The IP over ATM Mailing List Archive by date

[onvural@vnet.ibm.com]

*** Resending note of 03/29/96 09:34                                           
 - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -                 
 From:                                                                         
 Subject:                                                                      
Subject: Re: EE Times on IP over ATM                                           
Eric,                                                                          
When the queue occupancy reaches a threshold                                   
you start monitoring the (AAL 5) cell with SDU-type=1. After that cell,        
start dropping all cells which all have SDU-type=0 until you                   
receive the next SDU-type=1 cell. No assembly is required. With PPD,           
you don't wait to recieve the SDU-type=1 cell to start dropping,               
and drop all SDU-type=0 cells until type=1 cell is received. Repeat            
as necessary. Accordingly, no assembly is required as the SDU type             
is in the cell header.                                                         
Raif,                                                                          
                                                                               
>       How does EPD/PPD differ from what's being discussed here?  Both        
>(or either) require examination of the entire packet (thus assembly) -        
>at least in passing - by the network.  How else would the network know        
>to discard cells associated with a packet to be discarded (and how can        
>it otherwise know what cells to discard and when to stop discarding)?         
                                                                               
--                                                                             
>Eric Gray                                                                     
                                                                               
onvural@vnet.ibm.com wrote:                                                    
>                                                                              
> *** Resending note of 03/28/96 15:38                                         
>  - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -               
>  From:                                                                       
>  Subject:                                                                    
> Subject: Re: EE Times on IP over ATM                                         
> All these issues have been investigated in detail and the literature         
> is full of various conclusions under different assumptions. One              
> result of interest to the enclosed arguments is that upto 60-70%             
> link utilization, reassemble process introduces additional work              
> with NO recognizable benefit. Beyond 70%, checking corrupted SDUs            
> is beneficial. These conclusions did not consider early packet               
> discard/partial packet discard. I would think                                
> reassemble process inside the network may not buy much if EPD/               
> PPD is implemented. Nevertheless, I agree this is an implementation          
> issue. Raif                                                                  
> xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx          
> On Thu, 28 Mar 96 14:15:00 EST Hiroshi Esaki Wrote:                          
>                                                                              
> > I guess it depends on the implementation.                                  
> >  Whether really re-assemble AAL5 SDU (i.e., CRC-32 check) and assemble     
> >  again in MCS or not (just scheduling the cell departure) seems to be      
> >  implementation issue for each MSC vendor.                                 
>                                                                              
> Quite true.All you really need to do is to collect all the cells together    
> and then send them all out after all cells composing an SDU have been        
> collected.  Of course, re-assembling the packet does have the advantage      
> of being able to detect SDU corruption so as not to relay a bunch of         
> cells for a corrupt packets.  On the other hand just collecting cells and    
> being aware of SDU boundaries is probably more efficient in terms of both    
> memory usage and throughput.  But as you point out, this is entirely         
> an implementation decision.                                                  
>                                                                              
>  --- pete                                                                    
>