Priscilla, Comments inserted below. Since, I'm using stupid MS outlook, my comments are marked with a >> at the start of the paragraph only (which I had to manually type!).
WAYNE BAETY, MCSE, A1C, USAF Network Systems Trainer -----Original Message----- From: Priscilla Oppenheimer [mailto:[EMAIL PROTECTED]] Sent: Wednesday, November 14, 2001 3:54 AM To: [EMAIL PROTECTED] Subject: RE: 2 "Line Hit" Scenarios... [7:25928] You're obviously a trainer! I am too. >> Only because I'm nit-picky enough to actually go and read the protocol specs to prove a point, that programmers background in me. >> The CRC is handled by the NIC, not the OS of the computer. Just a nit-picky >> That is true in most cases, but Cisco (Kalplana) definitely throws us a curve ball with cut-through-forwarding / fragment-free switching. Actually, I was half thinking of how Cisco Routers & Switches handle Ethernet, unless of course you don't think of the ASICS as being part of the OS. And if I want to be completely nit-picky, the ASICS themselves are an OS of sorts, but only because Cisco's marketing engineers (if there is such a thing) wanted to call them 'programs' on a chip. >> thing. However, I also have to object to the main message that you are giving. I think it's very important to mention that it is rare that B1 (a bridge or router) would retransmit. >> With line hits the scenario completely changes. Unless of course the line hit gremlin has the audacity to make it-self look like the Ethernet jam signal. >> The "right answer" for certain tests is that the bridge or router won't >> NOTE TO AUDIENCE: She's completely right in this respect; most tests will want you to say that retransmissions are an Application/Session/Transport occurrence. In the case of TCP over LLC1, that would make it the responsibility of PCA to retransmit a TCP segment (with PCB humbly requesting a retransmission). The reason for this is because all the other lower layers LLC1, IP, Ethernet are unreliable. When a line hit occurs, Ethernet will drop the frame at the data-link layer because of a CRC Error. This will then timeout the TCP stream on the receiving end (it won't notice an error, just an absence of a segment) and prompt it to send an ACK message with the sequence number of the erred segment - 1. This will force the sender to retransmit the missing segment after a timeout expires or maximum threshold has been reached. >> retransmit, unless the question is asking about a particular protocol where this isn't the case. >> The questioned mentioned a serial interface, and even went so far as to indicate that that was the focus of the question. Serial is traditionally the WAN arena. (or even more traditionally the unreliable circuit communications arena) IF they only want to test on TCP, the test writers should stick to TCP scenarios. >> Retransmitting is usually the job of the end node. The end node doesn't hear its frame get acknowledged, so it retransmits. >> I agree with this. 99% of retransmissions that occur when dealing with modern networking LAN technology are implemented by the end nodes (That is of course you are not talking about bridging the token passing protocols with Ethernet). However, when dealing with reliable (unreliable?) communications occurring in the WAN environment, that figure could and usually is much lower. I also believe it's an overstatement to blindly say that an end node is responsible for all frame retransmissions, especially with line hits. In fact, oftentimes the end nodes aren't even aware that an error has occurred when a line hit happens; some provision is made at the lower layers to keep them from suffering the effects of retransmitting. ATM's resiliency with 1 bit error recovery comes to mind. X25 has node-to-node retransmissions built in to the protocol. Satellite communications have and always will have node-to-node retransmissions built in to any protocol that ever will be used with it, simply because propagation delay makes it very expensive not to. And yes, Cisco Routers are in (from outer) space. That's why I mentioned 'physical layer protocol', to get him studying and thinking along the lines of the complete picture as well as keeping my email as short as possible. >> Retransmitting is the job of TCP or an upper-layer application. In the case of SNA or NetBEUI running above LLC2, it is the job of LLC2 at the end station (unless you use Local ACK). At some layer, for most applications, there is reliability. You mention UDP and say that there are no retransmissions. UDP doesn't handle retransmissions, but with most protocols that run above UDP, the host retransmits if there is no reply to its message. >> None on Cisco Routers. I can remember a few games of old (and possibly new) that were written to ignore erroneous UDP packets, instead of request a retransmission. In a SNMPv1 environment I am almost 99.99% sure that no retransmissions occur, even in the face of erred messages. In RIPv1, one of the reasons for the dreaded routing loop is brought about by the fact that a host could for some reason or another somehow not receive a poison update, and start a count-to-infinity war. This means that RIP messages are inherently unreliable and not acknowledged. Erred RIP frames are just 'silently discarded'. The silence is because there is no provision in the protocol made for acknowledgement and recovery. >> Here's a list of cases where a router or bridge would retransmit instead of the end host. BISYNC - YES X.21 - YES SDLC - YES The various LAPs (LAPB, LAPF, etc.) - YES? Cisco HDLC - NO!! (remember it's Cisco's own variety of HDLC) PPP - NO Frame Rely - NO Ethernet - NO Token Ring - NO FDDI - NO LLC1 - NO LLC2 - NO, unless you're using DLSw or RSRB with Local ACK LLC3 - who cares? ;-), but I think the answer is YES Ethernet causes some confusion for people because a data-link-layer interface monitors for collisions while sending and retransmitting if one occurs. I don't think this fits into the same category as we're dealing with in this question, but the neophytes think it does, so it's worth addressing. I consider sending without collisions a basic part of Media Access Control, analogous to getting the token on Token Ring. It's not the same as monitoring for an ACK and retransmitting if you don't get one, which Ethernet NEVER does. >> I thought of this, but then I was thinking that maybe the line hit doesn't trip the collision detect circuitry, in most cases it won't. Maybe the gremlin feels conniving today and feels like letting Ethernet believe the transmission was a success. This is true more often than not. >> This is an extremely easy Networking 101 question where I come from. It's really frustrating that it results in so much confusion. >> Networking 101, but not Networking 303, which is why I included a paragraph of disclaimer somewhere in the middle. >> Priscilla >> I welcome discussions like this on this group. I wish there were more of them. Now the really good stuff comes out. Thanks, Priscilla. Wayne >> At 11:21 PM 11/12/01, Baety Wayne A1C 18 CS/SCBX wrote: >Line hits are caused by physical disturbances, electronic influences >on the transmission medium. The question draws attention to the serial >connection between B1 and B2, and a possible difference between Ethernet >connections. Ethernet makes no provision for physical layer protocol >retransmission in the face of erred communications. An explanation follows. > > > There are different physical layer protocol entities for Ethernet, >notably MLT-3 for fast Ethernet, Manchester for Ethernet, etc. These are >actual protocols for transferring bit streams over a common medium and only >serve to perform line encoding. When an error presents itself, most often >these errors register as invalid code signals to the distant end, which >somehow gets translated into a data signal, forcibly in the case of >Ethernet. After this process is complete the bit streams are compacted and >provisioned into 8-bit boundaries and are passed up to the data link layer. >At this point, the communication enters the prevue of a central processing >unit. The OS controlling the CPU would naturally do a CRC function on the >received data stream and extract the CRC that was computed by the sending >node, and do a comparison of the two. Actual implementations could vary. >This in essence is an overview of Ethernet Technology. The important thing >to remember is that there is not a protocol function that occurs at the >point the bit streams are sent from the hardware to the main CPU (channel >access functions are handled in hardware on a NIC). All communication is >accepted carte blanche, and naturally so. Ethernet is primarily a LAN >technology were error prone communications (caused by EMI or other naturally >occurring phenomenon) is tightly controlled and minimized. In serial >communication technology there is such a protocol function because there is >a higher chance of their being electromagnetic influences, propagation >delay, etc. > > In serial communications at the point that the bit streams are >decoded into logical binary words (that 8 bit provisioning scheme >aforementioned). There is a protocol function implemented to control the >actual reception and healthiness of the bit streams. HDLC is the default >protocol for Cisco Routers, but there are other notables. Such as Bi-Sync, >SDLC, LAPB, PPP, etc. In some of these protocols there are provisions for >the retransmission of frames when errors are detected, channel multiplexing, >stream windowing as well as frame sequencing and acknowledgements. > > Why this long answer? Remember the question draws attention to the >physical layer when 'line hits' are mentioned. Further clues were given >when the only difference mentioned was a change in physical composition. >It's up to you to decide if the test maker in this instance is testing to >see if you know all of this, judged by the overall difficulty of the exam. > > To answer your question if there is a line hit between B1 and B2, B1 >will always retransmit. In most serial encapsulations method, the frame >never clears the buffers on B1 until B2 acknowledges reception to B1. > > There was an effort to add this amount of reliability to Ethernet >and Token Ring environments, hence LLC which is a spin off of sorts of HDLC >for serial communications. With LLC there are actually 3 different modes of >communication. Type 1 is the normal mode that you would normally see in >modern networking environments (Type 2 is more usual for Token Ring). Type >2 is modeled after communication qualities that serial communications need >to overcome. Type 3 is not commonly used. To be succinct, it is like >slapping a serial protocol over Ethernet or Token Ring. When Ethernet is >behaving like a serial interface it will buffer, acknowledge and retransmit >erred frames just like a serial interface (In that case, each intermediate >device is responsible for retransmitting any frames with errors). Like >everything else in life, the true answer depends on what you are doing. > >To be safe, let me point out that all of this nonsense usually is spoken of >in the books as residing at the Data Link layer. I believe the test >question may also be trying to confuse you with what are the >responsibilities of the Transport layer (TCP to be exact). But what if you >aren't even using TCP, What if you are using UDP over IP over Ethernet? >There is clearly no retransmission effort going on here. All confusing >isn't it? Don't worry in time you'll sort it all out. > >Cheers and Good Luck, > >WAYNE BAETY, MCSE, A1C, USAF >Network Systems Trainer > > >-----Original Message----- >From: Todd Carswell [mailto:[EMAIL PROTECTED]] >Sent: Monday, November 12, 2001 11:09 PM >To: [EMAIL PROTECTED] >Subject: 2 "Line Hit" Scenarios... [7:25928] > >Here's the setup for my 2 questions... > >PCA-------B1-----------B2--------PCB > >Bridge 1 and Bridge 2 are running Transparent Bridging between them. > >Question 1: There's a SERIAL connection between B1 and B2. B1 and B2 are >configured for transparent bridging. If PCA sends a packet to PCB and the >frame is errored somehow, who takes care of the retransmission? I assume >it's PCA because it's a serial connection. Am I right? > >Question 2: There's an ETHERNET connection between B1 and B2. The bridges >are still using Transparent Bridging and PCA sends a packet to PCB. Again, >the frame has an error. Will B1 be the device to handle the retransmission? > >Thanks, guys! > >Todd ________________________ Priscilla Oppenheimer http://www.priscilla.com Message Posted at: http://www.groupstudy.com/form/read.php?f=7&i=26172&t=25928 -------------------------------------------------- FAQ, list archives, and subscription info: http://www.groupstudy.com/list/cisco.html Report misconduct and Nondisclosure violations to [EMAIL PROTECTED]