UNIX for Advanced & Expert Users

Hi all
I know AIX is big-endian machine.But does it read bytes in normal way from LSB.


Does it happen in some machine that at multi-byte integer level it is Little-endian and while reading a single byte it is Big-Endian.
This is urgent
Thanks in advance.

Endian-ness in a cpu arises because we need to specify multi-byte integers with a byte address.

In the '60's I worked on an IBM 1130. It was a 16 bit machine and had no byte addressing. Address 0 was a 16 bit word. Address 1 was the next 16 bit word. There is no way to determine an endian-ness on a system like this.

I have never heard of any cpu that even allows bit addressing. If there was one, and if it required the programmer to specify a bit address for a byte, then we would need to worry about the endian-ness of a single byte.

The only time that byte endian-ness becomes apparent is when a byte is transmitted across a serial data line. TCP/IP is an example of this, and in TCP/IP, bytes are big-endian. (or more accurately, octets are big-endian, since TCP/IP does not assume that bytes are 8 bits). Contrary to your comment, this is what I regard as "normal". Most other serial protocols are also big-endian, at least most of the modern ones are.

On the other hand, RS-232 is little-endian. So is the old current-loop interface. The earlier teletype protocols are before my time, but I think that they were little-endian (but I'm not sure). These protocols never need to transit a multi-byte integer and they need to support both 7 bit and 8 bit characters.

So unless you are designing an i/o card to transmit bytes in a bit serial fashion, you can safely ignore this issue with every cpu that I have heard of.

But it's hard to be sure that something doesn't exist, so if someone out there knows about a cpu that requires a bit address for addressing its bytes, I would be very interested in the details of it.

Hi Perderabo
I am basically trying to read a radar message following ASTERIX format(which is bit oriented) which comes on HDLC(little Endian) line.Now I have to put the bit stream coming into standard structures as per protocol.This decoding has to been implemented on Compaq(little endian) and i have to implement rather port it on HP(big endian).Since there is relative difference in LSB and MSB bits(I mean zeroth bit and 7th bit) on two platforms ,will redefining bit pattern will be a good solution to solve the problem.
##########
Also i have one general query regarding TCP and endian.Suppose I transmit from my application one long integer(say 8 bytes size) which is ruuning on littel endian and at other end on big endian system read it directly,will this cause some error.
################
Which is the ideal way or recommended way to transmit integers on network.Should we convert them to chars(one byte) and transmit 4 chars for one integer and regenerate integer using program at other end.
############
The last two question's answer will help me in my general understanding of transmission on TCP/IP and other lines.
###############
hope Perderabo you will be again kind enough to share some more knowlege of yours with all of us.
################
What difference it makes when u say TCP transmits octet and not byte?

I guess thats too many questions in one query.
sorry for that.

First, let's tackle TCP/IP and octets and integers. Not all computers have 8 bit bytes. Some weirdos have 6 bit bytes. I only one I know of is the CDC Cyber series, but I believe that there are more. Or were. Most everyone agrees on 8 bit bytes these days. But the designers of TCP/IP wanted it to work on all computers. They use the term octet as a politically correct term for an 8 bit quantity. TCP/IP does not deal with 8 octet integers. But it does have 2 octet integers which it called "short", and 4 octet integers which it calls "long". Both longs and shorts travel over the wire big-endian. But both are placed on the wire and extracted from it via macros. The macros are called ntohl, htonl, ntohs, and htons. (ntohl, for example, means network to host long). On hp-ux, just type "man htonl" to see a man page on them. These macros solve the problem of integers across a network. On a big-endian system like HP, they are just null macros and are removed by the pre-processor. On some other systems they expand into whatever is required.

It is obvious that this concept can be expanded in both directions. I would write new macros for 8 byte integers. And if I really needed it, I would write macros for bytes as well. It isn't hard to reverse the bits in a byte if that is truely required.

I have never worked with HDLC, and I just looked it up. I can't find proof of this, but I remain skeptical that it reverses its bytes. But if you are dumping data into the information frames with the bytes reversed, I guess that you are stuck with it.

Anyway, what is an application program doing using HDLC? HDLC is a level 2 protocol. That makes as much sense as an application program opening the ethernet device and writing its own ethernet frames. When you do stuff like that, it really starts to sound like you are talking to a directly attached device rather than using a network. Could you describe your network topology?