[arin-ppml] Routing Research Group is about to decide its scalable routing recommendation
rw at firstpr.com.au
Sat Dec 19 07:55:50 EST 2009
Short version: More on latency of 3G and GPRS wireless links.
Historical approaches to scalable routing and
addressing: 8+8, GSE and MHAP.
Before replying to your message, I want to mention that there's a
much more important factor in the latency of 3G wireless data than
the modulation technique - the time-slicing of the upstream and
Downstream, the device has to wait its turn for data to arrive for
it, rather than some other device. Upstream, it has to wait for the
right time to transmit - and there are various ways this can be
Here are some observations done quickly. Using a 3G HDSPA link and
pinging from WinXP I find a RTT of 114 to about 300ms to the first
router at the ISP. But the first packet, after a few seconds of
inactivity, might have a RTT of 800 to 1500ms. (The 3G network is
Optus and the router is at Internode in my city - Melbourne.)
When the link switches itself to WCDMA, for reasons unknown, the RTT
is typically 400 to 500ms with initial packet RTTs around 1000 to
Forcing the modem to GSM (GPRS data), I get RTT times 500 to 1300ms,
with no obvious extra time for the first packet.
I think for the 3G modes, the initial packet extra delay reflects not
so much the slots for upstream and downstream, but the modem going
from inactive to active, which takes a while to reserve some upstream
and downstream bandwidth. Without this, the modem would be chewing
spectrum and time all the time - which I suspect it may be doing in
You wrote, in part:
>> Which raises an interesting question, why hasn't SCTP taken off more?
> I have met in person with some of the guys heavily involved in SCTP ways
> back when; very interesting work. I think that in the end the reason it
> has not taken off more is basically the same reason none of the
> host-based multihoming solutions has taken off either: too complex.
> Anything that involves multiple addresses per host and heaven forbid
> even worse multiple interfaces per host is a nightmare to TE and
> troubleshoot. Imagine trying to troubleshoot a network issue with your
> typical tech support subcontracted overseas when the thing involves
> multiple interfaces with multiple addresses crossing multiple backbones.
> Good luck; as of today there are no tools for this and no money to build
> Sadly, nothing matches the raw simplicity of this unique PI prefix that,
> unfortunately, makes the DFZ big.
This is with host-based multihoming - core-edge elimination schemes.
A core-edge separation scheme does the work in ITRs and ETRs, so
there are no questions about host software. This is more complex
than at present - but extra complexity seems unavoidable.
>> Robin Whittle wrote:
>> I am sure we can achieve scalable routing with a
>> core-edge separation scheme. C is the current
>> situation and the idea is that we won't need to
>> be so fussy once there is a scalable form of PI space.
> I will point out that this approach has been tried many times
> before: 8+8, GSE, MHAP, etc.
I don't know much about these, but 8+8 and GSE are mentioned in:
I don't know anything about your MHAP - but its ca. 2003 home is:
I can't find mention of MHAP in my archive of RRG messages. There
are 42 mentions of "8+8" and 160 mentions of GSE. Ran Atkinson wrote
a little timeline of GSE:
In June 2008, Tony Li (RRG co-chair) wrote to me:
>> Please, please, please go read GSE. You may not like it, you may
>> not agree with it, but until you grok it, you haven't seen a big
>> chunk of the solution space.
I had a quick look at it, but it is for IPv6 only so I didn't study
it in depth.
More information about the ARIN-PPML