A test for
mobile broadband speed
Feb. 12, 2014
Test Setting (1/2)
• Installed an original measurement application on 11 iPhones
and measured QoS on a 10 minute interval for 2 hours.
• Date and time
Jan. 24, 2014 16:00‐18:00 (Japan Time)
Jan. 31, 2014 16:00‐18:00 (Japan Time)
A classroom at the 5th floor of the building of
the Faculty of Economics at Kyushu University.
Each iPhone cut Wi‐Fi connection and used only
The measurement server was located at Tokyo.
Test Setting (2/2)
• Our application conducted QoS measurements every 10
minutes for 2 hours.
• 12 measurement windows for each handset.
• In each measurement window
• RTT was measured 150 times at the interval of 0.2 seconds.
• Download/Upload speed was calculated based on throughputs for
• Upload speed was measured once.
• Download speeds were measured in three different settings:
one on a single TCP stream, the others are on 3 and 5 TCP
• The number of hops was counted 3 times using “traceroute.”
• Softbank’s LTE network offered faster but more volatile speed than
• Since the variations in the measurement outputs were significant, a
small number of speed measurements could not produce a
• Throughputs changed quite dramatically in a short amount of time.
• Considering the size of standard deviation, the difference that comes
from the number of TCP streams in the download speed
measurement was not statistically significant, and therefore can be
ignored in a practical sense.