I skimmed through the article and towards the end there is a graph of the throughput of this drive. It seems to max out at around 8 MB/sec. It is probably the slowest 4200 RPM hard drive you will be likely to find since the platters are so small. A 3.5 inch drive and a 1 inch drive using the same technology will both have save bit density.
I am sure the platters are not quite these sizes, but... The outter most track on a 3.5 inch platter is nearly 11 inches long. The outter most track on a 1 inch platter is 3.14 inches long. So the 3.5 inch drive will have over three times as much data pass under the read/write head each revolution. I wouldn't be at all surprised if the inner most track on the 3.5 inch platter is still bigger than the outtermost track on this 1 inch drive.
Seriously, this reminds me of a LiveCD of some kind.
That is precisely what it is... A live cd on read/write media.
I love the idea and think that some people will find the H2 invaluable, but to me it just doesn't seem very practical.
I think the biggest problem is the lack of machines that can boot USB drives, not to mention the machines where you won't have access to change the setting. I would imagine you'd have better luck carrying a business card cd with a bootloader, though.
I have thought about putting a bootable Linux image of some kind on my flash drive. I have tons of space free, but I don't own a machine that can boot it:p.
If client-based firewalls ar eso great, then why doesn't IBM and Ford and the Fortune 500 have all their PCs connected directly to the web and install personal firewalls? Answer?
When were "client-based firewalls" mentioned? He was talking about using a real firewall to filter traffic going to/from machines with public IP addresses. Since when do you need NAT to do this?
Having direct connections to the web for each terminal is more expensive than having them all behind the NAT
First, "the web" does not equal "the Internet." With IPv4 it is both wasteful and expensive to assign a public IP address to every workstation in an organization. With IPv6 the cost goes away and can potentially solve some connectivity issues. You will still need all the same firewalls you do today, they just don't need to NAT anymore.
These companies will not lose any security, because the firewall should be set to deny traffic by default (just like it should be now).
You can't trust your employees to keep a secure environment
No, that is why we are all talking about keeping your firewalls in place. Nobody wants to replace tham with software installed on each workstation.
Thus, corperations have no need or desitre to have all their terminals directly connected to the internet.
If a corporate environment switched over to IPv6 and gave all their workstations addresses in the public space they don't have to be any more "directly onnected to the internet" than they are today. They would still filter traffic at all the entry points to their network.
Thus, they don't need IPv6. Thus, the vast majority of computers *in the world* (business use still trumps home use by a factor of like 5 to 1) do not need it.
Businesses get at least two advantages to using the public address space. First, your firewall doesn't need to do anything strange to NAT protocols that are very NAT unfriendly. They can just allow particular addresses to get out with those protocols. Besides, if you still REALLY want to NAT with IPv6 nobody is stopping you:).
It is also a huge advantage when there is a merger. It is not much fun trying to merge two networks that already have overlapping addresses. If everyone can get large blocks of public addresses there would never be overlap.
There isn't currently enough incentive for anyone in the US to want to make the switch. These aren't compelling enough reasons to convert a network. If I had to guess, I would say the US won't hop on the bandwagon until a large enough percentage of the rest of the world actually starts converting.
agee with Mr Coward there... laptop drives are not my choice... good power management is.
If all you wanted to know about was how to power down the hard drives, Google would have given you the answer very quickly. Just use hdparm to set the spin down time, it has done a great job of keeping my laptop 100% silent the last 5 minutes.
You seem to have gotten plenty of good answers to this part of your question:
advice on processor/mainboard combos, low power HDDs and a distro with the best power management
Laptop hard drives are low power, and are a good answer to the question you asked. If you want lower power you will be paying a premium, it is as simple as that. I did not do exhaustive research, but a quick google search turned up these links:
Both drives are 7200 RPM. The desktop drive requires 10.6 W for read/write and 1.3 W when sleeping. The laptop drive needs 2 W for read/write and 0.1 W in standby (my math says it needs 10 times more to sleep than standby, I must be reading something wrong). I would imagine a good 4200 RPM laptop drive beats a 5400 RPM desktop drive by a slightly better margin.
The largest laptop hard drive I can find on pricewatch is 120 GB. Four of those would likely require 20% less power when running, and 10 times less power in standby as a single 500 GB desktop drive. I think the biggest advantage in your case is the standby power consumption. You can almost leave a single laptop hard drive running for every desktop hard drive that is asleep.
If I wanted low power and redundancy... I would probably JBOD a bunch of mirrored pairs of laptop drives. Since each pair would be tacked onto the end of the previous pair the odds should be in your favor that only one pair would be on at any given time (assuming you are only working with one video file at a time). If you set your spin down times to a minute or so, it would work pretty well.
120 gig laptop drives are about 200 bucks, 500 GB desktop drives seem to be about 360 bucks. That puts laptop drive only a little more than twice as expensive as desktop drives. In my JBOD setup it is likely that you would average 1/5th the power when running, and less than 1/10th the power at idle if you went with the laptop drives. If you really want to save power that doesn't sound expensive to me.
Here is a question for you. Wouldn't you save a good bit of juice by somehow running DC straight into the PC, instead of going from DC-AC and then back to DC in the power supply? I recall this making a pretty big difference in efficiency when I was thinking about putting a PC in one of my cars. I am entirely ignorant of what you have to work with in a solar powered home, so this may not be at all easy.
Guess what? That wicked dual-core CPU actually runs games slower than its single core cousin.
Is this actually a true statement? I can't do any current testing since I don't have a reasonable 3D card in my machine, but I remember testing Quake 3 on my old dual Celeron machine with a TNT2 card. top showed Quake was using 95% or more of one CPU, and the X server was using 30% or more of the other CPU.
I don't expect the numbers to be the same today, but shouldn't there be at least some slight increase in speed if the GUI is running in a separate process? I am not saying that the increase would justify the price, but I haven't run a single CPU desktop in something like six years. I am not about to start now:).
The big question is, how many of the remaining 99.5% are using Linux, and if so, why did they switch.
My parents computers and my girlfriend's father's computer are running Debian or Ubuntu. When I originally set up my parents computers I installed Windows 2000 on them. This was about 4 years ago. Since I was 1500 miles away, I thought it would be more useful to have a system that other people nearby could easily support. Boy, was that a mistake.
They mostly surf the web and take pictures. I have has much fewer calls from them with problems since I switched them, and they are happier with the machines now. I have no idea what percentage of your 99.5% people like my parents fall into, though. I do assume it is a fairly large number of people.:)
Also: try using linux as a desktop for 2 years and see if it doesn't start slowing down when you install a new program once every week or two, new hardware every 6 months, and new graphics drivers and security patches once a month.
I installed a copy of Debian on my old desktop machine sometime before Woody went stable. I am not sure exactly how long ago that was, I will just say it was more than 2 years ago. Since then I have upgraded every piece of hardware in the case (it is still a fairly nice case). I have been tracking unstable for a very long time now (which, of course, has caused a few minor irritations).
When I got my current laptop almost 2 years ago, I copied the current installation on to it, and tweaked some configuration. It has been running just fine ever since.
What sort of problem am I supposed to be noticing, exactly?
Dude, Windows Registry == Linux/etc. Config files have to go somewhere. Yes, you can get to them easier in linux, and yes I prefer it that way, but if programs are coded properly, there shouldn't be a need to go poking around in the windows registry, ever. Everything should be adjustable from inside the program, or shouldn't need user adjusting.
Having your configuration in text files has quite a few advantages over the windows registry. It is easier to back up, and much easier to restore an individual programs configuration. It is also much easier to copy configuration from one machine to another.
I honestly can't believe you weren't modded down into oblivion and bitchslapped by everybody for using the word "stealing". Bravo, sir. I salute you.
It is OK to refer to copyright infringement as stealing as long as you are not accusing someone else. If you are accusing someone else it is quite rude to exaggerate their crime. Since he is speaking negatively about himself it is perfectly alright to claim that the crime worse than it is.
For power I was refering to a Rack ups or the like to sustain power, not provide it so much.
You are going to have pretty similar UPS requirements for a home grown solution or a big commercial SAN. It won't have much impact on the overall price.
If you want any sort of operational reliability you'll want Raid 51/15 or something similar so Double your minimum estimate
I suppose it depends on your level of paranoia, and how much you want to spend. If I wanted more redundancy than RAID 5 I would probably go with RAID 10 and eliminate the write penalty you get from having to compute the parity. That would only require 50% more hardware, not double. Read performance would suffer a bit, but writes would be drastically better. That might not matter if the network turns out to be the bottleneck.
Don't forget power solutions, rack costs and power utilization in your estimate either.
Racks are cheap enough they would easily be covered in my quick cost estimate (I inflated my math by nearly $2000 per 4u server). A SAN from EMC would also use quite a bit of power, since the biggest drain in either case will be the drives. The SAN will have the advantage that it doesn't need a server for every 16 drives. Since you won't need top of the line machines for this, I would guess that the extra power requirements would quite a bit less than 200 watts per server, or 2000 watts per 40 - 60 TB of usable space (80 TB raw). How much more than 100k would I have to pay for a low end 80 TB SAN?
My cheap soution would be nearly as dense as the old EMC Clariion I used to work with. The 4U chasis I am thinking of was almost nothing but hard drives on the front side.
I am not saying a solution like this fits all, or even most, situations. I can recall a number of situations in the past where something like this would have been a very good fit.
Yeah, right. You *may* want to triple this figure for adequate hardware, though.
He may also want to do it "right" as well:p. 500 GB SATA drives seem to be around 400 bucks a piece. I have seen 4u cases with 16 hot swap SATA bays for about 1200 bucks. A pair of 8 port 3ware SATA controllers would be about 700 bucks, and figure 1000 bucks for motherboard/cpu/nic/etc. 16 drives will set you back 6400 bucks. Assume you go RAID5 with a hot spare on each controller, that works out to 6 TB for well under $10,000.
I don't think any SAN can compete with that price. I also don't think this machine can compete with a SAN in any scenario where you should actually be using a SAN:). However, 10 of these machines could be squeezed into a single rack for a total of 60 TB for $100,000. I have no idea what someone would use a rack of these machines for... Online backups or something, maybe?
You obviously underestimate the effort it takes to create a system like this, much less the effort to make it perform reasonably and maintain it.
The problem only becomes difficult when you want this to appear as a single volume. If you need a reasonably priced 25 TB in volumes of less than about 3 TB each it becomes quite easy.
If your business really plans with and depends on 1PB of storage but at the same time a 3mio investment makes you pee
then there's something seriously wrong with your business plan.
Maybe their business plan involves selling hardware to store up to 1 PB on a single volume?:p
And, last but not least, if you have to "ask slashdot" about it then you're most certainly not in a position to pull it off either way.
I absolutely agree with you. I don't think this fellow has any idea what he is looking at getting himself into... He wants to make it to 25 TB with 1-2 GB per node and scale up to 40x as much from there, and he thinks it will survive with no redundancy.:)
The point is, though, that the seal is used to indicate official documents, etc.
Using an easily reproduced seal is a ridiculous way of assuring that a document came from a particular organization. If they want to be sure that the public can verify that a document did indeed come from the White House they should digitally sign the documents.
I wouldn't want to trust important data in possibly harsh conditions to magnetic or electronic media. If you actually had to swim safety in New Orleans, what would salt water do to a flash drive? I have heard all the stories about what people accidentally running assorted flash memory through the washer/dryer. That doesn't mean I am going to trust it in a crisis:).
I am going to also assume that you haven't worn out any flash based media due to too many writes?
CDs, otoh, need just a bad scratch and they are done for.
Scratches are easily buffed out. You are probably correct that a CDR wouldn't be safe enough, because the data surface is generally just a sticker on the back of the disc. Writable DVDs on the other hand have the data surface sealed in plastic.
I am also going to imagine that you would at least have this valuable data disc in a jewel case to prevent scratches. I doubt salt water would do anything to the plastic, and if there is anything in the water that would disolve the plastic it would do the same to a flash drive.
Not to mention that for me, CD-Rs showed the habbit of becoming sponaniously unreadable after a few months/years quite often, which is something you _dont_ want for an emergency storage.
There are plenty of claims as to how long a CDR will last. I still have one or two discs that were burned right after I got my first 2x CD burner (back when they were about $350, I assume that is approaching 10 years). They are still readable. Every CDR I remember encountering problems with was either scratched, or just wouldn't read in a particular drive.
I would hope that if you were using a CDR or DVD to store this information that you woulnd't just let it sit in the closet for 10 years. I would hope you would be keeping it up to date every year or two. That would very easily fix the bitrot issue.
I really think the better option is to just keep a backup of this data as far away from where you live as is possible. If there is a disaster that impacts the east and west coast at the same time I doubt anyone will care about credit card numbers and birth certificates:p.
they suggest making a list of all of things like Social Security and credit card numbers, scanning birth certificates, marriage license and tax returns, and saving it all on a USB flash drive.
Why not just use a CD (full size, or 180 meg)? They are cheaper and more durable than a flash drive. Before I had my new, larger, flash drive I used to carry a 50 meg business card CD in my wallet. It would have to be replaced every 3-6 months from being repeatedly sat on:). I would imagine they would hold up better outside of the pocket, though:).
Since this would be a complete identity kit, encryption is of utmost importance. What's the best solution? A flash drive that claims to encrypt or a platform-independent, self-extracting, encrypted file on a regular drive?
I wouldn't use the software that comes with the drive. If I were doing this I would use GNU Privacy Guard. You should probably store the key in a safe location far away from home, and preferably with a strong passphrase.
Any suggestions for sturdy drives?
I currently have a PQI I-Stick. I have only had it about a year so far and I haven't doen anything stupid with it yet. It mostly just sits in my wallet in its little wallet case. I very much prefer keeping my flash drive in my wallet as opposed to my keychain. I also like that the little wallet insert will hold two drive. The only thing I dislike is that the wallet holder is so much thicker than the drive.
What other data would you put on this piece of "contingency hardware",
I have all of my revision control repositories mirrored to my flash drive and also any documentation or notes that I write. That is basically everything that I created myself and would have to do work to replace.
how would you protect the drive itself in case you did have to "swim for it"?
I would probably make sure the data was out of town before I was. Most of this data either doesn't change often (credit card numbers), or it never changes (SSN, birth certificates). Encrypt it, put it on some media of some sort, and send it out of town. Most people probably have friends or family living out of town that they can trust, send it to them. If this is not an option for you, you can probably get a box at a bank out of town I suppose...
remember usage isn't constant though and charges are usually by 95th percentile or capped bandwidth so you nearly always will have bandwidth during off hours that you are paying for but not using.
How about if I make some assumptions since I have no real data?:)
I have no bandwidth pricing available except my own personal provider, and I am paying for disk and cpu, but it is a reasonable place to start. Lets say they can use 70% of my 512mbit for 8 hours per day and for the other 16 hours they use 15%. Unless I am much more tired than I think I am at this late hour, that would be using one third of my total payed for bandwidth.
That triples my cost per song from 0.005 cents to 0.015 cents. Even if Apple pays 100 times more money for bandwidth than I do they are still paying 1.5 cents per song. If they are paying $1500 a month for one third of a T1 they are simply out of their minds:).
Here is a dirty little secret about 95-99% of web hosts. Any time they say 'unmetered' or 'unlimited', they are totally lying.
You might want to shop for a more honest host:p. My current plan is one of their old plans, and it is 30 GB transfer per month. So, I suppose I fibbed just a bit. The new plans have no monthly cap, but are limited to 512mbit for the low end plan. I am planning to switch to one of the new plans, I just have not decided which one yet:p.
I dealt with a bigish webhost a few months back that totally was devious.
I suppose technically this isn't a web host. I have a slice of a Linux box running Virtuozzo. From my point of view, it is similar to User Mode Linux. Mine is running Debian Sarge, although you can get just about whatever you like.
Before you ever get attached to a web host, max out your account for a month. If they offer "5GB of disk space", make a 5GB random text file and upload it. See what happens. If they offer 100GB of transfer, try it out.
My math makes me doubt that my provider is overselling disk space. I am guaranteed a minimum amount of CPU, and I did some math before I signed up. If you bought enough of my small machines to get 100% of the cpu on a single box you came out just shy of the common size SCSI drive. I cannot verify this with the currently listed plans, however, because they list the CPU as "fair share" now.
I am not sure exactly how Virtuozzo handles disk space. UML uses disk images... So if you have a 10 gig UML disk, there is a 10 gig file sitting somewhere for you (my provider also offers UML machines). I have a feeling that Virtuozzo does not use disk images, although I have not cared to research whether that is true or not. Anyway, I am currently using about one third of my available disk, but I have spiked up to near 50%.
I am willing to guarantee that you get shut down on probably 9 out of 10 times.
This is why you should do some research before you buy. I have been quite happy with the service I have gotten from my current provider. I have only been with them for about a year now, though. Before that I had a plain old web hosting account for about 5 years or so... When I got it it was a good deal, although the machine was unavailable for minutes at a time fairly regularly towards the end.
My hosted VM currently has an uptime of 172 days. That reboot was a scheduled outage for an upgrade of Virtuozzo. I believe that is the only time the machine has actually gone down since I've had it.
I think I have drifted rather far from the topic at hand. I was only attempting to point out that if I can get enough bandwidth to transfer 32,000 5 meg songs per month at this price I doubt Apple could be doing much worse. You could transfer 3 times as much data over a T1, and that would still be under a penny per song (assuming of course that the T1 were saturated).
Bandwidth is the cheapest piece of their puzzle. They have to pay for machines, hard drives, and facilities (whether they do it in house or they colocate).
Bandwidth is nearly free, yes. It probably costs Apple some fraction of 1% of 99 cents to cover the bandwidth used to download a single song. The $15 dollar a month plan at my web host offers enough bandwidth to serve over 32,000 5 MB songs (unmetered 512mb, about 160GB per month). That works out to 0.005 cents per song.
I would hope Apple is buying enough bandwidth to get a better discount than I do...:p
I love AMD to death, and have for years, but they really didn't have anything that could stand up to Intel's best until the K7/Athlons got to market.
My old AMD 386DX 40Mhz would be very disappointed to hear you say that:).
I was still in high school when I got that machine and at the time it seemed very important to have a faster machine than my friends:p.
That machine was cheaper and almost exactly as fast as one of my friend's 486SX 25Mhz machine. In some benchmarks it was a few percent faster, in others a few percent slower. In the real world you couldn't tell the difference. Thank god we had no use for floating point at the time:).
And Cyrix wasn't even on the same page.
They were for a while. Their integer unit was quite a bit more efficient than Intel. I remember that my Cyrix P200 ran with an FSB of 75Mhz and a core speed of 150Mhz. It was easily as fast as an Intel P200, except for floating point. It would have been lucky to keep up with an Intel P100 or P133 in that category.
Unfortunately for Cyrix, that was when everyone wanted to play Quake and all the latest 3D games were no longer cheater 3D like Doom.
I still have one of these, running at 450. That is to say, it ran at 450 the last time it was powered up:p.
But they're different parts, what the hell are you babbling about, you say?
Yes, they absolutely are different parts.
In fact, a lot of the Malaysian 300As were perfectly functional PII 450s that had a bunch of their cache burned out and their clock dropped to 66MHz instead of 100, as mine was, and as a result would run perfectly fine when clocked up to 450 MHz without even needing to change the heatsink and fan from the stock one.
No, sorry. A PII (Deschutes core) 450 had 512k of half speed, off chip cache. The 300A (Mendocino core)had 128k of on core full processor speed cache. Since the cache was twice as fast the Celeron ran as fast (or faster) than an equally clocked PII.
If the 300A and a PII 450 were the same chip with less cache it would not have performed as well.
So if Intel could afford to sell a chip that they had to do MORE work on (burning out the cache) for $500 less,
I believe you are thinking of the later cores... IIRC the Coppermine Celerons were Coppermine Pentium IIIs with a slower FSB and less cache. In all likelyhood many Coppermine Celerons were Pentiums with some faulty cache. Just like the early 486SX chips were 486DX chips with a faulty FPU.
It makes much more financial sense to "burn out" the faulty cache on some chips instead of throwing them away.
Bottom line. Intel is not your friend. They're a business. If they think they can get a $300 profit on a part even after their R&D costs, they'll do it in a heartbeat.
Icecast on Debian Linux already has great performance for streaming, if not "realtime" interrupts
I have no idea why you would need anything remotely like 5 microsecond latency for "streaming." When you stream media you have a buffer at the client end. If you are a bit late with a single packet it shouldn't make any difference.
But I do wonder about loads people have seen with standard datacenter server HW. For example, how many 128Kbps streams can a P4/4.3GHz/128K-cache/512MB-RAM Icecast2 server stream from the local filesystem (preencoded to 128Kbps MP3) to a 100Mbps ethernet, if that's all it's running under the Debian v2.6.10-5-686 kernel/OS?
I would hope this machine would be more than up to the task of saturating a 100Mb link. You need very little CPU and very little disk throughput. I wouldn't be surprised to find out that a 486 with a PCI 100 megabit NIC could do the job just as well.
I want my car to authenticate drivers just like SSH does. Everyone could carry a standard fob with their private key on it, and maybe some other people's public keys. If I want to let a friend borrow my car I just load his public key on my car and tell my car that he is authorized to drive.
Take it a step furthur. You can authorize a valet to open the doors and drive the car with a low rev limiter so he cannot go tooling around town.
The best part about this is that if I lose my fob I only have to remove my public key from the car.
You can take this as far as you like. You can make this same system work for locks in your home and office. I the system is remotely accessible you can do neat things with it... You could temporarily authorize people while you aren't home.
The people who sell copies of keys probably won't like it, though:).
Actually, TCP does a pretty bad job on the error correcting side- packets aren't buffered properly (in that, if a machine suddenly becomes unreachable for a couple of days, TCP will simply never deliver the packets. Zmodem, on the other hand, will resume upon reconnection, discarding spare packets and assembling it's file correctly. Just try to do THAT with a TCP protocol like FTP).
The whole point is that you are running zmodem over ssh which is a protocol that runs over TCP. You are adding the overhead of zmodem's error correction when you already have perfectly good error correction. If you want a more robust way to copy files than sftp and scp (I never mentioned ftp) you should really use rsync instead of zmodem. rsync will have no problem resuming downloads, and it will be able to syncronize files that changed.
Zmodem rules. I use it all the time. SecureCRT (an ssh client for Windows) has support for it and it's great for logging in somewhere and quickly sending files to/from your local machine, no filesharing necessary.
So you add the overhead of an error checking and correcting protocol (zmodem) over top of a protocol that already checks for and corrects errors (tcp)? Why not just use sftp/scp and skip installing an extra piece of software on your server?
Just to plug SecureCRT, so far it's the fastest, most convienent, best SSH client I've ever used on any platform, which is sad considering it's a Windows app.
What are the advantages over Putty on Windows? What are the advantages over OpenSSH?
Slashdot Mirror
4400 RPM Hard Drive... 4-5 minute boot time?
I skimmed through the article and towards the end there is a graph of the throughput of this drive. It seems to max out at around 8 MB/sec. It is probably the slowest 4200 RPM hard drive you will be likely to find since the platters are so small. A 3.5 inch drive and a 1 inch drive using the same technology will both have save bit density.
I am sure the platters are not quite these sizes, but... The outter most track on a 3.5 inch platter is nearly 11 inches long. The outter most track on a 1 inch platter is 3.14 inches long. So the 3.5 inch drive will have over three times as much data pass under the read/write head each revolution. I wouldn't be at all surprised if the inner most track on the 3.5 inch platter is still bigger than the outtermost track on this 1 inch drive.
Seriously, this reminds me of a LiveCD of some kind.
That is precisely what it is... A live cd on read/write media.
I love the idea and think that some people will find the H2 invaluable, but to me it just doesn't seem very practical.
I think the biggest problem is the lack of machines that can boot USB drives, not to mention the machines where you won't have access to change the setting. I would imagine you'd have better luck carrying a business card cd with a bootloader, though.
I have thought about putting a bootable Linux image of some kind on my flash drive. I have tons of space free, but I don't own a machine that can boot it :p.
If client-based firewalls ar eso great, then why doesn't IBM and Ford and the Fortune 500 have all their PCs connected directly to the web and install personal firewalls? Answer?
When were "client-based firewalls" mentioned? He was talking about using a real firewall to filter traffic going to/from machines with public IP addresses. Since when do you need NAT to do this?
Having direct connections to the web for each terminal is more expensive than having them all behind the NAT
First, "the web" does not equal "the Internet." With IPv4 it is both wasteful and expensive to assign a public IP address to every workstation in an organization. With IPv6 the cost goes away and can potentially solve some connectivity issues. You will still need all the same firewalls you do today, they just don't need to NAT anymore.
These companies will not lose any security, because the firewall should be set to deny traffic by default (just like it should be now).
You can't trust your employees to keep a secure environment
No, that is why we are all talking about keeping your firewalls in place. Nobody wants to replace tham with software installed on each workstation.
Thus, corperations have no need or desitre to have all their terminals directly connected to the internet.
If a corporate environment switched over to IPv6 and gave all their workstations addresses in the public space they don't have to be any more "directly onnected to the internet" than they are today. They would still filter traffic at all the entry points to their network.
Thus, they don't need IPv6. Thus, the vast majority of computers *in the world* (business use still trumps home use by a factor of like 5 to 1) do not need it.
Businesses get at least two advantages to using the public address space. First, your firewall doesn't need to do anything strange to NAT protocols that are very NAT unfriendly. They can just allow particular addresses to get out with those protocols. Besides, if you still REALLY want to NAT with IPv6 nobody is stopping you :).
It is also a huge advantage when there is a merger. It is not much fun trying to merge two networks that already have overlapping addresses. If everyone can get large blocks of public addresses there would never be overlap.
There isn't currently enough incentive for anyone in the US to want to make the switch. These aren't compelling enough reasons to convert a network. If I had to guess, I would say the US won't hop on the bandwagon until a large enough percentage of the rest of the world actually starts converting.
agee with Mr Coward there... laptop drives are not my choice... good power management is.
If all you wanted to know about was how to power down the hard drives, Google would have given you the answer very quickly. Just use hdparm to set the spin down time, it has done a great job of keeping my laptop 100% silent the last 5 minutes.
You seem to have gotten plenty of good answers to this part of your question:
advice on processor/mainboard combos, low power HDDs and a distro with the best power management
Laptop hard drives are low power, and are a good answer to the question you asked. If you want lower power you will be paying a premium, it is as simple as that. I did not do exhaustive research, but a quick google search turned up these links:
Desktop HD specs
Laptop HD specs
Both drives are 7200 RPM. The desktop drive requires 10.6 W for read/write and 1.3 W when sleeping. The laptop drive needs 2 W for read/write and 0.1 W in standby (my math says it needs 10 times more to sleep than standby, I must be reading something wrong). I would imagine a good 4200 RPM laptop drive beats a 5400 RPM desktop drive by a slightly better margin.
The largest laptop hard drive I can find on pricewatch is 120 GB. Four of those would likely require 20% less power when running, and 10 times less power in standby as a single 500 GB desktop drive. I think the biggest advantage in your case is the standby power consumption. You can almost leave a single laptop hard drive running for every desktop hard drive that is asleep.
If I wanted low power and redundancy... I would probably JBOD a bunch of mirrored pairs of laptop drives. Since each pair would be tacked onto the end of the previous pair the odds should be in your favor that only one pair would be on at any given time (assuming you are only working with one video file at a time). If you set your spin down times to a minute or so, it would work pretty well.
120 gig laptop drives are about 200 bucks, 500 GB desktop drives seem to be about 360 bucks. That puts laptop drive only a little more than twice as expensive as desktop drives. In my JBOD setup it is likely that you would average 1/5th the power when running, and less than 1/10th the power at idle if you went with the laptop drives. If you really want to save power that doesn't sound expensive to me.
Here is a question for you. Wouldn't you save a good bit of juice by somehow running DC straight into the PC, instead of going from DC-AC and then back to DC in the power supply? I recall this making a pretty big difference in efficiency when I was thinking about putting a PC in one of my cars. I am entirely ignorant of what you have to work with in a solar powered home, so this may not be at all easy.
Guess what? That wicked dual-core CPU actually runs games slower than its single core cousin.
Is this actually a true statement? I can't do any current testing since I don't have a reasonable 3D card in my machine, but I remember testing Quake 3 on my old dual Celeron machine with a TNT2 card. top showed Quake was using 95% or more of one CPU, and the X server was using 30% or more of the other CPU.
I don't expect the numbers to be the same today, but shouldn't there be at least some slight increase in speed if the GUI is running in a separate process? I am not saying that the increase would justify the price, but I haven't run a single CPU desktop in something like six years. I am not about to start now :).
The big question is, how many of the remaining 99.5% are using Linux, and if so, why did they switch.
My parents computers and my girlfriend's father's computer are running Debian or Ubuntu. When I originally set up my parents computers I installed Windows 2000 on them. This was about 4 years ago. Since I was 1500 miles away, I thought it would be more useful to have a system that other people nearby could easily support. Boy, was that a mistake.
They mostly surf the web and take pictures. I have has much fewer calls from them with problems since I switched them, and they are happier with the machines now. I have no idea what percentage of your 99.5% people like my parents fall into, though. I do assume it is a fairly large number of people. :)
Also: try using linux as a desktop for 2 years and see if it doesn't start slowing down when you install a new program once every week or two, new hardware every 6 months, and new graphics drivers and security patches once a month.
I installed a copy of Debian on my old desktop machine sometime before Woody went stable. I am not sure exactly how long ago that was, I will just say it was more than 2 years ago. Since then I have upgraded every piece of hardware in the case (it is still a fairly nice case). I have been tracking unstable for a very long time now (which, of course, has caused a few minor irritations).
When I got my current laptop almost 2 years ago, I copied the current installation on to it, and tweaked some configuration. It has been running just fine ever since.
What sort of problem am I supposed to be noticing, exactly?
Dude, Windows Registry == Linux /etc. Config files have to go somewhere. Yes, you can get to them easier in linux, and yes I prefer it that way, but if programs are coded properly, there shouldn't be a need to go poking around in the windows registry, ever. Everything should be adjustable from inside the program, or shouldn't need user adjusting.
Having your configuration in text files has quite a few advantages over the windows registry. It is easier to back up, and much easier to restore an individual programs configuration. It is also much easier to copy configuration from one machine to another.
I honestly can't believe you weren't modded down into oblivion and bitchslapped by everybody for using the word "stealing". Bravo, sir. I salute you.
It is OK to refer to copyright infringement as stealing as long as you are not accusing someone else. If you are accusing someone else it is quite rude to exaggerate their crime. Since he is speaking negatively about himself it is perfectly alright to claim that the crime worse than it is.
For power I was refering to a Rack ups or the like to sustain power, not provide it so much.
You are going to have pretty similar UPS requirements for a home grown solution or a big commercial SAN. It won't have much impact on the overall price.
If you want any sort of operational reliability you'll want Raid 51/15 or something similar so Double your minimum estimate
I suppose it depends on your level of paranoia, and how much you want to spend. If I wanted more redundancy than RAID 5 I would probably go with RAID 10 and eliminate the write penalty you get from having to compute the parity. That would only require 50% more hardware, not double. Read performance would suffer a bit, but writes would be drastically better. That might not matter if the network turns out to be the bottleneck.
Don't forget power solutions, rack costs and power utilization in your estimate either.
Racks are cheap enough they would easily be covered in my quick cost estimate (I inflated my math by nearly $2000 per 4u server). A SAN from EMC would also use quite a bit of power, since the biggest drain in either case will be the drives. The SAN will have the advantage that it doesn't need a server for every 16 drives. Since you won't need top of the line machines for this, I would guess that the extra power requirements would quite a bit less than 200 watts per server, or 2000 watts per 40 - 60 TB of usable space (80 TB raw). How much more than 100k would I have to pay for a low end 80 TB SAN?
My cheap soution would be nearly as dense as the old EMC Clariion I used to work with. The 4U chasis I am thinking of was almost nothing but hard drives on the front side.
I am not saying a solution like this fits all, or even most, situations. I can recall a number of situations in the past where something like this would have been a very good fit.
Yeah, right. You *may* want to triple this figure for adequate hardware, though.
He may also want to do it "right" as well :p. 500 GB SATA drives seem to be around 400 bucks a piece. I have seen 4u cases with 16 hot swap SATA bays for about 1200 bucks. A pair of 8 port 3ware SATA controllers would be about 700 bucks, and figure 1000 bucks for motherboard/cpu/nic/etc. 16 drives will set you back 6400 bucks. Assume you go RAID5 with a hot spare on each controller, that works out to 6 TB for well under $10,000.
I don't think any SAN can compete with that price. I also don't think this machine can compete with a SAN in any scenario where you should actually be using a SAN :). However, 10 of these machines could be squeezed into a single rack for a total of 60 TB for $100,000. I have no idea what someone would use a rack of these machines for... Online backups or something, maybe?
You obviously underestimate the effort it takes to create a system like this, much less the effort to make it perform reasonably and maintain it.
The problem only becomes difficult when you want this to appear as a single volume. If you need a reasonably priced 25 TB in volumes of less than about 3 TB each it becomes quite easy.
If your business really plans with and depends on 1PB of storage but at the same time a 3mio investment makes you pee then there's something seriously wrong with your business plan.
Maybe their business plan involves selling hardware to store up to 1 PB on a single volume? :p
And, last but not least, if you have to "ask slashdot" about it then you're most certainly not in a position to pull it off either way.
I absolutely agree with you. I don't think this fellow has any idea what he is looking at getting himself into... He wants to make it to 25 TB with 1-2 GB per node and scale up to 40x as much from there, and he thinks it will survive with no redundancy. :)
The point is, though, that the seal is used to indicate official documents, etc.
Using an easily reproduced seal is a ridiculous way of assuring that a document came from a particular organization. If they want to be sure that the public can verify that a document did indeed come from the White House they should digitally sign the documents.
Don't be put off by the 666 tattooed on his forehead and the left leg
Yeah, it just means his is readable and writable by himself, his group, and others :).
Flash is nearly unkillable.
I wouldn't want to trust important data in possibly harsh conditions to magnetic or electronic media. If you actually had to swim safety in New Orleans, what would salt water do to a flash drive? I have heard all the stories about what people accidentally running assorted flash memory through the washer/dryer. That doesn't mean I am going to trust it in a crisis :).
I am going to also assume that you haven't worn out any flash based media due to too many writes?
CDs, otoh, need just a bad scratch and they are done for.
Scratches are easily buffed out. You are probably correct that a CDR wouldn't be safe enough, because the data surface is generally just a sticker on the back of the disc. Writable DVDs on the other hand have the data surface sealed in plastic.
I am also going to imagine that you would at least have this valuable data disc in a jewel case to prevent scratches. I doubt salt water would do anything to the plastic, and if there is anything in the water that would disolve the plastic it would do the same to a flash drive.
Not to mention that for me, CD-Rs showed the habbit of becoming sponaniously unreadable after a few months/years quite often, which is something you _dont_ want for an emergency storage.
There are plenty of claims as to how long a CDR will last. I still have one or two discs that were burned right after I got my first 2x CD burner (back when they were about $350, I assume that is approaching 10 years). They are still readable. Every CDR I remember encountering problems with was either scratched, or just wouldn't read in a particular drive.
I would hope that if you were using a CDR or DVD to store this information that you woulnd't just let it sit in the closet for 10 years. I would hope you would be keeping it up to date every year or two. That would very easily fix the bitrot issue.
I really think the better option is to just keep a backup of this data as far away from where you live as is possible. If there is a disaster that impacts the east and west coast at the same time I doubt anyone will care about credit card numbers and birth certificates :p.
they suggest making a list of all of things like Social Security and credit card numbers, scanning birth certificates, marriage license and tax returns, and saving it all on a USB flash drive.
Why not just use a CD (full size, or 180 meg)? They are cheaper and more durable than a flash drive. Before I had my new, larger, flash drive I used to carry a 50 meg business card CD in my wallet. It would have to be replaced every 3-6 months from being repeatedly sat on :). I would imagine they would hold up better outside of the pocket, though :).
Since this would be a complete identity kit, encryption is of utmost importance. What's the best solution? A flash drive that claims to encrypt or a platform-independent, self-extracting, encrypted file on a regular drive?
I wouldn't use the software that comes with the drive. If I were doing this I would use GNU Privacy Guard. You should probably store the key in a safe location far away from home, and preferably with a strong passphrase.
Any suggestions for sturdy drives?
I currently have a PQI I-Stick. I have only had it about a year so far and I haven't doen anything stupid with it yet. It mostly just sits in my wallet in its little wallet case. I very much prefer keeping my flash drive in my wallet as opposed to my keychain. I also like that the little wallet insert will hold two drive. The only thing I dislike is that the wallet holder is so much thicker than the drive.
What other data would you put on this piece of "contingency hardware",
I have all of my revision control repositories mirrored to my flash drive and also any documentation or notes that I write. That is basically everything that I created myself and would have to do work to replace.
how would you protect the drive itself in case you did have to "swim for it"?
I would probably make sure the data was out of town before I was. Most of this data either doesn't change often (credit card numbers), or it never changes (SSN, birth certificates). Encrypt it, put it on some media of some sort, and send it out of town. Most people probably have friends or family living out of town that they can trust, send it to them. If this is not an option for you, you can probably get a box at a bank out of town I suppose...
remember usage isn't constant though and charges are usually by 95th percentile or capped bandwidth so you nearly always will have bandwidth during off hours that you are paying for but not using.
How about if I make some assumptions since I have no real data? :)
I have no bandwidth pricing available except my own personal provider, and I am paying for disk and cpu, but it is a reasonable place to start. Lets say they can use 70% of my 512mbit for 8 hours per day and for the other 16 hours they use 15%. Unless I am much more tired than I think I am at this late hour, that would be using one third of my total payed for bandwidth.
That triples my cost per song from 0.005 cents to 0.015 cents. Even if Apple pays 100 times more money for bandwidth than I do they are still paying 1.5 cents per song. If they are paying $1500 a month for one third of a T1 they are simply out of their minds :).
Here is a dirty little secret about 95-99% of web hosts. Any time they say 'unmetered' or 'unlimited', they are totally lying.
You might want to shop for a more honest host :p. My current plan is one of their old plans, and it is 30 GB transfer per month. So, I suppose I fibbed just a bit. The new plans have no monthly cap, but are limited to 512mbit for the low end plan. I am planning to switch to one of the new plans, I just have not decided which one yet :p.
I dealt with a bigish webhost a few months back that totally was devious.
I suppose technically this isn't a web host. I have a slice of a Linux box running Virtuozzo. From my point of view, it is similar to User Mode Linux. Mine is running Debian Sarge, although you can get just about whatever you like.
Before you ever get attached to a web host, max out your account for a month. If they offer "5GB of disk space", make a 5GB random text file and upload it. See what happens. If they offer 100GB of transfer, try it out.
My math makes me doubt that my provider is overselling disk space. I am guaranteed a minimum amount of CPU, and I did some math before I signed up. If you bought enough of my small machines to get 100% of the cpu on a single box you came out just shy of the common size SCSI drive. I cannot verify this with the currently listed plans, however, because they list the CPU as "fair share" now.
I am not sure exactly how Virtuozzo handles disk space. UML uses disk images... So if you have a 10 gig UML disk, there is a 10 gig file sitting somewhere for you (my provider also offers UML machines). I have a feeling that Virtuozzo does not use disk images, although I have not cared to research whether that is true or not. Anyway, I am currently using about one third of my available disk, but I have spiked up to near 50%.
I am willing to guarantee that you get shut down on probably 9 out of 10 times.
This is why you should do some research before you buy. I have been quite happy with the service I have gotten from my current provider. I have only been with them for about a year now, though. Before that I had a plain old web hosting account for about 5 years or so... When I got it it was a good deal, although the machine was unavailable for minutes at a time fairly regularly towards the end.
My hosted VM currently has an uptime of 172 days. That reboot was a scheduled outage for an upgrade of Virtuozzo. I believe that is the only time the machine has actually gone down since I've had it.
I think I have drifted rather far from the topic at hand. I was only attempting to point out that if I can get enough bandwidth to transfer 32,000 5 meg songs per month at this price I doubt Apple could be doing much worse. You could transfer 3 times as much data over a T1, and that would still be under a penny per song (assuming of course that the T1 were saturated).
Bandwidth is the cheapest piece of their puzzle. They have to pay for machines, hard drives, and facilities (whether they do it in house or they colocate).
Are you saying bandwidth is free?
Bandwidth is nearly free, yes. It probably costs Apple some fraction of 1% of 99 cents to cover the bandwidth used to download a single song. The $15 dollar a month plan at my web host offers enough bandwidth to serve over 32,000 5 MB songs (unmetered 512mb, about 160GB per month). That works out to 0.005 cents per song.
I would hope Apple is buying enough bandwidth to get a better discount than I do... :p
I love AMD to death, and have for years, but they really didn't have anything that could stand up to Intel's best until the K7/Athlons got to market.
My old AMD 386DX 40Mhz would be very disappointed to hear you say that :).
I was still in high school when I got that machine and at the time it seemed very important to have a faster machine than my friends :p.
That machine was cheaper and almost exactly as fast as one of my friend's 486SX 25Mhz machine. In some benchmarks it was a few percent faster, in others a few percent slower. In the real world you couldn't tell the difference. Thank god we had no use for floating point at the time :).
And Cyrix wasn't even on the same page.
They were for a while. Their integer unit was quite a bit more efficient than Intel. I remember that my Cyrix P200 ran with an FSB of 75Mhz and a core speed of 150Mhz. It was easily as fast as an Intel P200, except for floating point. It would have been lucky to keep up with an Intel P100 or P133 in that category.
Unfortunately for Cyrix, that was when everyone wanted to play Quake and all the latest 3D games were no longer cheater 3D like Doom.
That thing was the Mendoceno Celeron 300A.
I still have one of these, running at 450. That is to say, it ran at 450 the last time it was powered up :p.
But they're different parts, what the hell are you babbling about, you say?
Yes, they absolutely are different parts.
In fact, a lot of the Malaysian 300As were perfectly functional PII 450s that had a bunch of their cache burned out and their clock dropped to 66MHz instead of 100, as mine was, and as a result would run perfectly fine when clocked up to 450 MHz without even needing to change the heatsink and fan from the stock one.
No, sorry. A PII (Deschutes core) 450 had 512k of half speed, off chip cache. The 300A (Mendocino core)had 128k of on core full processor speed cache. Since the cache was twice as fast the Celeron ran as fast (or faster) than an equally clocked PII.
If the 300A and a PII 450 were the same chip with less cache it would not have performed as well.
So if Intel could afford to sell a chip that they had to do MORE work on (burning out the cache) for $500 less,
I believe you are thinking of the later cores... IIRC the Coppermine Celerons were Coppermine Pentium IIIs with a slower FSB and less cache. In all likelyhood many Coppermine Celerons were Pentiums with some faulty cache. Just like the early 486SX chips were 486DX chips with a faulty FPU.
It makes much more financial sense to "burn out" the faulty cache on some chips instead of throwing them away.
Bottom line. Intel is not your friend. They're a business. If they think they can get a $300 profit on a part even after their R&D costs, they'll do it in a heartbeat.
I certainly don't disagree with you here :)
Icecast on Debian Linux already has great performance for streaming, if not "realtime" interrupts
I have no idea why you would need anything remotely like 5 microsecond latency for "streaming." When you stream media you have a buffer at the client end. If you are a bit late with a single packet it shouldn't make any difference.
But I do wonder about loads people have seen with standard datacenter server HW. For example, how many 128Kbps streams can a P4/4.3GHz/128K-cache/512MB-RAM Icecast2 server stream from the local filesystem (preencoded to 128Kbps MP3) to a 100Mbps ethernet, if that's all it's running under the Debian v2.6.10-5-686 kernel/OS?
I would hope this machine would be more than up to the task of saturating a 100Mb link. You need very little CPU and very little disk throughput. I wouldn't be surprised to find out that a 486 with a PCI 100 megabit NIC could do the job just as well.
Wouldn't that be "chmod -w"?
Of course it would be, my mistake. A non-readable file would be quite useless, eh? :)
I want my car to authenticate drivers just like SSH does. Everyone could carry a standard fob with their private key on it, and maybe some other people's public keys. If I want to let a friend borrow my car I just load his public key on my car and tell my car that he is authorized to drive.
Take it a step furthur. You can authorize a valet to open the doors and drive the car with a low rev limiter so he cannot go tooling around town.
The best part about this is that if I lose my fob I only have to remove my public key from the car.
You can take this as far as you like. You can make this same system work for locks in your home and office. I the system is remotely accessible you can do neat things with it... You could temporarily authorize people while you aren't home.
The people who sell copies of keys probably won't like it, though :).
Actually, TCP does a pretty bad job on the error correcting side- packets aren't buffered properly (in that, if a machine suddenly becomes unreachable for a couple of days, TCP will simply never deliver the packets. Zmodem, on the other hand, will resume upon reconnection, discarding spare packets and assembling it's file correctly. Just try to do THAT with a TCP protocol like FTP).
The whole point is that you are running zmodem over ssh which is a protocol that runs over TCP. You are adding the overhead of zmodem's error correction when you already have perfectly good error correction. If you want a more robust way to copy files than sftp and scp (I never mentioned ftp) you should really use rsync instead of zmodem. rsync will have no problem resuming downloads, and it will be able to syncronize files that changed.
It is a much more advanced tool than zmodem.
Now, though, write-protecting a disk image isn't as easy as that...
In my world it would be "chmod -r". I know it works in some of the emulators I have used in the past. YMMV, of course, since I don't have a Mac.
Zmodem rules. I use it all the time. SecureCRT (an ssh client for Windows) has support for it and it's great for logging in somewhere and quickly sending files to/from your local machine, no filesharing necessary.
So you add the overhead of an error checking and correcting protocol (zmodem) over top of a protocol that already checks for and corrects errors (tcp)? Why not just use sftp/scp and skip installing an extra piece of software on your server?
Just to plug SecureCRT, so far it's the fastest, most convienent, best SSH client I've ever used on any platform, which is sad considering it's a Windows app.
What are the advantages over Putty on Windows? What are the advantages over OpenSSH?