faster (throughput) then a single disk or a single RAID1 set. Net capacity is the same as RAID1, half the gross capacity.
I have a few RAID10 arrays now and I keep wanting to build even larger ones. (We either setup a 4-disk + hot-spare or a 6-disk + hot-spare configuration.) The nice part about RAID10 is that even when you get the array thrashing with random seeks, there's a decent performance floor that you can rely on. A decent RAID card will balance reads between the two disks in each RAID1 set.
And, of course, higher RPM results in better seek times which translates into better performance. (Although a 6-disk RAID10 array of 7200rpm 750GB SATA drives is nothing to sneeze at...)
Athlon 64 EE 3800+
Do you know off-hand what the wattage is for the EE vs non-EE parts?
Consumers are happy now. computers have stagnated hard for the past 3-4 years and the performance gains offered by this new stuff is only marginal for them.
That's pretty true, but inexpensive multi-core might change that.
The big issue since 2001/2002 is that processor power for a single-core CPU has only barely doubled over a period of about 4 years. Which means that a machine from 2002 feels slow, but not unbearably slow compared to a brand new single-core machine. (I should know, my laptop is almost 5 years old now and I use it daily...). That's a far cry from the rapid doubling of processor power every 12-15 months in the mid-late 90s when a 3-4 year old machine was 8x to 12x slower then a brand new machine.
However, now that inexpensive ($150 or less, the $300 price from last year was way too expensive) dual-core chips are here, I think we're going to see a large shift take place. Savvy consumers who have used a multi-core machine for more then a few minutes are going to notice the responsiveness compared to their old single-core machine.
Will they bury a dual-core CPU in work? No more then they buried their old single-core machine. But at least with the multi-core unit, they're more likely to have a free core available to respond to their input without lag or stutter.
(We switched over to buying dual-core CPUs this summer for all of our new machines at work. Even the low-end administrative people who only shuffle documents. A conservative guess is that we'll easily get 8 years out of these machines before they seem too slow to be useful. Maybe as long as 12 years if none of the parts fail.)
Bah, genie & compuserve. I laughed about you posers on usenet, using my university dialup account. Of course, I had to pay for internet after I graduated and moved away. Everyone thought I was insane. Paying for internet?!? What's an internet!?!
Back in the 1990 timeframe (give or take a few years), it seemed like almost every major software and hardware vendor had a message area on CompuServe. Some vendors only had a single forum, others had multiple forms (and their own "go" word).
It was an interesting time. CIS definitely didn't adapt fast enough to the advent of ISPs, SMTP e-mail and HTTP/WWW. I think I finally bailed out of CIS sometime in the late 90s after getting a dial-up account through IBM (or some other local ISP).
Heh, I remember the prevalence of "under construction" signs in the mid-90s on the web. It was definitely something that the advent of money interests shoved aside. There was a certain innocence and naivety at that point about putting up not-quite-complete pages and fixing them up later.
I don't think I've seen an "under construction" sign in quite a while. But then, the majority of web browsing that I do is either forums, blogs, news, or shopping sites.
Screen is still alive and kicking, and I'm finding that I now do more stuff in a shell then ever before (mostly configuring Xen domain servers...).
To build your own system you have to know exactly which CPU will work together with which motherboard
Nonsense... go with a motherboard bundle from a shop like MWave.com where they'll put together the bundle (and test it) for only $9. Well worth it (and I build half a dozen systems per year) because they do the research leg work for me.
Cases come with bags of screws (Antec, Lian Li).
You can do a dual-core, 2GB RAM machine for about $600 assuming all new parts. Office Pro and WinXP Pro will add another $430 to the price. The key components will have a 3 year warranty and you can shop around for decent warranties on the rest. If you want a good video card, you can drop one in for around $150.
Even tough technology is still improving, just how often are computer manufacturers expecting consumers to purchase a brand new PC? It sounds to me that the slowing growth is more in part due to market saturation than anything else. Computer sales have enjoyed double digit growth through more difficult times than these (Windows ME fiasco, I.LOVE.YOU viruses, massive job loss in the bubble burst, terror attacks and wars, various US and foreign stock upsets, etc).
The bigger issue is that PC speed increases have slowed dramatically over the past 6 years. Back in the mid-late 90s, performance doubled every 12-15 months and that 3 year old machine was in drastic need of replacement. (Since it performed close to an order of magnitude slower then a brand new machine.)
Now look at the rate of increase from 2002-2006 (excluding multi-core). A 2002 machine is still about half as fast as a 2006 machine. So there's no huge incentive to upgrade if your machine was made in 2002 or later. Only gamers and other CPU-hungry users are upgrading more often then every 5 years. A lot of companies are moving from replacing machines every 3 years to only replacing them every 5-7 years.
Multi-core is changing the landscape slightly. With the lower-prices on dual-core, and if you don't want to deal with Vista until 2008-2009 (or after the 1st or 2nd service pack comes out), it's a good time to build a dual-core WinXP machine. You'll get a machine that will last 5-8 years for only $1000 (including MS Office software) with 2GB of RAM and dual-core. With luck you can probably even stretch that lifespan out to 10-12 years.
All I have to say is RAID10... go play with a 4, 6 or 8 disk RAID10 array sometime. Net utilization is lower then RAID5, but makes up for it in speed and predictable performance (and predictable recovery times).
it's amazing what halving your seek time and doubling your throughput will do for your computer :P
I have yet to see 2-disk RAID0 halve the seek times. It definitely doubles your throughput, but I don't think it has any effect on seek times.
(I could be wrong, or Linux Software RAID just isn't that good for that particular stat.)
Because once they are, the duals I want will become the cheap alternative.
I hear that. We postponed and put off and postponed our desktop upgrade schedule for almost 2 years while waiting for dual-core CPUs to drop in price.
The price cuts by AMD in late-July this year were extremely welcome. All of the new machines (and I mean ALL, even the people who will never max it out) are now coming in the door as dual-core. I figure we'll easily squeeze 8-12 years of lifespan out of those machines before we have to replace them.
(We still have a bunch of Win98 machines from 1998-2000... long past their prime.)
And what I mean by that is that if you want to run the latest whiz-bang OS or Office suite from MS, all I can say is good luck trying. If you want a machine to last five years and you are a Windows user, you MUST pay the $2000+ pricetag for mid-level computing. If you buy an $800 box now with Windows XP and it runs OK, I can guarantee you that you'll be spending a lot more money trying to upgrade it to run Vista or whatever comes after vista than you would have had you just plunked down the extra $1200.
That was true in the late 90s, but ceased to be true past about 2001. In the mid-90s, my rule was you should spend $1000 per year of lifespan that you wanted. So a $3k machine would last you about 3 years before it was hopelessly outclassed by a brand new machine.
But time has marched on, and the rate of progress has slowed dramatically. Now you can build a machine for about $950 (dual-core, 2GB RAM, quality parts) that will easily last 5 years for a moderate user or as long as 10 years for a lightweight user. A power user machine where they need lots of CPU power can be done for $1500.
And that power user's machine can be handed off after 4-5 years to a less demanding user who can use it for another 5 years.
Dual and quad core drastically change the landscape for how long a machine will seem "fast enough" for a user. By having multiple cores in the system, you gain a lot of responsiveness in the UI. Which means the user doesn't necessarily care that task X takes 30 seconds instead of 10, because it isn't slowing them down for doing the other 3 things that they're working on.
Well, I'm in the process of paring down from about half a dozen machines to just one. And that's going to be just a lowly dual-core X2 w/ 4GB of RAM and lots of hard disk space. Unless you're running the CPUs flat-out across all 16 machines, you could probably already start making the move to dual-core or twin-CPU dual-core boxes.
(Long live Xen...)
Being able to setup scratch servers in a virtual environment without having to put hardware together is rather addictive. Instead of giving every developer their own physical "play" box, I can set them up with virtual guest servers instead. Which is a huge savings over the old method of individual boxes.
(And yes, we'll be buying the AM3 quad-cores when they come out next year and upgrading, hopefully, our existing X2s.)
$28K per terabyte?
Well, that's only $28/GB which isn't too far off once you get into high-level hardware, 20% net utilization (dual RAID-10 arrays plus hot spares), SCSI drives that were probably built a few years ago (36/72GB disks with a higher $/GB), backups, the SAN hardware, etc.
Even a low-end SATA storage unit runs about $2-$3/GB. And with SCSI drives you're typically looking at $6-$8/GB at the low-end with it being pretty easy to get up into the $15/GB range once you include stuff like backups.
3. In a laptop, HDD speed makes a huge difference in the overall performance. The jump in performance on a 5 year old laptop between a 5400 and 7200 RPM drive is not only noticible, but amazing. So anything that further bottlenecks one of the worst bottlenecks for a laptop in the first place seems like a bad idea to me.
Encryption speed is not likely to be a bottleneck, as long as you're using a CPU speed over 1.5GHz against a laptop drive (probably even 7200 RPM). Go look at the TrueCrypt forums and you'll find a benchmark thread that lists the potential encryption / decryption speeds of the various algorithms on different hardware. Or you can download and install TrueCrypt and run the built-in benchmark tool.
On a newer dual-core CPU, it's even less likely to be CPU-bound.
Would Seagate really attempt to market a drive that was going to protect pedophiles and terrorists? (Not to mention us ordinary citizens who don't wholly and utterly trust the organs of the state to act systematically in our best interests.)
Wrong target market... think Digital Restrictions Management (DRM) in things like PVRs.