TCP/IP vs. FireWire, USB2, and other interfaces

[onix]

I still cannot understand why there are so many interfaces for peripherals, when TCP/IP over Cable/10B-T/100B-T & now 1000B-T has been around much longer and is capable of delivering so much data so much faster. Why, why, why?

Why can't the computer act as a router for all it's peripherals ... instead of using BlueTooth, we'd be using 802.11x and not deal with the headache of all these new interfaces....

 

[Lynx516]

A few things:

You cannot produce drivers easily for a ethernet based joystick. And who would use it? It would be accessible to anyone on the network. Some times the hastle of producing more programming to work around problems such as the one I just shown is not worth it. Also ethernet cost more, Both USB and Firewire can be in one chip. Look at your NIC that has a few more chips on.

You canactually use TCP/IP through firewire btw.

The basic reason is you dont need the data to come in faster becasue what you have is fast enough. Why do we need an 802.3 based mouse?

 

[onix]

Originally posted by: Lynx516
A few things:

You cannot produce drivers easily for a ethernet based joystick.

You can't? Why?

And who would use it? It would be accessible to anyone on the network.

Not if the computer hosted it's own subnet, and acted as a router.

Some times the hastle of producing more programming to work around problems such as the one I just shown is not worth it.

I can't imagine it being more costly then (1) developing a new standard, (2) prototyping & implementing it, (3) finding a market for it, (4) building infrastructure for making it in large quantites, (5) technical support, etc.

Also ethernet cost more, Both USB and Firewire can be in one chip. Look at your NIC that has a few more chips on.

Ethernet costs more? I could go to a scraps dealer and pick up and old NIC for super cheap, even on PCMCIA, new it costs less than $30, and that might include a modem, etc.

You canactually use TCP/IP through firewire btw.

The basic reason is you dont need the data to come in faster becasue what you have is fast enough. Why do we need an 802.3 based mouse?

Speed would benefit hard-drive speed access, for example. But that's not the real reason. The real reason is simplicity, commonality and standardization.

 

[Lynx516]

OK if the pc acted as a router that would require another layer of programming. You say a NIC is cheap $30 but a USB controller is about $8 for a company to implement. Take a mouse it will have a USB controller imbeded into its microcontroller and the SUB will have 4 pins on the packaging. a MAC for ethernet will posess around 30 as you have to implement a PHY as well which cannot be done on board the package.

You say it would be simpler. But ethernet is not a simple standard by anyones definition (the definiation is about 1500 pages long). You have to deal with so much more than you would with USB. Howmany USB devices out there require you to have large enough volumes of data to saturate a 10Mbit ethernet link let alone a 1Gigabit connection.

Standardiztaion is not always a good thing. There is somethign called overkill

 

[Gannon]

You'd still need a rather large hub if you were to connect all your devices to ethernet, and ethernet hubs/switches have not been all that inexpensive in the past. When I got my 100mbit switch it was $170(CAD). Either way plug and play via ethernet would be a hassle since it was not designed with "ease of use" in mind. I like how I can just plug in things to USB and firewire and have them work straight away with no rebooting or whatever.

I think there are solid reasons why we have so many interfaces... much of it is because people can't agree to a standard and theres more then one way of solving a problem. Being that the PC industry is full of geeks/engineers who want to make stuff as much for the sake of it as for other real "killer app" type of practical purposes I dont think its hard to see why.

The huge issue was bandwidth improvements over time and ease of use/connectivity for the user end peripherals. I believe thats why Firewire and USB were developed rather then trying to piggyback on ethernet standards (which take quite a while to emerge). The original firewire is 400Mbit of bandwidth thats like 4X 100Mbit ethernet that most people have, and the new 1394b is 800Mbit thats ~7x faster then 100base ethernet (in theory anyway). Also USB 2.0 has higher bandwidth then 100Mbit ethernet at 480Mbit. So I think your point is pretty much useless since both firewire and USB 2.0 are backwards compatable and are higher bandwidth then even lowly 100Mbit ethernet. Also USB and firewire are standard on many peripherals, even my audigy, and ASUS P4P800 included a firewire port at negligable cost to me. I also like having tonnes of USB ports for peripherals included on the motherboard, it makes my life that much easier.

 

[klah]

PCIe will be competing in some overlapping areas:

http://www1.la.dell.com/conten...=hn&l=es&s=gen

The PCI SIG is also working on the PCI Express cable specification. Because PCI Express has high data rates and low-pin-count connectors, it is likely to be used as a high-speed interconnect between components in client and server systems. Modular systems with separate high-speed components can be connected with PCI Express cables. Figure 14 illustrates the concept of a "split" system that separates components that generate heat such as processors, memory, and graphics from other components such as removable storage, display devices, and I/O ports. It may also make sense to separate high-end graphics subsystems, which require more power and generate heat, from the main processor chassis. This approach would make it easier to deliver appropriate power and cooling to the graphics subsystem.

 

[onix]

Originally posted by: Gannon
I think there are solid reasons why we have so many interfaces... much of it is because people can't agree to a standard and theres more then one way of solving a problem. Being that the PC industry is full of geeks/engineers who want to make stuff as much for the sake of it as for other real "killer app" type of practical purposes I dont think its hard to see why.

I think Gannon is on the right track - probably because I feel the same way. 😀

As for an easier port interface, nothing suggests we cannot use better plugs than the 8-pin RJ45's. The blasted parallel port (with 25-pins) on the back of my laptop takes up as much space as my ethernet, modem and 2 USB ports combined.

As for chip costs, it's a matter of volume. I am in the sensors industry, and we can sell (in low volume) sensors for $200-$1000 which are at least two orders of magnitude easier to fab than any of today's CPU. In high volume and with standardization, these costs can come down significantly. I would say IEEE is not doing its job to standardize. (Aside to Lynx516: Standardization is not overrated, especially in this case.)

Ethernet lines are hardware, but protocols which are implemented on it are many, e.g. TCP/IP, Microsoft Networking, EtherTalk. And software implementation comes for cheap these days... It's possible I'm missing something here, but someone did say we can now run TCP/IP over USB, now isn't that sayin something about the complexity of USB? By the way, 100B-T is not the fastest any more. It's gigabit LAN (1000B-T). So even then, we have faster transfer speeds than the latest FireWire and USB.

Once speeds get superfast, we'll have to go to fiber optics or perhaps, co-ax -- now wouldn't that be a take back!

 

[imported_devnull]

Xmm, I think you are oversimplifying things. Don't stand behind your point and try to defend it no matter what.

USB, FireWire, Bluetooth blah blah all have a purpose to exist and it is not really easy to find it out if you have no real knowledge of the purpose they were created or what was the application that drove their development. This certainly includes marketing!!

For example USB is NOT a networking standard. So you don't have for example to implement a whole protocol stack in order to use a device like that. On the other hand, it is designed to connect peripheral devices to a host device and as a result you may see that it has certain features that the ethernet does not. For example it carries power through the cable (please don't quote that you can have power over ethernet it is not the same and it costs much much more), you have cheaper and smaller cables, you have a more dummy protocol stack that is much easier and less costy to implement and so on.

FireWire is a network standard that has been invented in order to provide a medium to create a network without a central host that could include all kind of home appliances, from TVs and VCRs to camcorders and High-End Stereo equipment (for example a standard has been developped in order to carry the DVD-Audio high resolution signals digital data from a player to a receiver). In these multimedia environments the standard has been designed so as to fulfill specific requirements - low and guaranteed latency for example that an ethernet based network cannot guarantee unless it implements "expensive" (in terms of complicacy) QoS schemes.

 

[IEC]

Originally posted by: Lynx516
OK if the pc acted as a router that would require another layer of programming. You say a NIC is cheap $30 but a USB controller is about $8 for a company to implement. Take a mouse it will have a USB controller imbeded into its microcontroller and the SUB will have 4 pins on the packaging. a MAC for ethernet will posess around 30 as you have to implement a PHY as well which cannot be done on board the package.

You say it would be simpler. But ethernet is not a simple standard by anyones definition (the definiation is about 1500 pages long). You have to deal with so much more than you would with USB. Howmany USB devices out there require you to have large enough volumes of data to saturate a 10Mbit ethernet link let alone a 1Gigabit connection.

Standardiztaion is not always a good thing. There is somethign called overkill

Agreed, Ethernet is for networking and TCP/IP

Firewire and USB are so much cheaper and easier to implement for peripherals, let's keep it that way.

 

[AndyKH]

Also, as bluetooth is designed for small, mobile devices, the power requirements are lower than that of 802.11b for example.

 

[Abix]

Originally posted by: onix
Speed would benefit hard-drive speed access, for example. But that's not the real reason. The real reason is simplicity, commonality and standardization.

Incorrect. The raw ability to achieve 1000MBit/s wouldnt help a damn with hard-drive access speed. The hard-drive has to internally spin its platters to get at the data, and as it is even with UltraATA 133 which has a 133MB/sec max transfer speed, the most that any single drive uses is usually ~50MB/sec. Nowhere near the full potential of the "mere" 133MB/sec. In order to benefit hard-drive speed access, you would need to make more efficient mechanical parts inside the hard-drive.

Let the individual computer fields develop their own standards for what they need. If you want to make it all "standardized," feel free to try, but be prepared for dissapointment. Also, as multiple people have said, its not as easy as you would make it out to be.

 

[irwincur]

Another question... Why don't they just use USB2 or Firewire internally in a computer for peripheral hook up? Would be easier (and probably faster/cheaper) than SATA or IDE. I have wondered for years why externally USB2 is good enough for an HDD, but not internally. I have also wondered why companies invested so much time and money into the development of SATA when multiple viable options already existed.

 

[Peter]

IDE has survived because it IS the cheapest method of getting data to the HDD. External USB2 drives have a USB-IDE translator chip inside, which costs money.

The answer is the exact same as to your previous question: As soon as you're starting to stack up protocol layers and make "easy to use" physical interfaces, you need extra software and extra interface hardware. This is neither cheaper nor faster than the established direct-connect interfaces.

SATA's beauty lies in the software compatibility to traditional IDE, maintaining the simple, short-latency, direct register access, point-to-point interface between the core system and the drive's controller. Talking through a high level protocol like USB, FW, or whatever, makes things slower and more expensive. Way to go, huh?

 

[DoubleE]

Another point about Ethernet (not so much TCP/IP), is that the CSMA/CD algorithm in 802.3 causes it to get much less efficient with a significant number of peers on the bus. Obviously, this can be mitigated by using some sort of central hub, but in this configuration, why not use a light-weight implementation (like USB)?

I can see where you're coming from, though - hardware 'standards' seem to be arbitrary at times. But, it is probably important to note that TCP/IP is one of the great successes in 'standards' history. It was not all that long ago that communicating from TokenRing to ATM was a big deal (or it was supposed to become a big deal). Remeber good 'ol NetBIOS for Windows machines? I believe Apple had their own proprietary protocol as well (AppleTalk?). Rarely has a protocol been more succesfully adopted than TCP/IP. It just doesn't make a lot of sense for peripherals (as others have illustrated).