Slashdot Mirror

Yep.. and guess who created the 2.88 floppy?

None other than Toshiba.

I'm not sure about the BIOS, but you're correct regarding the controller. PCGuide says the 500Kbit limitation of existing floppy controllers was insufficient; the 2.88 floppies required a 1Mbit transfer rate. I'm not sure why the drives couldn't be slowed down for the sake of compatibility though. Seems easy enough to throw a jumper on there to toggle 500Kbit/1Mbit transfer rates, but I'm no EE.

http://www.pcguide.com/ref/hdd/op/actMultiple-c.ht ml

In fact, such hard disks have been built. Conner Peripherals, which was an innovator in the hard disk field in the late 1980s and early 1990s (they later went bankrupt and their product line and technology were purchased by Seagate) had a drive model called the Chinook that had two complete head-actuator assemblies: two sets of heads, sliders and arms and two actuators. They also duplicated the control circuitry to allow them to run independently. For its time, this drive was a great performer. But the drive never gained wide acceptance, and the design was dropped. Nobody to my knowledge has tried to repeat the experiment in the last several years.

It looks like Conner Peripheral's Chinook line of hard drives did that.

http://www.pcguide.com/ref/hdd/op/act_Servo.htm

The dedicated servo track that would make it possible to read multiple tracks at a time is obsolete, as this method is sensitive to temperature variations between the platters.

The embedded servo track is used by all modern drives, which intersperses the servo information with the data.

For clarification, I should note that I worked on the OS driver for a 4-head parallel drive, rather than the actual disk -- which explains my ignorance about the details of the drive technology.

http://www.pcguide.com/ref/ram/logic-c.html

I used to work for city government here in SoCal, USA. In contrast to our Aussie friends, they were super paranoid about data leakage.

This does not, in any way, reflect on "Aussies" or their awareness of the importance of media sanitization.

When there was actually a situation where the red tape was momentarily pierced and we were authorized to give away outdated equipment to schools, they made us do a multiple-pass low-level format on each and every HDD that left the building.

Are you sure you were "low level" formatting those drives? That is a term that gets used often when it should not. Modern IDE drives cannot be low level formatted outside the factory and this has been the case for many years. A true low level format actually re-writes tracks, aligning them again as it goes.

"all modern hard disks are low-level formatted at the factory for the life of the drive. There's no way for the PC to do an LLF on a modern IDE/ATA or SCSI hard disk, and there's no reason to try to do so."

Unfortunately, this term has become so misused, that even hard drive manufacturers are now providing zero-fill utilities labeled as low-level-format utilities.

I have worked for the Australian Government in sanitizing machines prior to them being decommisioned. Luckily, I am a contractor who takes his contracts, customers and their needs seriously and I did not have anything to do with this case. I don't think this reflects on Australia in any way. I'm sure I could dig up similar stories regarding US or UK blunders.

A 32bit PCI bus at 33 MHz is only has a "theoretical" bandwidth of 127.2MB/s or a little over what 2 drives should do in a RAID 0. http://www.pcguide.com/ref/mbsys/buses/funcBandwid th-c.html depending on the Mother Board you may need to share with the network card and the sound card and everything else.

So, you have to read from all four disks in a four-disk array. You read from three to get the data, read from the fourth to get the parity bit, then calculate the parity to verify that your read was valid. Ok think about what your saying if you need all 4 disks then the array would stop working if one disk went bad. Now it might be a good idea to do what your saying IF you don't trust the system to detect if a disk goes bad, but you did not talk about doing this with the RAID 10 setup so either you accept the slight chance of a read error and get to use all disks independently on a read or you don't get to do this for any array.

PS: Once again I am suggesting using 2 disks + the parity disk to read sequential data and then have the 3rd disk read some other data. With a good buffer you can rotate which disk is doing the independent read and get the full sequential read speed of all 4 disks over time.

... I always imagined he was some CS heavyweight rather than a pundit.

http://www.pcguide.com/ref/kb/layout/alphaDvorak-c .html
The guy who made the keyboard is a different Dvorak, it just made the revision fit better to complain about the keyboard too :)

Does that motherboard have boot block support to still let you reflash the bios from a floppy even when your system won't boot properly? or did that also fail for you?

AGP (1x) does 254.3MBps, PCI (v1.0) does 127.2MBps. CD data is 172.4KBps. So PCI does 738 CD streams, while AGP does 1475. At 128Kbps, each stream IO is 192.4KBps duplex, so AGP: 1322 encoder streams, PCI 661 encoder streams. A P4/3GHz does 6GFLOPS, including host apps (OS, etc), while a $115 GeForce FX5900 does 20GFLOPS; a $470 GeForce 6800 Ultra does 40GFLOPS. Even at the slowest AGP and PCI speeds, and the fastest theoretical GPGPU speeds, the GPU is still slower than the bus, and 5 of those cards in PCI would have to do over 100 streams each to fill the bus, even leaving 25% PCI capacity for the host P4. That's over 500 streams for about $1000, which can't be done any other way I know of.

USB started appearing on Intel chipsets in early 1996 with the Intel 430 chipset family. You are off by "several years" plus two.

I agree that the iMac helped create a market for USB peripherals (and USB adapters) because Apple gave them no other choice. But PCs had USB in their chipsets two years before the iMac. PC makers have this strange habit of including legacy ports, so peripheral makers and OS makers put off USB support until 1998.

if microsoft is advertising pro and home as two distinct products, when in fact they are not, isn't this false advertising? it's like buying a ford with a V6 and finding that it's really a V8, just two cylinders turned off...

Intel has been doing this for years with their SMP vs nonSMP chips and years back with the math coprocessors. An example is here.

I understand that Intel's processors have actually featured a RISC arcitecture for a long time, but they run a CISC emulation layer on top of it. Some information about RISC vs CISC can be found here, and some info on the CISC emulation in Intel chips here.

The Turbo button acctually made the CPU operate at slower speed. It was really a "Turbo Off" button:
http://www.pcguide.com/ref/case/switchTurbo-c.html

- Peder

While it is true that NTFS was written from scratch. My point is that they borrowed heavily from their work on OS/2. What the two are similar in is the work that Microsoft contributed to OS/2 and learned from, which were then refined and redesigned for Windows NT. This is evident when seeing what aspects Microsoft worked on (which were well done in NT) and what IBM did (which were often less thought out in NT).

My point about converting HPFS to NTFS was that they are similar enough to make it possible. It is easy to convert from a basic filesystem (FAT) upwards. When you move down, though, its much harder because so much information is lost. The two filesystems share a number of design goals (see here) making it relatively easy to convert over without losing any information.

If you search for articles on OS/2 history, you will find many aspects where the two borrowed from each other. OS/2's original presentation manager was based on Window's 2.0's (press release) and the API was close.

Don't confuse the video card's video memory with VRAM. They are not synonymous. I've had a card in the past that had VRAM, but most newer cards do not. Most use standard SDRAM.

http://www.pcguide.com/ref/video/techVRAM-c.html

"Note: Don't confuse VRAM with the generic term "video RAM" or "video memory", which just refer to the memory in the video subsystem in general."

This card is intended for and marketed towards hard core gamers. If it had something to do with Tiger's release, it would have been worthy of mention and ATI would have mentioned it in their press release.

And on the link you provided, Apple used incorrect terminology. They call the memory on the video card "VRAM", which is incorrect. The video memory on all the cards they listed is SDRAM, none of those cards have VRAM. My old Diamond Stealth 64 VRAM had VRAM on it. VRAM is a particular type of specialized high-speed memory, not used much (if at all) anymore.

http://www.pcguide.com/ref/video/techVRAM-c.html

Do most low end motherboards handle parity or error correction in memory? I think they just break. I know that higher end computers generally do handle it.

See here

I think I still have a computer that can read an MFM drive. A quick search shows, "Compaq began selling PCs with integrated hard disks using Western Digital controllers starting with their IBM-compatible Deskpro 386 in 1986." Ok, I guess I didn't make it clear. I don't use hard disks for archive, just data transfer. Final archives are normally on tape.

I have to admit you did catch me. My 10 year old removable drives are SCSI. We added removable IDE five years ago. I haven't had one of those fail either.

I still stand behind my claim that properly stored IDE drives last a lot longer than those used 24/7 inside computers. I never claimed they were a good way to archive data.

Hmmmmmm... One of the major differences between SCSI and IDE is that when accessing an IDE drive, your system processor does a fair amount of the work. With SCSI, this work is offloaded to the SCSI controller and the drive, leavinf the system processor free to do other stuff. This is why SCSI is a much better option for servers (and also why it's more expensive).

I think part of the reason that SCSI has been such a pain is that there have been so many SCSI 'standards'. IIRC, SCSI3 defined 5 properties that the system should have, but mandated that a SCSI3 device had to implement any 2 of these. This is insane IMHO.

Things seem to have returned to some sort of sanity with the current Ultra160 and Ultra320 standards.

This article is a good description of all the SCSI standards.

Maybe they use some from of PRML.

As long as that's the only time you spin it down, sure. But note that most drives have a finite (and quite small) rated number of powercycles; I think a Maxtor I had recently was rated for around 10,000. Say you set aggressive PM on a machine that's actually used fairly frequently, and it ends up cycling 20 times per day. Great - now you have an estimated life of about 500 days. I'm reasonably sure (although I haven't done the math) that just leaving the drive running will actually save money on the replacement cost versus electrical savings, and I'm certain it'll save environmental costs.

In my experience, it's the drives that get used all the time that get hosed, not the ones that spend their nights sleeping. I don't think you need to shut down every time you go to the bathroom or anything, but not rebooting a crappy drive for 180 days is gonna eat up way more of the drive's hourly life than its start/stop cycles. For a drive rated at 150000 hours and 10000 start/stop cycles, that eats up a whole 2.88% of hours versus 0.01% of start/stop cycles. Whereas, if you had it running only 8 hours a day for those 180 days, you'd've spent only 0.96% of hours and 1.8% of cycles. Factoring in weekends, that's 1.32% of cycles. And we're talking about a shitty 10000 start/stop cycle, 150000 MTBF drive. Most desktop drives are 30k - 50k according to this PC Guide. At just 20000 cycles, as several 10000RPM drives are rated today, the figure drops to just 0.66% of cycles.

Bleh, math, anyway, point being, shutting down overnight and weekends won't hurt the hard drive within any reasonable timeframe.

-l

Why would you use RAID 01 over RAID 10?

(Answer appears to be that your controller sucks).

Here is an example:

First if three data blocks and one parity are written at once:
1010110010011 - data
1100101101100 - data
0111011010110 - data
-------------
0001000101001 - parity (XOR of all data)
(Four block writes.)

Next, if two data blocks and parity are written then later a third data block and new parity is written.
1010110010011 - data
1100101101100 - data
-------------
0110011111111 - parity (XOR of original data)
(3 block writes)

0110011111111 - parity (possibly read from disk)
0111011010110 - data
-------------
0001000101001 - parity (Note same result.)

One disk read and two writes... exactly as grandparent poster said. No need to read any of the original data blocks. Only the original parity block. Even then only if it isn't still cached in RAM which is quite possible.

The drive draws 5.31W during reading. The whole system, that is including optical drive, CPU, display and everything, draws 19W when playing an Audio CD. It uses 30W in a graphics benchmark. It's reasonable to assume that DVD or MPEG4 playback would fall somewhere in between,

And incidently: a CD-ROM does not, or should not, spin at the same speed regardless of what's being read. Audio CD players (used to?) maintain constant linear velocity, spinning the disk at different speeds depending on what point was being read. CD-ROMs typically use constant angular velocty, spinning the drive at a constant speed no matter what is read.
But when I'm streaming some media from the drive, it really should run at the lowest speed that sustains the bitrate, and not at it's maximum. The reason being that if I'm watching a video clip that would run fine at 10X speed I wouldn't want to endure the sounds of a drive at 40X speed. I used to set this manually, but I think some drives do it automatically these days.

"two RAID-5s mirrored with a hotspare is a very common configuration. be it a 9, 13, or 17 disk"

OK, so that's what you meant by RAID5+1. I have never heard the term used that way. I read your RAID5+1 as meaning something like this, which was the the most common meaning back when I was more aware of current terms and trends in the storage industry.

My argument was based on the belief that that was your meaning, so pretty much none of it applies to what you said.

RAID 10 I can see if you really need high redundancy/availability, but 5+1 is just way too slow and too disk-hungry for any practical use. (what company or person wants to buy 16 disks for every 7 disks worth of storage they get to actually use?)

Lots of different terminilogy here.. when talking about 5+1 in a RAID5 setting, it's often a short way of saying "5 data disks +1 parity disk" for each set. 5+1 is a common configuration for write intensive tasks, since parity will be a limiting a factor. For more read intensive tasks, 8+1 isn't uncommon

But to add to the confusion, you do have raid 51 too... for business critical systems for large companies, using more disks isn't that much of an issue either.

Lots of people are claiming that this is much cheaper to run than a dedicated ATX server, but they are forgetting you need to power the hard disks too. In my (limited) experience, powering desktop 7200rpm disks from USB is very dicey, so you need externally powered hard disk boxes for them.

Based on UK prices turned up in 30 seconds by Google, so probably not the cheapest to be had, but never mind.

NSLU2: £60, 5V/2A power into device

Cheap USB hard disk box:£35, 50-80VA power into the PSU brick (based on the one on my desk). I'll use 70VA, to be on the safe side.

So, outfitting one of these for two hard disks would cost around £130. Assume a 60% efficiency plugpack for the NSLU2 (which seems conservative) and total power consumption would then be around 160VA.

In comparison, my server has an Athlon 900Mhz, a couple of fans, the same two hard disks, and a 300W PSU. Let's assume it's highly loaded and actually draws around 250VA; I'll ignore power correction factor for these calculations.

At 10p/kWh, the NSLU2 costs 39p per day to run, and the server 60p. If I upgraded to the NSLU2, it would take over 3 years to get a ROI from a purely financial point of view. Unless I've gotten something wrong, in which case I'm sure some clever slashdotter will correct me in a few seconds :o)

So, on purely financial grounds, perhaps hard to justify. Still, it's nifty, it's a hell of a lot smaller than my existing server, and it would reduce the noise in this room nicely by eliminated a few fans too.

Update: hmmm, PC guide reckons it's more like 10W for a hard disk under use, suggesting the rather high sounding 50-80VA max draw are probably for 10,000rpm disks spinning up or something. Even assuming 15W to be on the safe side changes things around a lot; assume 75% efficient PSU plugpacks just to look on the bright side, and we get 20W per hard disk and 13W for the device = 13p per day. Break even is now about 9 months; not too shabby, given the other benefits.

But a hard drive needs a FAT or something similar, which is generally going to live in one spot on the disk...

One address on the disk. It isn't that difficult for the wear-leveling or related logic to remap logical CHS 0/0/1 (or whereever the FAT is stored in your particular file system) to physical CHS 0/0/1 or 37/6/9 or whatever.

This type of bad sector remapping to spare sectors (mostly invisible to the operating system) has been happening in the internal hard disk logic for some time now:
http://www.pcguide.com/ts/x/comp/hdd/errorsBadSect ors-c.html

Dvorak allready has keyboards designs for both left and right hand only. These could probably be put on the new kb easy enough.

No, these are designed for standard 102/104-key keyboards.

There's one catch: someone will have to design a keying pattern and hack up software for it.

Inspired by this guy I hacked together a software for finding optimal keyboard layouts using genetic algorithms a couple of months ago. The tool I started works similar to his, but allows the user to specify arbitrary input texts for simulating the keystrokes and to define custom weights for keystrokes so that layouts for non-pc104 keyboards and/or for people with disabilities can be evolved. It's also meant to be released as open-source.

It's not finished, but the GA and the basic code for generating a layout is in place, along with a (simple) realtime GUI display of the evolution process. What's missing is basically a way to load all parameters from a file - currently, many are hardwired.

I won't have time to continue working on it anytime soon, but if someone is interested in picking it up (it's written in Java), drop me a note: julian@sektor37.ddvorake (remove name of HCI designer).

Julian

You could always use the right-hand Dvorak layout here - windows can switch to it pretty easily (i use the standard Dvorak layout myself on MS Natural Keyboard Pro), and there are online stores where you can get 'custom' keyboards and keytop labels if you want them. (I'm not impressed with the keyboards though, which appear to have just had the keytops switched)

Because it is the PC that swapped the location of caps lock and control. No, really. Traditionally old keyboards had control next to A and sun stayed with it for most of their keyboards (to not piss off UNIX users), if you don't like it I think sun makes PC style layouts too. Actually, the earliest PC keyboards had control next to A as well.

Do read up on some of the fundamentals of RAID: Everything you need to know (and lots you don't) is probably at least mentioned in the PC Guide on RAID. Look through that. Things like hot swap and hot spares are important to understand. Finally, you should remember to check compatability. Unfortunately, I for instance have not been able to find much of anything in the way of controller cards that is compatable with OS X (except the obvious, the XServe RAID). So I have something set up on a BSD box in my server closet that I then link to, more like a storage appliance. Happily, the 3ware cards and many others are now compatable with a wide variety of *nix and BSD flavors along Windows, but do check to make sure.

Last but not least, remember this!: RAID is *not* a backup solution, but an highly redundant onsite storage system. Have another form of backups, even if it is just a RAID 1 off site, or DVD-Rs, or something. If a disaster happens (thieves, fire, nuclear destruction, John Ashcroft) on site storage won't save you.

You should go with RAID 0+1

An alternative is to do what Conner did a while back... use double the number of heads. Basically the head and actuater was duplicated, one set on the left, and one set 180 degrees away on the right. This cut the speed-induced-latency by half and doubled the thoroughput, all without going to a denser substrate or doubling the drive speed. But, the linked article gives a good reason noone has tried this since... it was costly (the doubled parts weren't cheap) and competed too closely with RAIDs.

With buffer cache you're always going to have to touch the disk at some point and that point wastes billions of CPU cycles.

Bus Mastering
PCI bus mastering

If you're going to be useing a PC (with NICs in the PCI slots) to drive the T3, you're already throwing away money. The PCI bus is unable to handle that much bandwidth.
PCI bus = 127Mbps = 15.8mbps
T3 = ~50mbps
Congradulations! You've just thrown away 35mbps! There is a reason to go with a Cisco router: the asic is able to handle that much bandwidth w/o overloading.

You are obviously a troll. The 286 does not have a pipeline, idiot. The 80386 was the first Intel processor to use pipelining, but it was severly limited, and was only 5 stages. The i486 was the first Intel processor to really use full-blown pipelining, and was able to execute some instructions in a single clock cycle.

Look here

Did this guy really just use the abbreviation "GiB?" That stands for gibitye. Quite possibly the gayest name I think anyone could have come up for the supposedly official binary power prefixes. At least it's not as bad as a mebibyte (WTF???)

"This is because Intel refused to grant them licenses to manufacture Slot-1, and then Socket-370 compatible CPUs."

Cite sources, please? I've never heard this before. Ever.

The "Super" Socket 7 may have been aging, but the K6-3 was still a kickass chip that used Socket7 to its full potential.

You need a Gateway Anykey Keyboard, it has Diagonal Arrow Keys!

You need a Gateway Anykey Keyboard, it has Diagonal Arrow Keys!

-Lucas

Bullshit. If it locks as soon as POST is complete, memtest-86 wouldn't even start since the computer would be locked. If you really want to send me the memory, email me at lucasz@hotmail.com.

I just returned two Maxtor drives that passed multiple "extended" tests with their diag utils. BOTH have entire tracks that aren't readable -- sector mark not found... they aren't there anymore.

That's how hard drives work. When parts of the disk are bad, it marks them as such and doesn't use them. Besides, didn't I say that you can't trust a disk 100% just because it passes the extended test? You were right in returning them since they were new. Drives should only develop bad sectors after lots of use.

-Lucas

Get a Gateway Anykey It has the left-hand side function keys (a set at the top, too), and is completely reprogrammable so you can put CTRL wherever you want.

It's obvious that ctrl-alt-delete was NOT intended to be triggered automatically on the original model keyboards. Definitely a two-handed operation:

http://www.pcguide.com/ref/kb/layout/std_XT83.htm

Found a site with a better picture.

Sounds a lot like the keyboard my brother use to swear by . . . an old 124-key Gateway Anykey. An image of it is in this article. Another rave about it is here.

Google for 'gateway anykey' and you'll find lots of comments about it, the programmable macros, remappable keys, etc etc. It's a pretty slick keyboard, but I never liked how it had the 8-way arrow keys instead of the standard inverted-T.

The chips that test better are marked as DX and the chips that have minor flaws are downgraded and marked SX.

This is untrue.
cf. here and here. Rumors and FUD, man. There is no way in hell a corporation would ship defective parts. The legal liability aspect would be horrifying.

It is a marketing move taken straight from intel, though -- scale back the clock speed to compete in lower-end markets.