IP characteristics for 3G and WiFi links

Owen DeLong owen at delong.com
Thu Aug 26 20:25:50 UTC 2010

I think you have your latency numbers backwards. In my experience, HSDPA
has higher RTT than WiFi. Why would you limit your Wifi to A/G? If you're using
5Ghz (A), much better to go to N than A. N has slight advantages over G in the
2.4Ghz realm.

Your stated packet loss rates are obscenely high and loss at those rates would
severely degrade user experience. Loss over 1% is enough to cause significant
slow-down in TCP short-lived or interactive flows and 2% is more than enough
to effect even longer continuous data flows.

I have not done comparisons of average loss rates between WiFi and 3G
services where the loss rates were so high because either one would be
basically unusable until the packet loss was reduced.


On Aug 26, 2010, at 9:32 AM, Richard Barnes wrote:

> On Thu, Aug 26, 2010 at 6:26 AM, Daniel Migault <mglt.biz at gmail.com> wrote:
>> Hi,
>> We are testing protocols on our lab platform and we would like to simulate
>> communication 2 types of communication :
>>   - From terminals to service platform using a 3G (HSPA / HSPA+) Access
>> connection
>>   - From terminal to service platform using a WiFi Access connection
>> We are using dummyNet to simulate the links so we are interested  in IP
>> characteristics layers for Packet loss Rate, bandwidth and latency. Values
>> depends on multiple factors, but we would like to know what mean values are
>> considered when services are deployed.
>> Currently we are considering the following values. Packet Lost Rate for L2
>> seems 7% for Wifi and 5% for 3G. We are wondering how L3 is affected?
>> Parameter           | Wifi (802.11a/g) | 3G (HSDPA)
>> Latency                100 ms               60 ms
>> Bandwidth             5 Mbps               3 Mbps
>> Packet Lost Rate   XXX                   XXX
>> Any comment links will be appreciated.
>> Regards,
>> Daniel
>> --
>> Daniel Migault
>> Orange Labs / Security Lab
>> +33 (0) 1 45 29 60 52
>> +33 (0) 6 70 72 69 58

More information about the NANOG mailing list