97 new of 122 responses total.
I'm curious how you arrived at that number. To answer Sindi's question, it's pretty plausible that a 384kbit/s connection might deliver about 30kByte/s useful throughput, as you need to adjust for 8 bits / byte and also need to add on substantial protocol overhead, first for the wireless link, then for network and transport protocols (TCP/IP), and so on. My general rule of thump with the ISP we run at work is that we expect about 20% of the speed we limit the customers to to be protocol overhead, meaning the customer gets about 80% of what we set them to.
I was at the library sharing a connection with at least 30 other people. Could that reduce the speed of a 1.5MBit to what I got? I downloaded Opera, and it started off slow but by the end was up to 30Kbit, which implies there is some buffer and it might have gotten even faster after the first 5MBytes. The librarian (not the one that told me I needed to fill in my library card number to use the internet there, which was not needed) said the internet would go faster on my computer via wireless, than on their computers, because their computers were 'old' and 'basic'. We took our fastest laptop along, but I don't see how the computer speed would limit the internet connection speed, for just a download. Maybe a game player would notice a difference. The people at the desk there seem to know less than I do, and gave us a wrong answer about getting a USB memory stick recognized the previous time.
Re #26: 30 * 1024 = 30,720 bytes/sec 30,720 x 8 = 245,760 bits/sec ~= 246 kilobits per second of payload data. I made no attempt to guess the overhead, but 384 kbits/sec sounds entirely reasonable to me. Re #27: Earlier I thought you said 30 kbytes/sec, but now you say 30 Kbit (which isn't a sensible measurement in any case). Keep in mind that during downloads, the bottleneck need not be at your end. You could have a gigabit connection to the Internet, but if the chap at the other end only has ISDN BRI, his upstream speed isn't going to exceed 128 kbits/sec (discounting any software compression) and no amount of improvement at your end can significantly change that.
Ooops. In my reply to #27, I meant to say 30 Kbytes/sec, not 30 kbytes/sec.
In your mind, what does the capital "K" mean? I ordinarily would assume you were just mis-capitalizing the metric prefix abbreviation representing "kilo". Are these Kelvin bytes?
Regarding #28; It depends on how many bits are in a kilobit. If one takes it to be 1024, then the 1024's would just cancel out and you've have a 240kbit/sec. You are clearly taking a `kilobit' to mean 1000 bits.
One Kilobyte is 1,024 bytes. One kilobit is 1,000 bits.
Hmm.
The computer which downloaded at the library (from opera.com) at 30 per second, and in which the orinoco Linksys card does not work at al in linux but works fine in Win98, won't work in Win98 with two hardware modems, 56K and 33K (per second?) that work in another computer adn also work on this computer in linux. One can't open port, the other says no dial tone. ??? Another laptop also works online in linux but not Win98, and wont' work with a different linksys card in Win98 (wireless). Looks like you need to buy sefverAL PCMCIA CARDS AND EXPERIMENT UNTIL ONE WORKS, THEN RETURN THE REST. tHEY also keep changing versions, so if Ver 1 works ver 2 won't. Excuse typos, downloading a 6.5MB file to test a 2wire card with.
56k and 33k (lower case k signifies 1,000 as in km, kg etc.) Are they PC-Card (16-bit PCMCIA) modems?
make that 33.6 kbits/sec ;-)
Yes, pcmcia modems. A winmodem sort of works in another laptop - it dials (usually, sometimes can't find the dial tone), connects (usually), but can't access any websites or ping. The 33K hardware modem does the same, but in linux it works perfectly. Last night I tried to test three cardbus modems in Win98. Netlux driver was not at driverguide, and netlux driver downloads requires a password. 2wire driver not available, I found one for the same model number of a Sceptre (6.5MB download, one hour) but Win98 won't accept it. Had the 30MB CD for a Linksys 54G but it refuses to install on my computer unless I give it Internet Explorer 5.5, and the Adobe Acrobat on the same CD refuses to install without IE 5.01. (I think I have 5.0 in Win98, maybe 4.5). Why does one need ANY browser to install a pdf reader or network card software????? I would not pay a penny for anything from Linksys. The non-cardbus card won't work at all on one computer in linux (unresolved symbols) and it crashes on the other computer with a 2.2 kernel.
It's a shame that MS Windows is such a pain when it comes to standard modems. I'm surprised that anyone would build Card -bus modems, since PC-Card provides adequate bandwidth.
Win98 would not dial with a Winmodem in one computer, and would dial and not access websites with a Winmodem in another computer (both laptops) besides having problems with the hardware modem that worked in linux (could not open port). Com2 was disabled in BIOS, modem was Com2 (ttyS1 in linux). I don't think these winmodems are cardbus, the wireless network cards were. 11Mbits per second 16-bit or 32-bit, so why cardbus for those?
That's a good question. PC-Card would seem to provide enough bandwidth for 802.11b. Perhaps vendors anticipate host computers that feature Cardbus slots but lack PC-Card?
I thought anything that took cardbus also took 16-bit pcmcia. The 56K modem that would not work in Windows appears to be a hardware modem that also won't work in linux, but came labelled '56K okay'. That may only mean the computer found it (at Kiwanis). Combo card ethernet and modem. I also got two other combo cards one with dead ethernet one with dead modem. this was too good to be working.
I think all the Cardbus slots I've seen were also PC-Card (16-bit PCMCIA) compatible, but that may not be the case for ever.
The new in the box Belkin adaptor that fits into a PCI slot and lets you put in pcmcia cards does NOT take cardbus cards. It also crashes Win98. Made in 2001. Jim is trying it in DOS then linux, in his computer with no ISA slots, so we won't need to hook up a laptop computer as gateway (or reprogram a router to pick up a wireless signal and pass it along). Today I stuck an older non-cardbus wired ethernet card into a Win98 laptop (the card works in linux there) and plugged into a router and ran ipconfig and could ping the router. Then I put another wired card into the linux computer, plugged that into the same (working) router and did udhcpc. I could ping between the two computers! (Also the router). ping 192.168.1.0 pinged all three alternately. The router is 192.168.1.1, the computers 101 (the first) and 102 (the second). Then I ran telnetd on the linux computer, was told I need /bin/login which this linux does not have, switched linuxes, did it again, and could telnet from Windows to Linux. I could not log in as root (I need to edit /etc/securetty to add ttyp0 through 4 to use root on telnet on four terminals) but I could as user. Since I have no /home it put me into /root and I could write and save a file, but not use zgv (which cannot be run as user due to something I forgot how to fix) or mutt (it made very loud noises in Windows). I should fix it to log in as root. Then I tried busybox httpd (busybox is a small binary that does the basic parts of many other things) on linux and tried to access 192.168.1.102 with Opera and got 404 not found. So I got a bigger (75K) mini_httpd and could access linux with Win98 via opera, and download files. But I want to move files from Win98 to linux and need either a Windows httpd (which may be included in the 100MB plus of Personal Web Server, or in something even bigger) or a linux ftpd which might be wu_ftpd which I need to get. This will let me unload the USB camera to an XP computer then transfer the files via network cable to the linux computer, which has no USB. (We have a computer with USB and bad CD-ROM drive, or one with no floppy drive, and one that won't work with wireless networking in Windows, but the one without USB does everything properly and is most reliable. Thanks Scott). Next project is to put linux on someone else's laptop which may already have XP on it, maybe using qemu and a 5MB linux image file.
It sounds as though you're making useful progress. I suggest creating a non-root user and using "su" when you need root privilages. I quite like thttpd, which is a light web server that is easy to configure and use. I made some progress myself on Saturday: installed NetBSD 3.1 on a spare PC (500 MHz AMD K6-2, 128 Mbytes RAM, 6 Gbyte hard disk). I should probably look for a PCI wireless LAN card to install in that machine.
I seem to need root privileges to do much of anything so I have been running as root for four years with no problems. 'user' could not use zgv, and when I typed mutt (telnetted with Windows to linux) I got very loud noise.
Literally a loud noise? I think the only noise you're likely to hear from a telnet session is the bell, but perhaps if the speakers were cranked up, that might be surprising. If you have to be root all the time, then your unix is broken.
Re resp:43: Some of the PCI adapters sold for use with wireless cards are not actually generic PCMCIA slots, but rather adapters specially designed for that company's cards.
The noise sounded like very loud static. Windows seems to be set up to make noises on that computer despite my checking off Mute - is there some place to tell it not to make system noises? My linux is designed to run as root. svgalib has problems when used as user, so does Xvesa, then I would need to give myself privileges to save files to various directories. Most of the time I spend doing administration (adding and modifying programs, compiling, etc.).
There is something wrong with the MS Windows machine (either with the hardware, or with the driver software) if it is making a loud hash noise while you have mute selected. It seems as though data that doesn't represent sound is somehow finding its way to the digital-to-analogue convertor (DAC). If your Linux is designed to run as root, then I consider its design broken.
That's the way "Linspire" (formerly "Lindows") is designed to run, though I doubt that's the distro she's using. I wouldn't want to run that way, but some people do.
/me shudders
I have the same software on several machines and can easily restore it. I am root in DOS, why not in linux? Win98 also played all the WIndows noises at the library despite being muted. How do I turn off Windows noises? I can turn off online Opera noises.
> I am root in DOS, why not in linux? For the very same reason you shouldn't be root in DOS (which, admittedly has no other options) or in Windows (which, nowadays at least, does) -- that always running at the highest privilege level makes it trivially easy for a rogue program to corrupt the entire system.
Some of us who have to follow accepted security practices in our work generally do the same at home because it keeps up our "A" game. On top of that I consider a lot of the data I keep at home to be fairly important and I like to see it protected against loss.
But I don't have any rogue programs and have not had problems in four years and if I did I would just copy back the software from one computer to another. Today I am trying to figure out why udhcpc does not work with my small laptop kernel but does with another, so I tried a third. Stuck in a wireless card in the kitchen, forgot to plug in the router, got an IP number and three dns numbers from some network with signal strength -139dB (noise -156dB) but it did not last long, went back to signal strength 1/48 to 17/48. For the 20 seconds it lasted I could go online with two browsers, so all I need now is to wait for the free county wireless signal, or take the linux computer to the library instead of windows. Victory! (I still need to fix my laptop kernel to use udhcpc). I have detected four networks with iwconfig too.
Regarding #55; That you know of, you mean. In general, running is root is considered bad form. If you can get away with it, then great; but don't be surprised if it bites you in the butt one day.
Where would a rogue program come from and what is it?
It could come from anywhere. The `rm' command, if used without caution as root, could become a `rogue' program.
That's a good point: Running as a user helps to protect you not just from other systems or people with hostile intent, but also from yourself. rm * in the wrong directory is a great example of that.
It also provides more protection from a userland application that loses its mind.
But I have an identical computer next to this one linked by ethernet cable. And I could also just copy everything to another partition here. It is a 10GB drive and my linux fits fine in 1GB or less. I keep what I have compiled at a shell account.
If they're separate partitions on the same disk, or even seperate disk drives connected to the same computer, then they provide no significant redundancy (since an errant program, run as root can simply erase all connected disks). Having an identical computer connected via a network cable is a handy thing (I do something similar myself), but if they're running the same system software, they're going to share any vulnerabilities that it has. In your case replacing the data may be trivial, but for most people that is probably not the case. Besides, isn't it a bit like saying "I don't lock my front door because I have home- owner's insurance"?
Keesan, I have fat fingered a single "." into "..". The guy who hired me at my current job has still never lived down a "chmod -R 600 .." instead of "chmod -R 600 .", which made everything in the entire machine unreadable by anyone except root and unexecutable by anyone including root. Had he been running as his normal login, he would have done very little damage, but because he was using the root login, he and a colleague wound up having to reimage the server and restore settings and data from backup, which is a royal P.I.T.A.
Sindi, you can do what you want. But like I said, it might bite you in the ass. If you don't care, then great, I really have *no* interest in trying to convince you otherwise, but it is bad form.
Most of what I do in linux is not doable by user. The only damage I have done is crashing things, and e2fsck fixed it all but once (when I think the drive was bad). I read email and browse mostly in DOS (kermit, lynx, pine). And replacing the contents of a house is a lot different from copying software between computers, which I do every time I change drives anyway. I don't want to have to change write permissions on a bunch of directories, or write only to one directory with subdirectories, wastes time. I discovered iwlist (on the card that supports it) shows 8 available connections. I need to compile a newer pcmcia module for the other card to work with iwconfig. I have the source code. Do I uncompress it into the pcmcia source code directory and make all again? Can I make just the one module and if so how? I think udhcpc (a small dhcpc) can choose between connections by address with -r .
I live inside my computer ;-)
Re resp:56: I learned not to run as root when I accidentally typed "rm -rf /usr" instead of "rm -rf ~/usr".
I would be extremely nervous to run as root all the time. On modern Unix and Linux systems, I find that the "sudo" command provides a rational middle ground, providing reasonable protection against unfortunate accidents while not requiring a full-blown root login every time I want to do something requiring administrative privileges.
Sudo has a lot of advantages: among them, it logs commands via syslog. So if someone does something mistakenly, there is at least a lot which one can look at to see what happened.
Regarding #32; I wanted to write more about this last week, but was tremendously busy. This is not universally true; in some contexts, a kilobit is canonically taken to be 2^10 bits. The wikipedia article, for instance, includes this interpretation.
S.I. says one k is 1,000.
The problem is that both pow(10,3) and pow(2,10) are correct, depending on the context. System Internacional uses it to represent the former, while computers (which 'think' in base-2 rather than base-10) use it to refer to the latter. As an approximation, they are close, but the difference does matter.
That's the great thing about standards: there are so many to choose from. Why is one kilobyte considered to be 2^10 = 1024 bytes? Why do disk vendors give capacities of hard drives measured in 1k = 1000 bytes, when the operating system views things as power of two block sizes? Which is more standard than the other? I'll grant that the 1,000 bits == 1 kilobit definition is standard, but it is not universal.
10^3 is k, per SI. 2^10 is K, per longstanding convention.
Disk vendors don't specify disk capacity in K or k. They use Gbytes or Mbytes and adhere to the S.I. definitions of those.
Regarding #74; I don't ever remember seeing that, but maybe I wasn't looking in the right places. Regarding #75; Actually, if you want to pick nits, they do: the G or M or K just refers to 9, 6, or 3 as an exponent for 10. So, technically, Gbytes are in the same equivalence class modulo 10. But my point was that disk vendors rate their products in terms of powers of ten, not powers of two. Saying KB was just convenient, as the kilobyte is essentially the first `real' unit in common usage after the byte (that is, few people talk in terms of decibytes or centibytes).
resp:76 Picking more nits, decibyte is 1/10th of a byte. Dekabyte is 10 bytes. Of course, with word-lengths in powers of 2 (32 or 64), dekabyte is sort of an awkward amount of data.
My bad.
Re #73: k != K
Like I said, that's the first I've heard of that. Got a citation?
K != k just as M != m (M is 1,000,000 and m is 0.001).
Ah, I see what you mean now. I thought you meant K = 2^n while k = 10^n or something. Yes, you are right.
Apparently the International Electrotechnical Commission (whoever they are) want us to use "Ki" in place of K for 1,024. Computer people have been using K for 1,024 for a very long time though.
Re #81: 1K = 1,024 1k = 1,000.
Erm, that was Re #82 ;-)
Regarding #84; You know, I've never heard that before. Like I said, do you have a citation?
I'll have a rummage for one.
This is interesting: http://en.wikipedia.org/wiki/Kibibit They seem to use ``Kib'' or ``Kibit'' (with a capital K) instead of ``Kbit'' or ``Kb.'' They do acknowledge that ``kilobit'' can be either 2^10 or 10^3 depending on context.
This is also interesting: http://en.wikipedia.org/wiki/Binary_prefix Note that they say that in the SI system, `K' (capitalized) stands for Kelvin, as a unit of temperature, and `k' (lowercase) only stands for `kilo.' They say that outside of SI, K and k are mostly interchangable, and can refer to either 2^10 or 10^3, as I had originally said. To wit: 'The one-letter abbreviations are identical to SI prefixes, except for "K", which is used interchangeably with "k" (in SI, "K" stands for the kelvin, and only "k" stands for 1,000).' However, they do say that as of 2005, the binary meanings are deprecated.
k (as a multiplier prefix) should only ever be used to mean 1,000. Everywhere I've ever worked or studies, K has been capitalised to differentiate it from k. Telecomms people talk in terms of kbits/sec, and mean 1,000 bits. Computer people talk in Kbytes and mean 1,024. It's not rocket science ;-)
Here's an example of K from a PDP-11 manual... http://pdos.csail.mit.edu/6.828/2005/pdp11/pdp11-40-000009.html
Here's a KIM-1 manual from 1976... http://users.telenet.be/kim1-6502/6502/usrman.html
re #90: You're right that "it's not rocket science", but it's not universally or consistently applied, either, which means making assumptions based on the use of "k" or "K" is dangerous if you need better than approximate numbers.
It's been consistently applied in my experience, but it's true that a few people don't use it correctly.
Take a random sample of computer manuals, text books (I hope they're right!) or EPROM / SRAM data sheets. K = 1,024 is a long-standing convention.
Yes, but you were talking specifically about K = 1024 and k = 1000, and in neither of the two references that you posted can I find such a distinction. Everyone knows that most computer manuals refer to K as 2^10 = 1024. Your claim was that they also refer to k as 1000, which is not universal, and in fact, is a convention I've never heard of before, and is not supported by your evidence. If telecom people refer to kbits as 1000 bits, that's great, but what McNally says is true: if you want to be exact, you've really got to specify.
k is 1,000 because of S.I. (km, kg, kW etc.) It's only necessary to specify because some people seem underinformed.
*sigh* It's not being underinformed, Andy, it's recognizing that standards aren't universally followed. I don't know how to explain it better than that. Really, though, it's true: not everyone follows the same standards.
Never mind. Let's talk about wireless networking. My next wireless networking task is to find a PCI 802.11g adaptor that works with NetBSD. This could take a while.
How are you searching, in BSD online discussions?
I'll probably start with the man pages for common device drivers such as ath(4) and perhaps wi(4). Hopefully I'll be able to find a card that has an appropriate chipset.
There are lists of linux-compatible pcmcia cards. Why don't you search on BSD PCI wireless?
The man pages that I mentioned include lists of cards that are supposed to work. Sadly some manufacturers will change the chipset in a product without changing the model number so it can be something of a lottery.
Re resp:75: My impression is that the computer world pretty universally used K=1,024 until marketing types realized they could put a bigger number on hard disk packages if they used K=1000. For a while they tried to avoid confusion (and presumably false advertising claims) by using the phrase "million bytes" instead of "megabytes."
I never saw them use K=1,000, but I did see them use M=1,000,000 which makes sense in the context of S.I. They could have excusably used k=1,000 but I never saw that either.
I now have a D-Link DWL-G510 802.11g PCI wireless network adaptor working under NetBSD-current.
What module(s) and what else did you need to do?
I didn't use any modules, but I had to upgrade to NetBSD 4 which is not quite released as stable yet (it's in Beta testing). My kernel includes the ath(4) driver and for some reason that I'm not clear about yet, bpf (the Berkeley packet filter) was also required. I have to launch a thing called wpa_supplicant because the wireless network uses WPA, which is supposedly less insecure than WEP. The usual procedure for launching the supplicant didn't work for me, so as a temporary measure I launch it from the rc.local script. Hopefully that will be fixed before 4.0 is released.
It annoys me that some Linux distributions no longer have an rc.local script. There are some applications where creating a full SYSV init script is major overkill.
Do you include SATCOM in the list of wireless networking techniques/technologies? When I deploy in the wake of natural disasters, I am responsible for backhauling unclassified voice and data communications over a satellite link, as well as management and maintenance of the solution. -DTK
Sure! Sounds reasonable to me.
Probably like a less cool version of the rig your brothers took out to the mountains or the sandbox. -DTK
Re. #110: That qualifies. Is that done using VoIP or something else? It's telling that even though years have passed since I asked the question, adding a NetBSD host to a WiFi network is still awkward to the point where I tell people not to bother.
We use VoIP phones, connecting back to a phone switch at the HQ. The phones use lightweight codec-specific signalling, common to both the phones and the switch. Nothing terribly novel.
Is latency much of an issue with the satellite link or is that better with today's LEO birds?
We do not bounce off a LEO bird; they move too much. Instead, we bounce off of a geostationary bird. We set expectations about latency, and people adapt pretty quickly to the latency, as long as it is consistent. Jitter is your big killer, not delay. Oh, and SAA on Cisco gear, or SmokePing on UNIX (or Linux) is absolutely your friend, followed closely by any NetFlow analyzer you can cope with.
I'll ask the packet pushers what SmokePing is.
http://oss.oetiker.ch/smokeping/
resp:115 Latency is almost always an issue for satcom. Pushing on Ka or X band to geo-sync/geo-stationary is going to be slow because, well, the speed of light isn't just a good idea, it's the law. :-) I've never had great luck with LEO for anything. Maybe DAMA voice, but I don't recall what birds those were bitting.
Here here. GEOS is great for resilience, and as long as you can tolerate delay, you can go anywhere in its shadow. That said, it takes people a while to get used to the delays involved in a voice call, but as long as the delay is pretty consistent (i.e. low jitter), people adapt. Oh, and for the fans following along at home, remember that the speed of light in the atmosphere is a lot slower than the speed of light in a vacuum. I never tried using a LEO-provider; having to track a bird that is in motion relative to your frame of reference either requires the dish to be in constant motion, or accept drop-offs frequently. Neither seems like much fun, and not worth the small improvement in round-trip-time. -DTK
Indeed. It's a pain in the butt.
re #114 Back in the stone ages, we used analog phones through a multiplexer over VHF. VOIP is a very specific protocol overhead for packet node sites. I haven't looked at Network44 in years, though.
You have several choices: