yes, I use them and have good results. I was surprised that the review got the worst write performance of any system on them. Really bad. It was very unclear why.
Have you used IDE recently? It doesnt use CPU time. It was only PIO back in the dark ages that created this myth.
For sustained transfers, IDE has a better price/performannce (15K SCSI is marginally faster sometimes, but 2 7200rpm IDE drives are cheaper and faster). For random access workloads SCSI is a little fatser (lower latency) but thoes of us with sequential access workloads dont care.
As for saturating GigE, probably can be done. I usually only need real time 50MB/s and thats fine.
> For #2, realize that some integer operations are > O(N) where N is the number of bits involved. > 64-bit multiplication and division are slower > than the same 32-bit operations. Period.
Not true. They are done in circiuts so you can use parallel algorithms. Mostly arithmetic ops are n log n so they dont take a significant amount of time longer. You do need more transistors though...
After going through the crappy registration process, what do I find: not much at all.
The (code not available) firewire stuff is a fix to allow sharing of firewrire disks. Which has been in the kernel for quite some time (perhaps they submitted it), but it is hardly radical (couple of lines of code, if your hardware happens to support it).
Thearticle was very vague, but you do wonder if the reason they are packing them in so close is ro use PCI on at least some of the interconnects. Many blade servers sit in a (sometimes 64/66) PCI bus per rack (most blade machines are built on PCI cards). You are not going to be able to do it with all the machines in the rack, because as you say you need some switching.
Of course that is where the future Cray Hammer based system wins out as you have hypertransport directly on chip to build these interconnects.
MIPS does have some nice things though, like being available now in very low power with lots of cores per package: eg http://www.broadcom.com/products/1250.html th is has 2 64bit MIPS cores, DDR memory controller, 3 gigabit controllers, hypertransport and 66MHz PCI, and 2 T3 interfaces, for about $500 in volume... and it dissipates 10W.
I think it does (but I need to write some programs using it). You just say I want to start reads async on all these connections, tell me what has finished (which will be the active ones). People on the list seem to be using it for these types of apps (thats why they want aio network io).
If Philips dont sue, then start to expect consumers to sue. In the UK the Trade Descriptions Act should cover passing off a copy protected CD as a red book CD. If everyone starts calling the local trading standards officer the retailers may fight Bertelsmann (as they can be criminally prosecuted for this).
As a customer for embedded products I have to say that the non_GPL drivers do stop me buying products. Having your kernel marked tainted means that no vendor or lkml support is available any more. Unless your product has no competition and is essential to me I am therefore unlikely to buy it.
I dont mind if things like formware are not available, but I do want the api.
If you were to give us some more technical information about what type of product / class of driver perhaps we could help you suggest a way to move parts into userspace in an efficient way.
If anyone actually read the lkml context, the remark was entirely in relation to the flood of recent patches making everything on 32 bit platforms support 64 bit sizes. Once upon a time it was just files over 2GB, then it was block devices over 2TB, now it is all sorts of shit because vendors are selling 32 bit machines that support 64GB of RAM.
Now Intel of course just reckons that people should buy Itaniums if they want this (and apparently they did actually ship 250 of the Itanium 1...) but someone is buying these. Even though you have to use 32 processes in order to use the memory.
Clearly these machines should be 64 bit, thats what Linus was commenting on. Then we could leave at least some of the limits for 32 bit machines without complaints.
The other problem is non-atomic 64 bit ops on 32 bit machines, incrementing counters and such. This has caused quite a few problems recently.
most gps units dont do very accurate time. Well the time is accurate but when it sends it down the serial line isnt. You need pps support which sends a pulse at the time the second actually changes. I use a gpsclock (www.gpsclock.com afaik). It is not as cheap as the $24.95 one though.
Unfortunately, at least in this part of the world, mingetty really is rather rude if you parse it right (ie wrong). And it is rather widespread in Linux distros.
It is of course true that the less appropriate words havent been trademarked yet, so they are available for open source projects.
Had been kind of expecting something like this to happen for a while. ssh (mostly openssh) is becoming a worthwhile target for hacking because it is so widespread, and people are becoming dependent on it. The right answer is to reengineer things so that remote logins are not necessary in so many situations. Even network switches come with ssh now.
Doesnt rreally work if you turn the power off before opening the case. Of course you may have another alarm for this, but it doesnt help you know whether the case was opened.
yes, thats what I wanted, so I have just bought a Psion 5MX on ebay and installed Linux on it. So I can go around with something that I can get batteries for anywhere, and just plug in to a network. I cant type ssh passphrases on the Ipaq...
hmm, no, flash is not a well-designed format. And although there is some openness to the spec, there is not much (like Java, you never know where ti is going next, like adding the Sorenson video codec to the latest version).
WHat is wrong with it? Well, it doesnt have decent generic matrix transforms for a start. SVG is better from this point of view. And there is more once you start to look at it
You can probably get some specifications from manufacturers, but I wouldn't really rely on them (you may for example get 8-bit grey scale, but it may be a non-linear response). WHy not calibrate it yourself: you have got the equipment. Also different models may vary slightly (some manufacturers will change components without telling you). Pay a bit more for good quality equipment, and bear in mind that for example the quality of a ccd camera depends a lot on the lens you use, and it may be better to buy greyscale equipment rather than use colour eqiupment in grey modes.
Also for image capture avoid anything that adds software artefacts (especially compression). firewire uncompressed cameras (we get ours from www.unibrain.gr, very good) are good for high framerate high res, with good Linux support.
Yes, they are very nice. I got a coupel of the 4-way (6410) cards for testing and will get more. As drives are cheaper you can do things you wouldnt do with SCSI, like use RAID 1 instead of RAID5 and get much better performance too. The only annoying thing is that the monitoring tools are not open source and are web based: give me a command line version please! They are supported by recent installers, and appear as a SCSI drive, so you can just disable your motherboard IDE.
Theya re supposed to do hot swap too, but I havent tried yet.
no serial line disciple = need to modprobe serio and serport.
If you want to know the protocol just look at the Linux kernel source. And there is a generic unix driver floating around somewhere. The kernel driver should work fine though.
Slashdot Mirror
yes, I use them and have good results. I was surprised that the review got the worst write performance of any system on them. Really bad. It was very unclear why.
Have you used IDE recently? It doesnt use CPU time. It was only PIO back in the dark ages that created this myth.
For sustained transfers, IDE has a better price/performannce (15K SCSI is marginally faster sometimes, but 2 7200rpm IDE drives are cheaper and faster). For random access workloads SCSI is a little fatser (lower latency) but thoes of us with sequential access workloads dont care.
As for saturating GigE, probably can be done. I usually only need real time 50MB/s and thats fine.
> For #2, realize that some integer operations are
> O(N) where N is the number of bits involved.
> 64-bit multiplication and division are slower
> than the same 32-bit operations. Period.
Not true. They are done in circiuts so you can use parallel algorithms. Mostly arithmetic ops are n log n so they dont take a significant amount of time longer. You do need more transistors though...
After going through the crappy registration process, what do I find: not much at all.
The (code not available) firewire stuff is a fix to allow sharing of firewrire disks. Which has been in the kernel for quite some time (perhaps they submitted it), but it is hardly radical (couple of lines of code, if your hardware happens to support it).
Seems more like a PR announcement to me.
Thearticle was very vague, but you do wonder if the reason they are packing them in so close is ro use PCI on at least some of the interconnects. Many blade servers sit in a (sometimes 64/66) PCI bus per rack (most blade machines are built on PCI cards). You are not going to be able to do it with all the machines in the rack, because as you say you need some switching.
h is has 2 64bit MIPS cores, DDR memory controller, 3 gigabit controllers, hypertransport and 66MHz PCI, and 2 T3 interfaces, for about $500 in volume... and it dissipates 10W.
Of course that is where the future Cray Hammer based system wins out as you have hypertransport directly on chip to build these interconnects.
MIPS does have some nice things though, like being available now in very low power with lots of cores per package: eg
http://www.broadcom.com/products/1250.html
t
I think it does (but I need to write some programs using it). You just say I want to start reads async on all these connections, tell me what has finished (which will be the active ones). People on the list seem to be using it for these types of apps (thats why they want aio network io).
If you have hundreds of thousands of connections you should be using aio, which is the new scalable replacement for lots of polls...
If Philips dont sue, then start to expect consumers to sue. In the UK the Trade Descriptions Act should cover passing off a copy protected CD as a red book CD. If everyone starts calling the local trading standards officer the retailers may fight Bertelsmann (as they can be criminally prosecuted for this).
As a customer for embedded products I have to say that the non_GPL drivers do stop me buying products. Having your kernel marked tainted means that no vendor or lkml support is available any more. Unless your product has no competition and is essential to me I am therefore unlikely to buy it.
I dont mind if things like formware are not available, but I do want the api.
If you were to give us some more technical information about what type of product / class of driver perhaps we could help you suggest a way to move parts into userspace in an efficient way.
usable. I think not. It is a complete pain to use. I only use mine when I need triband.
Nokia make much better user interfaces. And Benefon.
hmm, no in a few years time, the scsi roadmap says they are going to serial scsi. Curiously this is exactly the same as serial ata except:
1. max cable length is officially 3m, sata officically 1m, though in fact 3m should work on both.
2. the command set is scsi not ata6: irrlevent as ata6 was tcq etc.
3. there was going to be multiple devoces on serial scsi I think but they scrapped this as the termination didnt work.
4. er thats it.
So scsi will go anyway.
If anyone actually read the lkml context, the remark was entirely in relation to the flood of recent patches making everything on 32 bit platforms support 64 bit sizes. Once upon a time it was just files over 2GB, then it was block devices over 2TB, now it is all sorts of shit because vendors are selling 32 bit machines that support 64GB of RAM.
Now Intel of course just reckons that people should buy Itaniums if they want this (and apparently they did actually ship 250 of the Itanium 1...) but someone is buying these. Even though you have to use 32 processes in order to use the memory.
Clearly these machines should be 64 bit, thats what Linus was commenting on. Then we could leave at least some of the limits for 32 bit machines without complaints.
The other problem is non-atomic 64 bit ops on 32 bit machines, incrementing counters and such. This has caused quite a few problems recently.
most gps units dont do very accurate time. Well the time is accurate but when it sends it down the serial line isnt. You need pps support which sends a pulse at the time the second actually changes. I use a gpsclock (www.gpsclock.com afaik). It is not as cheap as the $24.95 one though.
Unfortunately, at least in this part of the world, mingetty really is rather rude if you parse it right (ie wrong). And it is rather widespread in Linux distros.
It is of course true that the less appropriate words havent been trademarked yet, so they are available for open source projects.
Will there ever be Linux drivers? I dont care about a bit of performance if there is source code.
Had been kind of expecting something like this to happen for a while. ssh (mostly openssh) is becoming a worthwhile target for hacking because it is so widespread, and people are becoming dependent on it. The right answer is to reengineer things so that remote logins are not necessary in so many situations. Even network switches come with ssh now.
yes. arp poisoning. Unless you have static arp. It works fine.
yes, though it did have a low budget.
You just want digital audio out from your PC (a nice RME sound card is fine). Then an external DA and AD box or boxen...
Doesnt rreally work if you turn the power off before opening the case. Of course you may have another alarm for this, but it doesnt help you know whether the case was opened.
yes, thats what I wanted, so I have just bought a Psion 5MX on ebay and installed Linux on it. So I can go around with something that I can get batteries for anywhere, and just plug in to a network. I cant type ssh passphrases on the Ipaq...
There is no 2GB filesize limit. Hasnt been for some time. I have many files larger than this. Go on, make one yourself with dd and you will see.
hmm, no, flash is not a well-designed format. And although there is some openness to the spec, there is not much (like Java, you never know where ti is going next, like adding the Sorenson video codec to the latest version).
WHat is wrong with it? Well, it doesnt have decent generic matrix transforms for a start. SVG is better from this point of view. And there is more once you start to look at it
You can probably get some specifications from manufacturers, but I wouldn't really rely on them (you may for example get 8-bit grey scale, but it may be a non-linear response). WHy not calibrate it yourself: you have got the equipment. Also different models may vary slightly (some manufacturers will change components without telling you). Pay a bit more for good quality equipment, and bear in mind that for example the quality of a ccd camera depends a lot on the lens you use, and it may be better to buy greyscale equipment rather than use colour eqiupment in grey modes.
Also for image capture avoid anything that adds software artefacts (especially compression). firewire uncompressed cameras (we get ours from www.unibrain.gr, very good) are good for high framerate high res, with good Linux support.
Yes, they are very nice. I got a coupel of the 4-way (6410) cards for testing and will get more. As drives are cheaper you can do things you wouldnt do with SCSI, like use RAID 1 instead of RAID5 and get much better performance too. The only annoying thing is that the monitoring tools are not open source and are web based: give me a command line version please! They are supported by recent installers, and appear as a SCSI drive, so you can just disable your motherboard IDE.
Theya re supposed to do hot swap too, but I havent tried yet.
no serial line disciple = need to modprobe serio and serport.
If you want to know the protocol just look at the Linux kernel source. And there is a generic unix driver floating around somewhere. The kernel driver should work fine though.