More and more vendors are supporting Linux, especially on server type hardware. 3ware is a notable example, and I'm sure it has helped their sales immensely.
Other things like dual processor AMD board seem to have been designed with Linux in mind. When I tried to get Linux working on a $500 Supermicro dual PIII motherboard, I ran into problems left and right with BIOS issues, IDE issues, etc. Pulled that board out and put in a Tyan TigerMP and it worked flawlessly right from the beginning.
I guess I am brilliant too, since if I ran a company with many billion dollars in cash laying around, and a monopoly on desktop operating systems, I'd stick my dirty fingers into every hardware pie I could too.
Fiber cables are not that cheap. The bulk media may be approaching CAT5 prices, but the connectors on the end and the associated labor in installing end connectors and splices when something breaks is huge compared to copper.
There's more than that. Hundreds of thousands (i would guess) field techs, who mostly don't have the skills to deal with fiber.
I've met a lot of these field tech types, and it's the rare one that understands much more than continuity, cable quality issues, and dialing 211 to get ANI.
They do their jobs well for the most part, they just don't need to know much more. It will be a huge investment in people to train these techs to lay and handle and even splice fiber. I can just imagine an untrained one pushing a bundle of excess fiber into a network interface box and forcing the door shut like they do with copper.
EXT does seem to not be designed with TB+ in mind. As I said, the file system creation is very slow with any more than one inode per 4 or 8 megs, taking up to 8 hours to complete. Also, fsck takes fscking forever.
One of the things is that the IDE layer stuff doesn't matter with 3ware cards, because of the SCSI virtual interface they provide, so most of us terabyters are immune to those.
I have seen people with promise type cards have major troubles getting tons of hdX type devices to work well.
Your point is taken, there are issues a home user would not normally be exposed to, but that happens anytime a user goes off the beaten track.
If it were trivial to build such large systems, it would be hard to impress your friends with it.:)
Re:My wedding ring too, and just to be extra geeky
on
The Sexiest Metal
·
· Score: 1
If you were going to use dots and dashes anyway, why didn't you just use morse code, which is a very efficient binary coding of the alphabet, and much more likely that people would be able to read it than a 5 bit encoding scheme.
I'd say that's probably it. Back when Al was hard to come by, it was almost a precious metal. I don't recall the actual buildings, but I believe some monument has an Al top part, because it was considered a huge thing back when it was made, to have that much Al in a building. It was comprable to covering a dome in gold leaf.
You shouldn't wear any jewerly when working with machines or power tools anyway. Didn't you have any basic safety training, or at least read the manual?
Oh, the reason the benchmarks on the page are slower than the ones quoted here was because of a kernel layer issue in 2.4.17 that was fixed in 2.4.18. It was causing a bottleneck.
43Megs/sec write, 128 Megs/sec read. 1.9TB. With IDE. 1/3rd the cost of SCSI. We built two of them for extra reliability and still saved money over SCSI. Even if a whole unit fails, we are still up and running.
Not that you would really need to do so. This thing only has 3 PCI slots, and even assuming you install three non-compliant PCI devices that don't play well with others IRQ wise, you should still have plenty of room using 2/9, 10, 11.
I don't think IRQs will be a problem, unless they designed the on board IDE ports so badly that they use tons of them in a non-shared sort of way.
Last I checked, most people were having problems creating single non-raid block devices larger than 1TB, but I think that is mostly fixed, except for some overflows that cause numbers to be reported incorrectly. This means you have hardware RAID that makes a single sda or hda device that is larger than one TB. This can be avoided by breaking up the hardware RAID into multiple devices and using software raid on those. (use it to your advantage and implement RAID 10, or RAID 0 (software) over 5 (hardware))
You are correct in that there are a lot of little gotchas. I built a Linux system at work with 1.9TB of space in an single md device using the XFS stock kernel with no extra patches though, so it isn't too bad.
I was able to put EXT3 on it, but if you don't have 2 days to waste waiting for mkfs to finish, you need to tell it to only put one inode per 4 megs. This means that if most of your files will be less than 4 megs, you are going to run out of inodes before the file system is full.
XFS has no such problem, and only takes a few seconds to mkfs, but takes a little longer to mount each time.
It is definitely something that can be done at home, it's just not going to be quite as trivial as making a normal sized volume.
and here you are talking about a victory for the user over the PACs.
Hey, don't put words in my mouth. The Slashdot population may be hailing this as a victory, but I noticed the same thing, all it was, was an announcement that they had gotten only negative emails, I don't see why people automatically assumed the bill is dead either.
Unless you are advocating a bloody revolution, I don't see how your attitude will help anything.
The rest of us can continue using peaceful means to try to change things for the better. (Or maintain the status quo when politicians try to make things worse)
There could be more dire consequences in the future for the Unix camp (or not), but a company can't sacrifice the present for the distant future and stay in business.
Put another way, which would you rather, fat expensive and slow Sun boxes getting taken over by MS boxes, or by Red Hat boxes. It's like thinning the herd. In the end it will only make us stronger.
I know, the guy asked about MP3s through. It was kind of a stupid question since these are obviously optimised for bandwidth and not space. That is why I showed him the IDE RAID.
TIFF files have an optional, but built in lossless compression. You would accomplish little by compressing it more. A typical 300 DPI CMYK 5 inch by 10 inch EPS with no spot colors is going to be something like 45 megs.
It could still hit legitimate users, if they move these types of files around a lot. They should probably have "business class" service if they are going to be doing that type of thing anyway.
In 1994-5 it was $19.95 a month for about 5 hours and $2.50 an hour after that.
Don't ask how I know. I prefer to not remember the dark past when the only way to anything close to the Internet in my small town was AOL, Prodigy, and Fidonet on the BBSs.
My cable ISP had a single First Tier T1 and one peering T1 to the local college. Not everyone lives in chicago or NY. 90% of everyone lives in a small to medium sized town.
Slashdot Mirror
More and more vendors are supporting Linux, especially on server type hardware. 3ware is a notable example, and I'm sure it has helped their sales immensely.
Other things like dual processor AMD board seem to have been designed with Linux in mind. When I tried to get Linux working on a $500 Supermicro dual PIII motherboard, I ran into problems left and right with BIOS issues, IDE issues, etc. Pulled that board out and put in a Tyan TigerMP and it worked flawlessly right from the beginning.
56k modems with proper drivers don't use hardly any processor on modern systems. Sorry. I'd like for your argument to be true, but it just isn't.
That doesn't mean winmodems don't have other issues. I had tons of trouble with flaky drivers on them back when I used windows.
The guy is brilliant
Uh What?
I guess I am brilliant too, since if I ran a company with many billion dollars in cash laying around, and a monopoly on desktop operating systems, I'd stick my dirty fingers into every hardware pie I could too.
Fiber cables are not that cheap. The bulk media may be approaching CAT5 prices, but the connectors on the end and the associated labor in installing end connectors and splices when something breaks is huge compared to copper.
There's more than that. Hundreds of thousands (i would guess) field techs, who mostly don't have the skills to deal with fiber.
I've met a lot of these field tech types, and it's the rare one that understands much more than continuity, cable quality issues, and dialing 211 to get ANI.
They do their jobs well for the most part, they just don't need to know much more. It will be a huge investment in people to train these techs to lay and handle and even splice fiber. I can just imagine an untrained one pushing a bundle of excess fiber into a network interface box and forcing the door shut like they do with copper.
EXT does seem to not be designed with TB+ in mind. As I said, the file system creation is very slow with any more than one inode per 4 or 8 megs, taking up to 8 hours to complete. Also, fsck takes fscking forever.
:)
One of the things is that the IDE layer stuff doesn't matter with 3ware cards, because of the SCSI virtual interface they provide, so most of us terabyters are immune to those.
I have seen people with promise type cards have major troubles getting tons of hdX type devices to work well.
Your point is taken, there are issues a home user would not normally be exposed to, but that happens anytime a user goes off the beaten track.
If it were trivial to build such large systems, it would be hard to impress your friends with it.
If you were going to use dots and dashes anyway, why didn't you just use morse code, which is a very efficient binary coding of the alphabet, and much more likely that people would be able to read it than a 5 bit encoding scheme.
I'd say that's probably it. Back when Al was hard to come by, it was almost a precious metal. I don't recall the actual buildings, but I believe some monument has an Al top part, because it was considered a huge thing back when it was made, to have that much Al in a building. It was comprable to covering a dome in gold leaf.
You shouldn't wear any jewerly when working with machines or power tools anyway. Didn't you have any basic safety training, or at least read the manual?
Oh, the reason the benchmarks on the page are slower than the ones quoted here was because of a kernel layer issue in 2.4.17 that was fixed in 2.4.18. It was causing a bottleneck.
It really depends on your needs.
Fast, cheap, (pretty) reliable.
http://www.bedford.smythco.com/storage/
43Megs/sec write, 128 Megs/sec read. 1.9TB. With IDE. 1/3rd the cost of SCSI. We built two of them for extra reliability and still saved money over SCSI. Even if a whole unit fails, we are still up and running.
Not that you would really need to do so. This thing only has 3 PCI slots, and even assuming you install three non-compliant PCI devices that don't play well with others IRQ wise, you should still have plenty of room using 2/9, 10, 11.
I don't think IRQs will be a problem, unless they designed the on board IDE ports so badly that they use tons of them in a non-shared sort of way.
It's OK if you use XFS and software RAID.
Last I checked, most people were having problems creating single non-raid block devices larger than 1TB, but I think that is mostly fixed, except for some overflows that cause numbers to be reported incorrectly. This means you have hardware RAID that makes a single sda or hda device that is larger than one TB. This can be avoided by breaking up the hardware RAID into multiple devices and using software raid on those. (use it to your advantage and implement RAID 10, or RAID 0 (software) over 5 (hardware))
You are correct in that there are a lot of little gotchas. I built a Linux system at work with 1.9TB of space in an single md device using the XFS stock kernel with no extra patches though, so it isn't too bad.
I was able to put EXT3 on it, but if you don't have 2 days to waste waiting for mkfs to finish, you need to tell it to only put one inode per 4 megs. This means that if most of your files will be less than 4 megs, you are going to run out of inodes before the file system is full.
XFS has no such problem, and only takes a few seconds to mkfs, but takes a little longer to mount each time.
It is definitely something that can be done at home, it's just not going to be quite as trivial as making a normal sized volume.
and here you are talking about a victory for the user over the PACs.
Hey, don't put words in my mouth. The Slashdot population may be hailing this as a victory, but I noticed the same thing, all it was, was an announcement that they had gotten only negative emails, I don't see why people automatically assumed the bill is dead either.
So what are you going to do, kill them?
Unless you are advocating a bloody revolution, I don't see how your attitude will help anything.
The rest of us can continue using peaceful means to try to change things for the better. (Or maintain the status quo when politicians try to make things worse)
There was an article on here with some people running a P4 at something like 3.5Ghz using a metal cup of Liquid nitrogen as a heat sink.
Is it too hard for you to go get pppd.tar.gz and install your own version?
Also, point me to the redhat bugzilla listing you are referring to.
There could be more dire consequences in the future for the Unix camp (or not), but a company can't sacrifice the present for the distant future and stay in business.
Put another way, which would you rather, fat expensive and slow Sun boxes getting taken over by MS boxes, or by Red Hat boxes. It's like thinning the herd. In the end it will only make us stronger.
As root.
In XP, normal users will have raw socket access.
So I guess a company that has a PA system with music over it connected to a CD changer is in the same boat?
What about the many many companies that buy a consumer digital satellite system and broadcast the musak channels throughout their building?
I know, the guy asked about MP3s through. It was kind of a stupid question since these are obviously optimised for bandwidth and not space. That is why I showed him the IDE RAID.
Charter Communications is mom-and-pop?
TIFF files have an optional, but built in lossless compression. You would accomplish little by compressing it more. A typical 300 DPI CMYK 5 inch by 10 inch EPS with no spot colors is going to be something like 45 megs.
It could still hit legitimate users, if they move these types of files around a lot. They should probably have "business class" service if they are going to be doing that type of thing anyway.
In 1994-5 it was $19.95 a month for about 5 hours and $2.50 an hour after that.
Don't ask how I know. I prefer to not remember the dark past when the only way to anything close to the Internet in my small town was AOL, Prodigy, and Fidonet on the BBSs.
My cable ISP had a single First Tier T1 and one peering T1 to the local college. Not everyone lives in chicago or NY. 90% of everyone lives in a small to medium sized town.