Project

General

Profile

Accelerate3g5 -- theresa » History » Version 8

theresa, 06/13/2017 03:11 PM
First performance measurement results

1 1 theresa
h1. Accelerate3g5 -- theresa
2
3
h2. Summary
4
5
We will integrate Osmocom 3G into our research testbed as one of multiple access networks that an end device can choose between as a data carrier.
6
The goal is to gather information on current network performance, e.g. cell load or congestion within the core network.
7
This information should be made available through a suitable API, e.g. for monitoring or to help clients choose between available access networks (Details TBD).
8
9
h3. Participants
10
11
* Theresa Enghardt (email: theresa@inet.tu-berlin.de, jabber: tenghardt@inet.tu-berlin.de)
12
13
h2. Details
14
15
Part of ongoing research at "INET":http://www.inet.tu-berlin.de/menue/internet_network_architectures/ (TU Berlin) on choosing between multiple access networks, related to the Socket Intents prototype (see "paper":http://conferences.sigcomm.org/co-next/2013/program/p295.pdf and "research prototype code":https://github.com/fg-inet/socket-intents).
16
17
The main research question is how to choose between different available access networks, such as 802.11 and 3G/4G, to optimize performance for the client. As opposed to current carrier's "offloading" approaches, we propose a solution controlled by the end device. Our prototype gathers statistics on the performance of the available access networks and aims to make better decisions on when to use which network. 
18
19
For this, it would be useful to get information about the current load of the network, e.g. of the cell or of the congestion within the core network. However, in proprietary cellular networks, this kind of information is usually not available to user devices. 
20
21
We want to investigate how to make it available within the network (e.g. for monitoring) and possibly to the user device.
22
More details TBD.
23
24
Also, we will carry out performance tests of the 3G data service under different load conditions.
25
26
h2. Journal
27
28
_2017-03-07_
29
Start. Set up wiki page.
30
31 2 theresa
_2017-03-16_
32
Inquiring about frequency regulations and how to obtain an experimental license.
33
34 3 theresa
_2017-03-28_
35
Request for Experimental radio license has been sent off.
36
37 4 theresa
_2017-04-24_
38
Request for Experimental radio license has been granted.
39
40 6 theresa
_2017-05-23_
41 4 theresa
First attempt to get the core network running with the example config with adjusted IP addresses. 
42
All components start up properly except the HNBGW which fails with "unable to bind socket: (null):0: Protocol not supported". Apparently it fails to set a local address/port before trying to connect IuPS/IuCS.
43 1 theresa
Will investigate further.
44 6 theresa
45
_2017-05-24_
46
Got the core network running. Apparently SCTP was not supported on my machine, now it is.
47
Configured the nano3G, but its NTP isn't working yet (85.10.199.217 does not answer). Looking for how to configure a different NTP server.
48 4 theresa
49 7 theresa
_2017-06-02_
50
The NTP server eventually worked without being reconfigured, so the HNBGW connection could be established. Connected a smartphone to the femtocell for the first time. Data service works.
51
52 1 theresa
h2. Conclusions
53
54 5 theresa
* In March 2017 in Germany it was possible to get a license for "Experimental Radio":https://www.bundesnetzagentur.de/EN/Areas/Telecommunications/Companies/FrequencyManagement/FrequencyAssignment/ExperimentalRadio/experimentalradio_node.html for the femtocell on UMTS band 5 (824-829 MHz and 869-874 MHz). 
55
** Granted for 24.04.2017 until 31.03.2018
56 8 theresa
57
h3. Performance measurements
58
59
*Setup*: Smartphone (LG K10 Android 5.1.1 Lollipop) -- Femtocell -- Core network VM -- Web server VM
60
*Workload*: Static web page, 32 objects of 200 KB each, downloaded via Chrome mobile browser
61
* about 30 concurrent TCP connections, each with an average throughput of 82 - 186 kBit/s
62
* total throughput varies between ~200 kBit/s and ~1.5 Mbit/s, moving average (8 seconds): ~1 Mbit/s
63
* RTT averages are between 90 and 500 ms per TCP connection (according to tcptrace), minimum values around 70 ms, maximum values around 800 ms
64
* see also graph "throughput-without-gtp.png"
Add picture from clipboard (Maximum size: 48.8 MB)