This is an optical switch - it's best IMO to think of it as a layer 2 switch connecting a number of different routers with very fast 'logical wires' - each link will use a number of wavelengths, and the switch presumably converts stuff from one wavelength to another (though if it's all mirrors maybe they have to stay on the same wavelength).
There is some other work using tunable laser transmitters and corresponding receivers, which lets you set up a LAN-like structure on DWDM - this seems more like a L2 switch, analogous to the way ATM virtual channels can be used to connect IP routers.
The world will still need very fast IP routers, though - if you have too many IP routers connected via a DWDM cloud, you run into problems with the routing process in the router 'seeing' too many neighbours and thereby working inefficiently (particularly if an interface somewhere starts going up and down frequently - known as route flapping).
The real issue for IP usage is how on earth do you build routers fast enough to keep up with all the wavelengths pouring out of a single fibre? Terabit routers will address this through massive parallelism, but they don't come cheap.
Most people in the Internet world use router to mean something that switches packets at layer 3.
Almost everything else is called a switch, so I think this should be called a DWDM switch, since it is working at layer 2 with DWDM (or maybe layer 1 depending on how you look at it). (DWDM = Dense Wavelength Division Multiplexing, or shoving several colours of light down a single fibre.)
This sort of kit will be very useful though - before too long it should be possible to run DWDM to the customer premises, but even before that heady prospect DWDM should make it possible for xDSL and cable modems to run closer to full speed.
All we need now is faster servers and 10-gigabit ethernet (which a standards group is working on I think).
I thought this was a pretty fair review - I had a similar experience installing OpenBSD, though most of my install hassles were related to getting X working.
As a mainly Linux user who's used Unix a lot in the past, I found OpenBSD very interesting, and I am planning to use it as the basis for a firewall, where its relatively small ports collection is a feature not a problem. My theory is that script kiddies & others will be less likely to have exploits for OpenBSD, due to its smaller user base and more stringent auditing.
I'm not a Firewall-1 expert but it does have stateful inspection - as with ipmasq modules for Linux that support ftp, realaudio, etc, stateful inspection looks at each flow to extract port numbers etc, then opens up appropriate ports for the corresponding data flows.
Stateful inspection can be defined for new protocols by writing a relatively simple script - quite a bit easier than writing an ipmasq module.
Firewall-1 also has quite sophisticated NAT facilities that can do static NAT, not just dynamic NAT. Last time I looked at Linux NAT there seemed to be quite a few packages for it, none of which seemed to be 'the one'.
Firewall-1 is also quite well packaged, with a decent GUI for viewing and modifying rules. Although it's a pretty complex product, it is well integrated and tested as a whole.
There are quite a few companies out there who prefer Solaris to NT for Firewall-1, and will no doubt jump at the chance to use Linux. Security gurus would probably be even happier if it used OpenBSD, but that doesn't have the same market share as Linux.
For those of you who want to run BSD on top of Linux or NT, have a look at VMware, www.vmware.com - FreeBSD is supported and OpenBSD has worked at least for some people.
Why is this useful? I'm intending to try out OpenBSD as a firewall, largely because of its security focus - script kiddies will be less familiar with it and there should be fewer exploits for it. VMware is one way of running a client-side firewall, particularly on laptops.
VMware is a good way to play with different OSs/distros - I've had Debian, NT, Win98 all running on top of Red Hat 5.2.
You might like to look at IBM's AS/400, and its predecessor System/38 - both have orthogonal persistence, i.e. everything is an object and some are just persistent.
Despite its proprietary nature, the AS/400 has an interesting architecture, and is highly popular as a mid-range system.
The rise of web-based applications (e.g. www.opendesk.com, an open sourced web desktop, and desktop.com, as well as the handy web-based database, www.flashbase.com) may make it much easier for radically different OS architectures to flourish. It also makes it easier for older architectures to do well - e.g. IBM's OS/390 mainframes remain a good way to do truly enormous amounts of back-end transactional processing, and they can run web applications just like anything else.
To paraphrase the New Yorker cartoon, "on the Web, nobody knows you're an AS/400."
VMware is certainly not slow for me - if it's slow for you, you most probably have too little memory. It's not too good for running games, but for running office type apps, it's excellent. I run it on a K6/2-350 and in practical terms it's as fast as running natively.
If you have performance problems, post them to the vmware.* newsgroups, and if they don't help, submit an incident report.
You can still buy Maltrons, details at http://www.maltron.com - not cheap but they are incredibly comfortable to use, once you adapt, and will work with PCs and Macs and maybe other systems.
Maltrons can use either the Qwerty or Maltron layout - the main benefit IMO comes from sculpting the keyboard so your hands rest in two separate wells with keys right at your finger tips. I have two Maltrons, one at home and one at work - both used in Qwerty mode.
Richard
Re:Be needs a hardware vendor
on
Be on the G4
·
· Score: 1
At least in the UK, Digital Networks (http://www.dnuk.com/) ship BeOS pre-installed on x86 boxes. They also do Linux - in fact I got an NT/Linux/BeOS box pre-installed from them in January, though this was a special setup I think.
Re:TELL ME ABOUT IT!!!!!
on
DVD for Linux
·
· Score: 1
Reminds me of May this year in Las Vegas, during Interop Week. As you may know, the casinos have these huge colour displays pushing their dubious attractions - one of them was displaying the Windows Login screensaver - i.e. it had crashed, presumably auto-rebooted (which is possible in NT), and got stuck at the login prompt.
The big difference is that Iridium was based entirely around low-bandwidth, high-cost mobile voice connections, while optical internetworking is aligned with the trend to converged voice+data networks, and for my money is a good investment. Not sure about the $7bn price tag, but it could well turn out to be reasonable if Cisco sells enough optical kit based on this technology.
Not sure what you mean by gigabit TCP stack - is this using standard TCP extensions such as window scaling and SACKs to handle long fat pipes (high bandwidth*delay product)? I'd have thought that optical networks have very low delay, so maybe this is something else.
At the IP level, I think Linux can go at gigabit ethernet speeds in any case, I recall some test where it beat NT. There are definitely several gigabit ethernet cards available for Linux.
[Having looked around on the Net...] Here's the URL for Trapeze - http://www.cs.duke.edu/ari/manic/tpz_www/trapeze_r oot.html - it seems to be layer 1 / 2 messaging software on BSD for a network called Myrinet - it doesn't include any IP stuff at all, and an unmodified BSD IP stack seems to perform pretty well - latency is in the 100-200 microsecond range, so TCP extensions are probably not necessary.
Quite a few laptops do this, e.g. IBM Thinkpads and no doubt most other name brands. I think there's someone working on 'snapshot to disk' equivalent functionality for Linux, but I don't have a URL.
What happened is that up to NT 3.51 inclusive, the Win32 subsystem was a completely separate process, as were the OS/2 and POSIX subsystems. In NT4, the Win32 subsystem was migrated into the kernel (in Linux speak; in NT it's the Executive, the Kernel is a very low level layer just above the HAL). There are rumours that this was resisted by key people on the NT team, but there was a strong push to improve GUI performance; there are also rumours that this is the reason for NT4 being less stable than NT 3.5x.
Certainly I never had a crash on NT 3.5x and have had many on NT4, but YMMV. On some configurations, NT does not really crash much (e.g. my Linux workstation on top of VMware!).
As for the HAL originally hiding all hardware - this is not true, otherwise why would device drivers be necessary? I think the HAL abstracts basic resources such as CPU, memory, bus, etc, but many other resources require drivers.
I'm not hugely surprised by Linux doing better on IA-64, since Linux has been 64-bit for some time and has worked out a lot of generic 64-bitness bugs, whereas Win64 is still alpha.
Right now, you might want to look at a sub-notebook with a built-in hard disk, or one of those tiny palmtops with hard disk. Although if your books fit on a Compact Flash card (say 32 or 64MB, not sure how big these are now) you might want to look at a Psion Series 5 or a WinCE machine. Palms don't support add-in Flash memory.
O'Reilly already do some CDs with 6 or so books on them, have a look at their website. Not sure of the format but I'd guess it's work on *nix since they are very supportive of open source etc.
I just read a Robert Silverberg book (from Peanut Press) on the Palm, and it was fine - as long as the light's good, it's fine. The print is about the same size as a magazine, which is what matters.
Most importantly, the book weighed nothing - in future I think I'll just download a few novels before I go on a trip, and cut the weight of my bag immensely!
Reading websites (via Avantgo.com) is also great for short Tube (metro) journeys in London, where in the rush hour you can be crammed in so tight there's no room to read a newspaper - here, the small size is an advantage.
My only concern is that there needs to be standards that are adopted by everyone - sounds like the eBook standard may help, but I don't know much about this.
From an article about Win2000, it appeared that only about 70% of Windows NT4 apps would run OK on Win2000 - allegedly MS has put stability as a higher priority than backward compatibility in this release. However, all previous releases of Windows have been highly upward and cross compatible - there is no need to modify source code much, if at all, as long as it was written right in the first place, to port to a different version of Windows (excluding WinCE, which is a smaller subset of Win32).
Cisco would be within their rights to charge for the features - as it happens, they don't. QoS features have been around in Cisco's for quite a few years. It's just that they're a touch hard to configure correctly by hand, particularly in a large network.
Limits have the great advantage that you only need to limit in one place, whereas guarantees need to be done end to end.
IPv4 was standardised in RFC 791, which defines 6 bits of priority information (3 bits precedence, 3 bits type of service) in what was known as the TOS byte, and is becoming known as the DiffServ field.
IPv6 is renaming this field to Traffic Class, I think, but adding no new features - DiffServ will work fine on IPv6 as well. The only new QoS feature in IPv6, IMO, is the Flow Label, which is a longer field used to quickly classify individual flows (e.g. a VoIP call) in RSVP (which is a finer-grained way of requesting absolute bandwidth/latency, rather than just 'better treatment' a la DiffServ.)
Some links on QoS and DiffServ: - Linux-DiffServ - working code for 2.2+ kernels: http://lrcwww.epfl.ch/linux-diffserv/ - QoS Forum - general information: http://www.qosforum.com/ - Orchestream Links page - http://www.orchestream.com - has links on RSVP, DiffServ, etc.
'Ethically questionable'... OK, how about if the same bandwidth limits are used to ensure that one cable modem user doesn't consume all the available bandwidth for 200 homes on a given segment? To make this pointed, what if this bandwidth limitation makes it possible for VoIP packets carrying an emergency services (911, 112) call to be carried over the same net? (I know people won't use cable phones for emergencies, it's a hypothetical example!)
The point is that quality of service and class of service technology can be used to guarantee or to limit bandwidth to or from any IP address, with any IP protocol or TCP/UDP port, etc - it's down to the user of these features (and your perspective and values of course) as to whether the resulting effect is morally good or bad.
Astonishingly, this is exactly like IP itself, which can carry hate material just as easily as charity donations...
The technology's morally neutral - for example, CAR is used to limit bandwidth consumed by Smurf attacks (presumably a good thing), but this same technology is (or could be) used to limit bandwidth to certain unaffiliated websites.
This is not really very different morally to using firewalls and access lists - you can use them to block access to certain websites (probably bad) or to block access by crackers to your domain (probably good, unless you are a cracker:).
It seems to me that a prime reason for putting Linux onto an RS/6000 is access to the emerging market for PPC Linux apps - while there are nowhere near as many as Intel Linux, they are coming along (even games, as recently announced). And porting from Linux/x86 to Linux/PPC should be pretty easy if the code's been written properly.
Probably the biggest advantage is access to skills - administrators, developers, etc.
It's worth pointing out that many ADSL deployments seem to run ATM PVCs from the DSLAM (DSL box at the CO) to the ISP - the PCR is the Peak Cell Rate (i.e. max burst rate) that you'll get, and the SCR is the Sustained Cell Rate that defines the guaranteed minimum rate (like frame relay Committed Information Rate).
For more information, see the ATM Forum or ADSL Forum websites.
Slashdot Mirror
This is an optical switch - it's best IMO to think of it as a layer 2 switch connecting a number of different routers with very fast 'logical wires' - each link will use a number of wavelengths, and the switch presumably converts stuff from one wavelength to another (though if it's all mirrors maybe they have to stay on the same wavelength).
There is some other work using tunable laser transmitters and corresponding receivers, which lets you set up a LAN-like structure on DWDM - this seems more like a L2 switch, analogous to the way ATM virtual channels can be used to connect IP routers.
The world will still need very fast IP routers, though - if you have too many IP routers connected via a DWDM cloud, you run into problems with the routing process in the router 'seeing' too many neighbours and thereby working inefficiently (particularly if an interface somewhere starts going up and down frequently - known as route flapping).
The real issue for IP usage is how on earth do you build routers fast enough to keep up with all the wavelengths pouring out of a single fibre? Terabit routers will address this through massive parallelism, but they don't come cheap.
Most people in the Internet world use router to mean something that switches packets at layer 3.
Almost everything else is called a switch, so I think this should be called a DWDM switch, since it is working at layer 2 with DWDM (or maybe layer 1 depending on how you look at it). (DWDM = Dense Wavelength Division Multiplexing, or shoving several colours of light down a single fibre.)
This sort of kit will be very useful though - before too long it should be possible to run DWDM to the customer premises, but even before that heady prospect DWDM should make it possible for xDSL and cable modems to run closer to full speed.
All we need now is faster servers and 10-gigabit ethernet (which a standards group is working on I think).
I thought this was a pretty fair review - I had a similar experience installing OpenBSD, though most of my install hassles were related to getting X working.
As a mainly Linux user who's used Unix a lot in the past, I found OpenBSD very interesting, and I am planning to use it as the basis for a firewall, where its relatively small ports collection is a feature not a problem. My theory is that script kiddies & others will be less likely to have exploits for OpenBSD, due to its smaller user base and more stringent auditing.
I'm not a Firewall-1 expert but it does have stateful inspection - as with ipmasq modules for Linux that support ftp, realaudio, etc, stateful inspection looks at each flow to extract port numbers etc, then opens up appropriate ports for the corresponding data flows.
Stateful inspection can be defined for new protocols by writing a relatively simple script - quite a bit easier than writing an ipmasq module.
Firewall-1 also has quite sophisticated NAT facilities that can do static NAT, not just dynamic NAT. Last time I looked at Linux NAT there seemed to be quite a few packages for it, none of which seemed to be 'the one'.
Firewall-1 is also quite well packaged, with a decent GUI for viewing and modifying rules. Although it's a pretty complex product, it is well integrated and tested as a whole.
There are quite a few companies out there who prefer Solaris to NT for Firewall-1, and will no doubt jump at the chance to use Linux. Security gurus would probably be even happier if it used OpenBSD, but that doesn't have the same market share as Linux.
For those of you who want to run BSD on top of Linux or NT, have a look at VMware, www.vmware.com - FreeBSD is supported and OpenBSD has worked at least for some people.
Why is this useful? I'm intending to try out OpenBSD as a firewall, largely because of its security focus - script kiddies will be less familiar with it and there should be fewer exploits for it. VMware is one way of running a client-side firewall, particularly on laptops.
VMware is a good way to play with different OSs/distros - I've had Debian, NT, Win98 all running on top of Red Hat 5.2.
You might like to look at IBM's AS/400, and its predecessor System/38 - both have orthogonal persistence, i.e. everything is an object and some are just persistent.
Despite its proprietary nature, the AS/400 has an interesting architecture, and is highly popular as a mid-range system.
The rise of web-based applications (e.g. www.opendesk.com, an open sourced web desktop, and desktop.com, as well as the handy web-based database, www.flashbase.com) may make it much easier for radically different OS architectures to flourish. It also makes it easier for older architectures to do well - e.g. IBM's OS/390 mainframes remain a good way to do truly enormous amounts of back-end transactional processing, and they can run web applications just like anything else.
To paraphrase the New Yorker cartoon, "on the Web, nobody knows you're an AS/400."
VMware is certainly not slow for me - if it's slow for you, you most probably have too little memory. It's not too good for running games, but for running office type apps, it's excellent. I run it on a K6/2-350 and in practical terms it's as fast as running natively.
If you have performance problems, post them to the vmware.* newsgroups, and if they don't help, submit an incident report.
Errr, I think he was kidding - in fact the whole article is an extended joke, though I do like the idea!
You can still buy Maltrons, details at http://www.maltron.com - not cheap but they are incredibly comfortable to use, once you adapt, and will work with PCs and Macs and maybe other systems.
Maltrons can use either the Qwerty or Maltron layout - the main benefit IMO comes from sculpting the keyboard so your hands rest in two separate wells with keys right at your finger tips. I have two Maltrons, one at home and one at work - both used in Qwerty mode.
Richard
At least in the UK, Digital Networks (http://www.dnuk.com/) ship BeOS pre-installed on x86 boxes. They also do Linux - in fact I got an NT/Linux/BeOS box pre-installed from them in January, though this was a special setup I think.
Only if you have a multi-region DVD player...
Reminds me of May this year in Las Vegas, during Interop Week. As you may know, the casinos have these huge colour displays pushing their dubious attractions - one of them was displaying the Windows Login screensaver - i.e. it had crashed, presumably auto-rebooted (which is possible in NT), and got stuck at the login prompt.
Unfortunately I didn't have a camera with me...
The big difference is that Iridium was based entirely around low-bandwidth, high-cost mobile voice connections, while optical internetworking is aligned with the trend to converged voice+data networks, and for my money is a good investment. Not sure about the $7bn price tag, but it could well turn out to be reasonable if Cisco sells enough optical kit based on this technology.
Not sure what you mean by gigabit TCP stack - is this using standard TCP extensions such as window scaling and SACKs to handle long fat pipes (high bandwidth*delay product)? I'd have thought that optical networks have very low delay, so maybe this is something else.
r oot.html - it seems to be layer 1 / 2 messaging software on BSD for a network called Myrinet - it doesn't include any IP stuff at all, and an unmodified BSD IP stack seems to perform pretty well - latency is in the 100-200 microsecond range, so TCP extensions are probably not necessary.
At the IP level, I think Linux can go at gigabit ethernet speeds in any case, I recall some test where it beat NT. There are definitely several gigabit ethernet cards available for Linux.
[Having looked around on the Net...] Here's the URL for Trapeze - http://www.cs.duke.edu/ari/manic/tpz_www/trapeze_
Quite a few laptops do this, e.g. IBM Thinkpads and no doubt most other name brands. I think there's someone working on 'snapshot to disk' equivalent functionality for Linux, but I don't have a URL.
I'm no NT expert, but this is confused.
What happened is that up to NT 3.51 inclusive, the Win32 subsystem was a completely separate process, as were the OS/2 and POSIX subsystems. In NT4, the Win32 subsystem was migrated into the kernel (in Linux speak; in NT it's the Executive, the Kernel is a very low level layer just above the HAL). There are rumours that this was resisted by key people on the NT team, but there was a strong push to improve GUI performance; there are also rumours that this is the reason for NT4 being less stable than NT 3.5x.
Certainly I never had a crash on NT 3.5x and have had many on NT4, but YMMV. On some configurations, NT does not really crash much (e.g. my Linux workstation on top of VMware!).
As for the HAL originally hiding all hardware - this is not true, otherwise why would device drivers be necessary? I think the HAL abstracts basic resources such as CPU, memory, bus, etc, but many other resources require drivers.
I'm not hugely surprised by Linux doing better on IA-64, since Linux has been 64-bit for some time and has worked out a lot of generic 64-bitness bugs, whereas Win64 is still alpha.
Right now, you might want to look at a sub-notebook with a built-in hard disk, or one of those tiny palmtops with hard disk. Although if your books fit on a Compact Flash card (say 32 or 64MB, not sure how big these are now) you might want to look at a Psion Series 5 or a WinCE machine. Palms don't support add-in Flash memory.
O'Reilly already do some CDs with 6 or so books on them, have a look at their website. Not sure of the format but I'd guess it's work on *nix since they are very supportive of open source etc.
I just read a Robert Silverberg book (from Peanut Press) on the Palm, and it was fine - as long as the light's good, it's fine. The print is about the same size as a magazine, which is what matters.
Most importantly, the book weighed nothing - in future I think I'll just download a few novels before I go on a trip, and cut the weight of my bag immensely!
Reading websites (via Avantgo.com) is also great for short Tube (metro) journeys in London, where in the rush hour you can be crammed in so tight there's no room to read a newspaper - here, the small size is an advantage.
My only concern is that there needs to be standards that are adopted by everyone - sounds like the eBook standard may help, but I don't know much about this.
From an article about Win2000, it appeared that only about 70% of Windows NT4 apps would run OK on Win2000 - allegedly MS has put stability as a higher priority than backward compatibility in this release. However, all previous releases of Windows have been highly upward and cross compatible - there is no need to modify source code much, if at all, as long as it was written right in the first place, to port to a different version of Windows (excluding WinCE, which is a smaller subset of Win32).
Cisco would be within their rights to charge for the features - as it happens, they don't. QoS features have been around in Cisco's for quite a few years. It's just that they're a touch hard to configure correctly by hand, particularly in a large network.
Limits have the great advantage that you only need to limit in one place, whereas guarantees need to be done end to end.
IPv4 was standardised in RFC 791, which defines 6 bits of priority information (3 bits precedence, 3 bits type of service) in what was known as the TOS byte, and is becoming known as the DiffServ field.
IPv6 is renaming this field to Traffic Class, I think, but adding no new features - DiffServ will work fine on IPv6 as well. The only new QoS feature in IPv6, IMO, is the Flow Label, which is a longer field used to quickly classify individual flows (e.g. a VoIP call) in RSVP (which is a finer-grained way of requesting absolute bandwidth/latency, rather than just 'better treatment' a la DiffServ.)
Some links on QoS and DiffServ:
- Linux-DiffServ - working code for 2.2+ kernels:
http://lrcwww.epfl.ch/linux-diffserv/
- QoS Forum - general information: http://www.qosforum.com/
- Orchestream Links page - http://www.orchestream.com - has links on RSVP, DiffServ, etc.
'Ethically questionable'... OK, how about if the same bandwidth limits are used to ensure that one cable modem user doesn't consume all the available bandwidth for 200 homes on a given segment? To make this pointed, what if this bandwidth limitation makes it possible for VoIP packets carrying an emergency services (911, 112) call to be carried over the same net? (I know people won't use cable phones for emergencies, it's a hypothetical example!)
The point is that quality of service and class of service technology can be used to guarantee or to limit bandwidth to or from any IP address, with any IP protocol or TCP/UDP port, etc - it's down to the user of these features (and your perspective and values of course) as to whether the resulting effect is morally good or bad.
Astonishingly, this is exactly like IP itself, which can carry hate material just as easily as charity donations...
The technology's morally neutral - for example, CAR is used to limit bandwidth consumed by Smurf attacks (presumably a good thing), but this same technology is (or could be) used to limit bandwidth to certain unaffiliated websites.
:).
This is not really very different morally to using firewalls and access lists - you can use them to block access to certain websites (probably bad) or to block access by crackers to your domain (probably good, unless you are a cracker
It seems to me that a prime reason for putting Linux onto an RS/6000 is access to the emerging market for PPC Linux apps - while there are nowhere near as many as Intel Linux, they are coming along (even games, as recently announced). And porting from Linux/x86 to Linux/PPC should be pretty easy if the code's been written properly.
Probably the biggest advantage is access to skills - administrators, developers, etc.
It's worth pointing out that many ADSL deployments seem to run ATM PVCs from the DSLAM (DSL box at the CO) to the ISP - the PCR is the Peak Cell Rate (i.e. max burst rate) that you'll get, and the SCR is the Sustained Cell Rate that defines the guaranteed minimum rate (like frame relay Committed Information Rate).
For more information, see the ATM Forum or ADSL Forum websites.