A few stats on the box: just under 2 weeks' uptime and 92G of traffic later, 36G (120M packets) is NFS. The loadavg hovers at around 0.25 but heavy NFS/IPSec stuff will peg it at 1.0 easily (P4/1700). This would indeed be a good test bed for NFS over TCP. Sadly the box is in production now.:-/
I know about NFS over TCP, but decided against it as it is relatively untested and further, the networks I deal with are all on the same provider so the connection set-up and tear-downs and other TCP overhead just wasn't worth it for me.
Interesting workaround for the routing damage though... Honest, it's HTTP traffic.:-)
I looked at AFS when I was doing research into this very topic but I could not get anywhere with the documentation. I'm sorry, but openafs.org sucked ass when it came to a simple HOWTO or anything other than "AFS is from IBM, thanks IBM!"
$IPTABLES -L INPUT -vn | $GREP -q 'nfs all -- !ipsec0+' if [ $? -ne 0 ]; then $IPTABLES -I INPUT 1 -i eth0 -j nfs fi
Basically it only allows incoming NFS-related connections over ipsec, dropping anything that is not. NFS port allocation is dynamic by default and I know you can force ports, but this seemed far easier to scale.
One thing I have noticed (and perhaps it's common knowledge to NFS experts) is that in order to get locking to work at all, my NFS clients had to be running statd and lockd. Without 'em everything worked but locking would fail every time.
Problem with 802.11b is that it is still CSMA/CD and as you add nodes you will start having troubles with hidden nodes and there simply not being enough time to give everyone access. It's the same as regular (non-switched) ethernet... above about 40% utilization things start going wrong.
If your uplink is on that same antenna it gets even worse. Best to go to a token-based or polled-node solution.
15:1 overcommit? You've got to be joking. I have not seen a cable provider or telco provider do anything under 100:1, with the average being around 300:1 overcommit on bandwidth. Throw a good forced-proxy in there and play your QoS rules right and you can probably get close to 500:1 without anyone really being offended.
It's not just higher voltage or amplified ethernet signalling. That would violate all crosstalk and interference rules that exist on those trunks. You are forbidden from causing more than xdB of interference, and you can't go much higher than about 300V or so with severely limited current (26-28AWG wire in new installations).
As another poster already pointed out, this is nothing more than an an ethernet-VDSL bridge. I install ethernet-SDSL bridges all the time, it's not magical. Consequently, your 56k modem limit is also (partially) due to this tenet of not violating your energy distribution and noise characteristics over POTS. It also has to do with the robbed-bit signalling and sampling rate of the switch, but it all plays in to the problem.
Depends on the length. They literally say it's $x + $y/kilometer. It's often not worth it unless the bandwidth is free. I am paying CDN$34/mo for the circuit between the ISP I work at and my home.
Nice, unfettered, full-duplex 2048kBps SDSL link for $34/mo, you can't beat it.:-)
Correction: a single PRI will carry 23 calls. (23B+D) PRI's are out-of-band signaled. Without the D channel, there's no way to signal incoming or outgoing calls.
More or less. It depends on the signalling -- If you're talking ISDN, you're absolutely correct until you get into NFAS (non-facility associated signalling) -- in which case you can band together 7 or 8 PRIs (I forget, we use 7 with our DS3s) so you can get 24B *6 + 23B+D to squeeze out as many calls as possible.
T1s can and do carry 24 voice calls using in-band (robbed-bit) signalling. In Canada it was called (confusingly so) PRIs as well, but they were also ear-marked CAS lines. I know they're not PRIs but you ordered them as CAS-signalled PRIs. It's similar in the US but I have no idea what they called the signalling. I thought it was D-letter-letter.
Paperless toilets are far more likely than paperless office. in most civilized nations, it is customery to wash your ass with water than just wipe it with paper.
Pardon my rudeness, but I have to ask -- how is it you can really clean your ass without some kind of supreme water pressure? I mean for normal shits sure, a little wash is all you need, but for some of those disgusting deposits there's scrubbing to be done, man!
So if you burn out a P4, why bother spending more money on a P4 when you could cheaply limp your computer on a celeron till the P5 comes out?
In all my years of computing I've never "burnt out" a processor. If you "burn out" your processor, what's to say the motherboard didn't take a hit too? What a lame-ass argument.
In fact, I've never "burnt out" any solit-state device in any of the computers I've used over the years. Hard drives, definately. But even if the fan goes on a processor it will throttle down/lock up, not "burn out."
I have a Duron running right now that has seen very little down time for over 3 years. It just keeps chugging away, overclocked and over-volted all to hell. It's outlived CPU fans, hard drives, video cards, a stick of RAM, even a motherboard.
That's fine, my story goes the other way: Two Duron problems: one a 1.2G and one a 750..er..1000..er..1000..er...1100 -- The heatsink tit on the socket snapped off and cooked the CPU and mobo. Didn't realize it was the mobo too until I put the 1000 in it and fried that. New 1000, new mobo, the bitch ran at 100oC. Far too hot, even with a copper base ThermalTake heatsink rated for the chip. Then I put an 1100 in a new system and I guess my 15+ years of taking care of computers wasn't gentle enough, since I crushed the corner of the die and zapped that one. Now there's a 1200 in there that's working well.
My wife's Duron 1.2 (Athlon 1.2? I don't remember) is flaky under high CPU load. New high-quality 400W (AMD-approved) power supply. RAM is fine. The CPU runs at a constant 55-60oC. High-quality heatsink and fan, installed properly. Ambient is 19oC. The damned thing just runs hot.
I've never had such a run of bad luck with Intel. Hell, I never believed the "Athlon goes up in a ball of flame" stories until that computer did it (it's a rackmount system and hadn't been moved in about a year and a half, don't give me the "you knocked it around" bullshit). -- At least they'll fold back and save themselves instead of practically exploding. And I've _never_ cracked the die on any coppermine.
AMD's nice from a price perspective but I won't buy another one.
MS Access can easily be replaced with Postgres and something like Data Architect and Rekall, both from TheKompany. Hell, there are some pretty decent OSS forms generators out there too which also help you get away from your Access habit.
Why? With Appgen running natively on win32, mac and linux it seems that Quicken/Quickbooks has some competition. I know that I am looking at the larger package to replace the shite that AccPAC/MiSYS is at this company.
That was a pre-emptive dupe... we KNOW slashdot'll dupe the story at least once more. It's time to liberate the submission queue; you're either with us or against us.
That's what's holding me back from going mythtv -- I am on ExpressVu and want to be able to save the MPEG stream directly to HDD or at with a minimum of screwing around.
There are some DVB PCI cards from Germany (ExpressVu is standard DVB, throw in your access card and you're done) but they're on the order of $400!! I've been through the schematics of my old 2700-series receiver and while I can tap off the digital audio, the unencrypted MPEG video stream seems to never leave the custom decrypt/decode chip.:-(
I suppose I could use RCA out and an IR mouse to change channels, but I am really looking for something I can put all in one box and, as I said earlier, not screw around.
I have to say I love my NexII. The IIe looks like a black version with the ability to use USB power. whoop de doo. And the IA seems to introduce voice recording and FM radio, which is nothing crazy...
Two things I really do wish the NexII-series of players had was
true SHUFFLE (not random) play
save your position so when you stop and power off the player it remembers what it did play and where you are
That's really all I would like out of it. Battery life is great, but the case is a little cheap feeling (and the battery cover flys off if you ever drop it). A pair of AAs gives TONS of juice for this thing.
I don't use/have OGGs, although I would recode my entire CD library again if I could use it on the NexII. My NexII was about $75 and with a couple of 256M CF devices, I can shuffle my music around and I don't feel constrained. It really is a superb little device.
IRC was worse than useless to her, and apparently to any newbie that pops in.
I dunno about that -- #linuxhelp on freenode is pretty decent, as is #freeswan, #kernelnewbies, #perl and #openembedded (for ipsec, kernel-specific stuff, Perl and anything OpenZaurus, respectively). I think that by far the problem is the attitude of the people asking. I hang out in those channels semi-frequently and over and over again I see people come in and ask questions that could have been answered by practically putting the exact question into Google or pulling up the man page on the program they're asking about.
Free help is very intolerant of stupidity. Helping someone out is not a problem, but if the stupid shit won't even try to help themselves they will be mocked. I don't think I've ever had trouble with IRC if I had taken even cursory steps to narrow the problem down.
Free help is also very intolerant of people who demand that they be helped. Go buy the boxed set of RedHat or Suse and demand that the paid tech support staff there help you.
Actually that brings me to another point: If you are a newbie, go to #distroname (#redhat, #debian, #gentoo, #suse, etc.) -- they seem to be far more tolerant of cluelessness and the don't know where to start-types. Except #slackware -- if you're running Slack they expect a lot more from you (which to me is a good thing, it's not a handholding distro). Conversely, if you really need to get technical, seek out a Slackware user. a good 75% of the time they're the ones answering the really odd questions anyway.
The PalmOS devices continue to be successful because they don't try to cram some variant of Unix or Windows in them and, instead, stick to an OS that is appropriate.
Actually I don't consider Palm all that successful anymore. Power-hungry colour screens, MP3/voice capabilities, cameras, wireless... They are running into the PocketPC/Zaurus arena and they will fail because their API was never meant to handle these things.
IMO, my Palm Vx (well maybe the m500 because it has the SD/MMC port) was the pinnacle of Palm's capabilities. More rugged than the plastic cases before them, enough memory to hit the 95% of what people want, easily 8-15 days runtime on a single charge and a clean, unencumbered API.
If you want an ultraportable computer, get yourself a PocketPC, Zaurus (I have one of these too) or even those mini Sony Vaios. If you want a PDA, get the Palm Vx or m500. They are for totally different markets.
Actually I found Palm's memory handling quite logical -- you're writing for a platform with a slow but energy-miserly processor and small amount of memory. Why bog it down with unnecessary memory buffers memory reorganization?
Access all memory as handles and lock what you need only when you need it. It's a pain in the arse compared to the normal "memory's always there and can be used as if it's never going to run out" but in the context I mentioned above, it's really not too bad. You work within your constraints.
802.16 is designed for fixed outdoor operation and the antennas are much too big to fit in a PCMCIA card, so it won't replace 802.11.
What frequency band does 802.16 work on? I imagine it's still in the high-MHz/low-GHz range which means a 1/4 wave dipole can easily be mounted on any notebook computer.
PS. Btw, Async IO in NT *does* rule... all of the threading mechanisms in NT rule, IMHO.
I thought NT's threading mechanism was inferior to Linux's "processes (pretty much) == threads" methodology. That and the "every thread has its own private message queue" on NT drives people batty. To each their own, I suppose.
You will see performance enhancements with Gentoo if you set up your compiler flags properly. I believe that the standard compiler flags for most distros these days is -march=i386 -mcpu=i686 -- produce code that'll run on anything Linux'll run on, but order them and optimize them for execution on P6 class machines.
By specifying the proper -mcpu you can see some significant enhancements, especially (I'm speaking from history here, I haven't kept up with instruction ordering optimization since P5) since some of the instructions used on i486 would crawl on i586 and again be significantly different on i686.
The downside, of course, is that people will use -march=mpentium4 or something and then wonder why nothing works when their system dies and they have to plug their hard drives into an old P2 or P3 sitting around.... (not that I've learned that the hard way, noooooooo...
Since I'm ranting about ease of use anyway, and maybe using my Imac is starting to jade me, but I'm getting tired of running./configure, chasing down libraries, fixing compiler errors, and such in order to setup programs.
Honestly that's not an issue on practically any distro. Personally I don't like the fact that I can almost never get the package I want, even with Debian and it's 10k packages. GNU-Radiusd, PostgreSQL with SSL support, qmail with the patches and options I think are relevant, (same with courier-imapd), vpopmail... it's tiring to look only to find that it's not there and you need to build anyway, and then realize that if you're out to make it fit in with your system properly you have to do a lot more work with the Mandrakes and Debians out there.
Slackware's cool because it's package manager is tight enough to keep track of what's where, and it stays the fuck out of the way for everything else. That is a package manager for someone like me. I find I have LESS hassle managing packages on Slack than I have ever with RedHat, Mandrake or Debian, even when it comes to security updates like the recent Samba update or the SSH fiasco of way back.
The simple truth of the matter is that for most updates, the libraries do not have to be updated unless the problem is in the library or there's been some major update. And for production systems, you're almost never going to run across that.
Slashdot Mirror
Dammit I hate replying to myself.
A few stats on the box: just under 2 weeks' uptime and 92G of traffic later, 36G (120M packets) is NFS. The loadavg hovers at around 0.25 but heavy NFS/IPSec stuff will peg it at 1.0 easily (P4/1700). This would indeed be a good test bed for NFS over TCP. Sadly the box is in production now. :-/
I know about NFS over TCP, but decided against it as it is relatively untested and further, the networks I deal with are all on the same provider so the connection set-up and tear-downs and other TCP overhead just wasn't worth it for me.
Interesting workaround for the routing damage though... Honest, it's HTTP traffic. :-)
I looked at AFS when I was doing research into this very topic but I could not get anywhere with the documentation. I'm sorry, but openafs.org sucked ass when it came to a simple HOWTO or anything other than "AFS is from IBM, thanks IBM!"
I use a very simple script to help keep NFS secure:
Basically it only allows incoming NFS-related connections over ipsec, dropping anything that is not. NFS port allocation is dynamic by default and I know you can force ports, but this seemed far easier to scale.
One thing I have noticed (and perhaps it's common knowledge to NFS experts) is that in order to get locking to work at all, my NFS clients had to be running statd and lockd. Without 'em everything worked but locking would fail every time.
Problem with 802.11b is that it is still CSMA/CD and as you add nodes you will start having troubles with hidden nodes and there simply not being enough time to give everyone access. It's the same as regular (non-switched) ethernet... above about 40% utilization things start going wrong.
If your uplink is on that same antenna it gets even worse. Best to go to a token-based or polled-node solution.
15:1 overcommit? You've got to be joking. I have not seen a cable provider or telco provider do anything under 100:1, with the average being around 300:1 overcommit on bandwidth. Throw a good forced-proxy in there and play your QoS rules right and you can probably get close to 500:1 without anyone really being offended.
Incorrect.
It's not just higher voltage or amplified ethernet signalling. That would violate all crosstalk and interference rules that exist on those trunks. You are forbidden from causing more than xdB of interference, and you can't go much higher than about 300V or so with severely limited current (26-28AWG wire in new installations).
As another poster already pointed out, this is nothing more than an an ethernet-VDSL bridge. I install ethernet-SDSL bridges all the time, it's not magical. Consequently, your 56k modem limit is also (partially) due to this tenet of not violating your energy distribution and noise characteristics over POTS. It also has to do with the robbed-bit signalling and sampling rate of the switch, but it all plays in to the problem.
Depends on the length. They literally say it's $x + $y/kilometer. It's often not worth it unless the bandwidth is free. I am paying CDN$34/mo for the circuit between the ISP I work at and my home.
Nice, unfettered, full-duplex 2048kBps SDSL link for $34/mo, you can't beat it. :-)
Correction: a single PRI will carry 23 calls. (23B+D) PRI's are out-of-band signaled. Without the D channel, there's no way to signal incoming or outgoing calls.
More or less. It depends on the signalling -- If you're talking ISDN, you're absolutely correct until you get into NFAS (non-facility associated signalling) -- in which case you can band together 7 or 8 PRIs (I forget, we use 7 with our DS3s) so you can get 24B *6 + 23B+D to squeeze out as many calls as possible.
T1s can and do carry 24 voice calls using in-band (robbed-bit) signalling. In Canada it was called (confusingly so) PRIs as well, but they were also ear-marked CAS lines. I know they're not PRIs but you ordered them as CAS-signalled PRIs. It's similar in the US but I have no idea what they called the signalling. I thought it was D-letter-letter.
Paperless toilets are far more likely than paperless office. in most civilized nations, it is customery to wash your ass with water than just wipe it with paper.
Pardon my rudeness, but I have to ask -- how is it you can really clean your ass without some kind of supreme water pressure? I mean for normal shits sure, a little wash is all you need, but for some of those disgusting deposits there's scrubbing to be done, man!
So if you burn out a P4, why bother spending more money on a P4 when you could cheaply limp your computer on a celeron till the P5 comes out?
In all my years of computing I've never "burnt out" a processor. If you "burn out" your processor, what's to say the motherboard didn't take a hit too? What a lame-ass argument.
In fact, I've never "burnt out" any solit-state device in any of the computers I've used over the years. Hard drives, definately. But even if the fan goes on a processor it will throttle down/lock up, not "burn out."
I have a Duron running right now that has seen very little down time for over 3 years. It just keeps chugging away, overclocked and over-volted all to hell. It's outlived CPU fans, hard drives, video cards, a stick of RAM, even a motherboard.
That's fine, my story goes the other way: Two Duron problems: one a 1.2G and one a 750..er..1000..er..1000..er...1100 -- The heatsink tit on the socket snapped off and cooked the CPU and mobo. Didn't realize it was the mobo too until I put the 1000 in it and fried that. New 1000, new mobo, the bitch ran at 100oC. Far too hot, even with a copper base ThermalTake heatsink rated for the chip. Then I put an 1100 in a new system and I guess my 15+ years of taking care of computers wasn't gentle enough, since I crushed the corner of the die and zapped that one. Now there's a 1200 in there that's working well.
My wife's Duron 1.2 (Athlon 1.2? I don't remember) is flaky under high CPU load. New high-quality 400W (AMD-approved) power supply. RAM is fine. The CPU runs at a constant 55-60oC. High-quality heatsink and fan, installed properly. Ambient is 19oC. The damned thing just runs hot.
I've never had such a run of bad luck with Intel. Hell, I never believed the "Athlon goes up in a ball of flame" stories until that computer did it (it's a rackmount system and hadn't been moved in about a year and a half, don't give me the "you knocked it around" bullshit). -- At least they'll fold back and save themselves instead of practically exploding. And I've _never_ cracked the die on any coppermine.
AMD's nice from a price perspective but I won't buy another one.
MS Access can easily be replaced with Postgres and something like Data Architect and Rekall, both from TheKompany. Hell, there are some pretty decent OSS forms generators out there too which also help you get away from your Access habit.
Quicken is also supported.
That's a biggie.
Why? With Appgen running natively on win32, mac and linux it seems that Quicken/Quickbooks has some competition. I know that I am looking at the larger package to replace the shite that AccPAC/MiSYS is at this company.
That was a pre-emptive dupe... we KNOW slashdot'll dupe the story at least once more. It's time to liberate the submission queue; you're either with us or against us.
That's what's holding me back from going mythtv -- I am on ExpressVu and want to be able to save the MPEG stream directly to HDD or at with a minimum of screwing around.
There are some DVB PCI cards from Germany (ExpressVu is standard DVB, throw in your access card and you're done) but they're on the order of $400!! I've been through the schematics of my old 2700-series receiver and while I can tap off the digital audio, the unencrypted MPEG video stream seems to never leave the custom decrypt/decode chip. :-(
I suppose I could use RCA out and an IR mouse to change channels, but I am really looking for something I can put all in one box and, as I said earlier, not screw around.
I have to say I love my NexII. The IIe looks like a black version with the ability to use USB power. whoop de doo. And the IA seems to introduce voice recording and FM radio, which is nothing crazy...
Two things I really do wish the NexII-series of players had was
That's really all I would like out of it. Battery life is great, but the case is a little cheap feeling (and the battery cover flys off if you ever drop it). A pair of AAs gives TONS of juice for this thing.
I don't use/have OGGs, although I would recode my entire CD library again if I could use it on the NexII. My NexII was about $75 and with a couple of 256M CF devices, I can shuffle my music around and I don't feel constrained. It really is a superb little device.
IRC was worse than useless to her, and apparently to any newbie that pops in.
I dunno about that -- #linuxhelp on freenode is pretty decent, as is #freeswan, #kernelnewbies, #perl and #openembedded (for ipsec, kernel-specific stuff, Perl and anything OpenZaurus, respectively). I think that by far the problem is the attitude of the people asking. I hang out in those channels semi-frequently and over and over again I see people come in and ask questions that could have been answered by practically putting the exact question into Google or pulling up the man page on the program they're asking about.
Free help is very intolerant of stupidity. Helping someone out is not a problem, but if the stupid shit won't even try to help themselves they will be mocked. I don't think I've ever had trouble with IRC if I had taken even cursory steps to narrow the problem down.
Free help is also very intolerant of people who demand that they be helped. Go buy the boxed set of RedHat or Suse and demand that the paid tech support staff there help you.
Actually that brings me to another point: If you are a newbie, go to #distroname (#redhat, #debian, #gentoo, #suse, etc.) -- they seem to be far more tolerant of cluelessness and the don't know where to start-types. Except #slackware -- if you're running Slack they expect a lot more from you (which to me is a good thing, it's not a handholding distro). Conversely, if you really need to get technical, seek out a Slackware user. a good 75% of the time they're the ones answering the really odd questions anyway.
The PalmOS devices continue to be successful because they don't try to cram some variant of Unix or Windows in them and, instead, stick to an OS that is appropriate.
Actually I don't consider Palm all that successful anymore. Power-hungry colour screens, MP3/voice capabilities, cameras, wireless... They are running into the PocketPC/Zaurus arena and they will fail because their API was never meant to handle these things.
IMO, my Palm Vx (well maybe the m500 because it has the SD/MMC port) was the pinnacle of Palm's capabilities. More rugged than the plastic cases before them, enough memory to hit the 95% of what people want, easily 8-15 days runtime on a single charge and a clean, unencumbered API.
If you want an ultraportable computer, get yourself a PocketPC, Zaurus (I have one of these too) or even those mini Sony Vaios. If you want a PDA, get the Palm Vx or m500. They are for totally different markets.
Actually I found Palm's memory handling quite logical -- you're writing for a platform with a slow but energy-miserly processor and small amount of memory. Why bog it down with unnecessary memory buffers memory reorganization?
Access all memory as handles and lock what you need only when you need it. It's a pain in the arse compared to the normal "memory's always there and can be used as if it's never going to run out" but in the context I mentioned above, it's really not too bad. You work within your constraints.
802.16 is designed for fixed outdoor operation and the antennas are much too big to fit in a PCMCIA card, so it won't replace 802.11.
What frequency band does 802.16 work on? I imagine it's still in the high-MHz/low-GHz range which means a 1/4 wave dipole can easily be mounted on any notebook computer.
PS. Btw, Async IO in NT *does* rule... all of the threading mechanisms in NT rule, IMHO.
I thought NT's threading mechanism was inferior to Linux's "processes (pretty much) == threads" methodology. That and the "every thread has its own private message queue" on NT drives people batty. To each their own, I suppose.
You will see performance enhancements with Gentoo if you set up your compiler flags properly. I believe that the standard compiler flags for most distros these days is -march=i386 -mcpu=i686 -- produce code that'll run on anything Linux'll run on, but order them and optimize them for execution on P6 class machines.
By specifying the proper -mcpu you can see some significant enhancements, especially (I'm speaking from history here, I haven't kept up with instruction ordering optimization since P5) since some of the instructions used on i486 would crawl on i586 and again be significantly different on i686.
The downside, of course, is that people will use -march=mpentium4 or something and then wonder why nothing works when their system dies and they have to plug their hard drives into an old P2 or P3 sitting around.... (not that I've learned that the hard way, noooooooo...
Since I'm ranting about ease of use anyway, and maybe using my Imac is starting to jade me, but I'm getting tired of running ./configure, chasing down libraries, fixing compiler errors, and such in order to setup programs.
Honestly that's not an issue on practically any distro. Personally I don't like the fact that I can almost never get the package I want, even with Debian and it's 10k packages. GNU-Radiusd, PostgreSQL with SSL support, qmail with the patches and options I think are relevant, (same with courier-imapd), vpopmail... it's tiring to look only to find that it's not there and you need to build anyway, and then realize that if you're out to make it fit in with your system properly you have to do a lot more work with the Mandrakes and Debians out there.
Slackware's cool because it's package manager is tight enough to keep track of what's where, and it stays the fuck out of the way for everything else. That is a package manager for someone like me. I find I have LESS hassle managing packages on Slack than I have ever with RedHat, Mandrake or Debian, even when it comes to security updates like the recent Samba update or the SSH fiasco of way back.
The simple truth of the matter is that for most updates, the libraries do not have to be updated unless the problem is in the library or there's been some major update. And for production systems, you're almost never going to run across that.
Gentoo is the Mandrake of Slashdot...
Proofread, dammit... Gentoo is the Mandrake of Slackware, not Slashdot...