[c-nsp] Is this throughput on ATM DS3 in 3640 normal?

[Mark Rogaski]

An entity claiming to be Peter Olsson (pol at leissner.se) wrote:
: 
: They get about 40 Mbps when they only push traffic in one direction.
: When they try duplex traffic they can't get higher than about 34 Mbps
: in one direction and about 27 Mbps in the other.
: Are these numbers normal?

What CoS (eg. VBR-nrt, UBR, etc) is provisioned, and what are the
parameters of the service?  What other VC's or VP's are sharing the ports
at either end?


An entity claiming to be Peter Olsson (pol at leissner.se) wrote:
: On Tue, 30 Aug 2005 12:18 -0600, james edwards wrote:
: 
: > ATM cell tax is about 20 % so 34 Mbs about right.
: 
: Should they then have 34 in each direction or is it ok to
: have 34 in one and 27 in the other?

The actual amount lost to overhead is dependent on the frequency
distribution of packet sizes across the link.  If you have the frequency
distribution from Netflow (or some such tool), you can calculate the wasted
bandwidth as follows.

SUM_{packets of size x} freq_x * { 1 - [ x / ( ceil( x / 48 ) * 53 ) ] }

If you have A) a site sending 64 byte packets 46% of the time and 286
byte packets 54% of the time and B) a site sending 64 byte packets
34% of the time and 232 byte packets 66% of the time, the first site will
be losing 24% to the cell tax and the second site will be losing only 14%.

And you need to use the frequency distribution, not just average packet
size.  The example above would show 13% lossage at both sites if average
packet size was used.


: 
: > Which framing are you
: > using ? C bit has quite a bit of overhead (to make DS2's) that you cannot
: > use on an ATM DS3 so you can squeeze a bit more out of an ATM DS3 with
: > Direct Mapping/ADM framing.
: 
: Current framing is DS3 C-bit direct mapping.
: Should I change it, and to what?
: (config-if)#atm framing ?
:    cbitadm   DS3 Framing C-bit ADM
:    cbitplcp  DS3 Framing C-bit PLCP
:    m23adm    DS3 Framing M23 ADM
:    m23plcp   DS3 Framing M23 PLCP
: 
: DS3 Scrambling is off if that is of relevance.


ADM (aka direct mapping) gives you 140268 cps versus PLCP's 96000 cps.  You
don't need to use scrambling unless you are doing PLCP to prevent loss of
framing due to framing-like patterns in the cell payload.  Direct mapping
derives framing by looking for cell headers with valid HEC fields, there's
a much lower risk of losing framing due to user data.

: 
: Another question: Should one of the routers have clock source internal
: on the ATM interface?
: 

Like James said, it all depends on whether the provider is providing timing
on the circuit (which they probably are).  But if you are not getting any
input errors, your timing configuration is probably correct.

Mark

-- 
[]                        |
[] Mark Rogaski           |       The very opposite of spinach is armor.
[] wendigo at pobox.com      |       -- Salvador Dali
[] mrogaski at cpan.org      |
[]                        |
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 189 bytes
Desc: Digital signature
Url : https://puck.nether.net/pipermail/cisco-nsp/attachments/20050830/d2c543fe/attachment-0001.bin