“sent_buf_size”,
and
“recv_buf_size”
. The buffer size is
chosen among five values: 1024, 2048, 4096 and 16.384 byte.
The ICMP packet header at the MAC layer is 64B and we add
4032B payload more, thus, the ICMP packet is 4kB in total.
The RTT results are shown in Fig.5.
Fig. 5.
RTT MEAN Latency and Standard Deviation on USRP E100
platforms
It is important to note that in Tx side, USRP FIFOs/SRAM
performs all of the buffering. Consequently, we cannot really
buffer anything in the embedded computer. The Linux kernel
in the embedded computer just waits for the notification that
space is available in the USRP before committing more
packets. In Rx side, GPMC driver performs all the buffering.
Therefore the USRP sample rate does not play a role in case of
buffering scheme. However, it contributes much to the latency
of generating samples.
The effect of UHD buffer sizes on GNU Radio/USRP1
platform is mentioned in [18, 19]. By our experiment on USRP
E100, Tx UHD Buffer size
“sent_buf_size”
has no role in the
Tx latency. It is because when samples are collected enough to
generate a packet, it is sent right away without waiting for Tx
UHD buffer to get full. In the other hand, Rx buffer size
“
recv_buf_size”
plays a key role to the latency since samples
received from GPMC bus are moved to this buffer. When the
Rx UHD buffer is fulfilled, samples are transmitted to the GNU
Radio. It is the same as our analytical latency in the section III.
In the case of timestamp approach, we create two programs,
one is very simple Tx program called
hw_latency_test
with no
GNU Radio component, generating sample from UHD driver.
Its purpose is to reduce the delay introduced in GNU Radio
software. Another is a GNU Radio program named
full_latency_test
which leverages some SDR radio components
such as QAM modulation, filters, and decoders. Because of
three places that we put the timestamp, the results indicate
three components
∆
software,
∆
UHD +
∆
UsrpB and
∆
GPMC +
∆
hardware (Fig.6).
Fig. 6.
MEAN Tx latency results of timestamp approaches
We transmit 512 samples (4kB) and conduct 1000 trials to
get the average latency and standard deviation for each part of
the latency. It shows that the experimental results of the
(
∆
UHD +
∆
UsrpB) and (
∆
GPMC +
∆
hardware) are just little
higher than the analytical latency in the section III (Fig.6).
In case of the approximation approach, we conducted 100
tests with different
rt
time values by calculating the number of
successful transmissions as well as the number of failures. In
order to compare with the analytical latency results and the two
other experimental methods, we also choose the same
parameters: 1024 samples (4kB) to transmit at 2Msps USRP
sample rate. The results are the total latency of the bus
transmission and USRP hardware:
∆
bus +
∆
hardware (Fig.7).
13
13
Fig. 7. The total bus and hardware latency results from experimental and
analytical approaches
The latency in the timestamp method (1.480ms) is smaller
than the approximation method (1.495). The latency from
analytical approach is the smallest (1.417). It is reasonable
because when analyzing the latency, we can only estimate the
minimum
∆
GPMC. The experimental
∆
GPMC could be
different. The timestamp method could be increased due to the
timestamp procedure running along with the main program. In
the approximation approach, the approximated time
rt
is
always higher than the real latency. This is the understandable
reason for the different results between two experimental
approaches.
For the relations between RTT and the latency from other
two approaches, note that when the UHD buffer size is equal to
GPMC packet size (4kB for example), we can assume that the
Tx latency = Rx latency. In this case, whenever an ICMP
packet is received, the Rx UHD buffer gets full and
immediately transfers the data to GNU Radio. Therefore, RTT
= 2*(Tx latency + Rx latency). In other words, Tx latency = Rx
latency = RTT/4 = 4.75ms.
Besides ∆bus
(0.667ms) and
∆hardware
(0.25 + 0.25 = 0.5ms)
, ∆software in this case
contains the time for reading and conducting TUN/TAP file
descriptor, reading and processing 4kB ICMP packets into
samples, user-space and kernel-space interaction. Assumed that
GNU Radio processes samples at the rate 2Msps, thus, it takes
at least 0.5ms to generate a 4kB ICMP packet. Thus, the delay
due to the TUN/TAP component and
ping
utility is about
3.083ms. The results basically show that the time for
processing at the host computer dominates the whole latency of
the GNU Radio/USRP E100 SDR platform.
A
CKNOWLEDGMENT
This research was supported by World Class University
program funded by the Ministry of Education, Science and
Technology through the National Research Foundation of
Korea (R31-10100).
R
EFERENCES
[1] Ettus Research LLC.,
“Universal Software Radio Peripherals”,
http://www.ettus.com.
[2] GNU Radio.,
http://gnuradio.org/redmine/projects/gnuradio/wiki.
[3]
Wikipedia., ”Software Defined Radio”,
http://en.wikipedia.org/wiki/Software-defined_radio.
[4] M. Cummings and S. Haruyama.,
“FPGA in the software radio”,
IEEE Communications Magazine, 1999.
[5]
“WARP: Wireless open Access Research Platform”,
http://warp.rice.edu/trac.
[6] ANSI/IEEE std 802.11 part 11.
, “Wireless LAN medium access
control (MAC) and Physical Layer (PHY) Specification”, IEEE
press, 1999.
[7]
K. Tan, J.Fang, J.Fang, et al., “SORA: High Performance
Software Radio Using General Purpose Multi-core Proce
ssors”,
6th USENIX Symposium on Networked Systems Design and
Implementation 2009, USENIX, 2009.
[8]
T. Schmid. O. Sekkat, and M. Srivastava., “An experimental
study of network performance impact of increased latency in
SDRs”, WiNTECH’07, 2007.
[9] S. Valentin, H.
von Malm, and H. Karl., “Evaluating the gnu
software radio platform for wireless testbeds”, Technical Report
TR-RT-06-273, 2006.
[10] G. Nychis, T. Hottelier, Z. Yang, S. Seshan and P. Steenkiste.,
“Enabling Mac protocol implementations on Software
-Defined
Radi
os”, Proceedings of the 6
th USENIX Symposium on
Networked Systems Design and Implementation, USENIX,
2009.
[11]
Y. Lin, Hyunseok Lee, Yoav Harel, Mark Woh, et al., “SODA:
A low-
power architecture for software radio”. Proceedings of
the 33rd International Symposium on Computer Architecture,
ISCA, vol.1, pp. 89-100, 2006.
[12]
A. Jow, C. Schurgers, and D. Palmer., “CalRadio: A Portable,
Flexible 802.11 Wireless Research Platform”, 1st International
Workshop System Evaluation for Mobile Platforms, San Juan,
Puerto Rico, 2007.
[13]
Wikipedia., “Universal Software Radio Peripheral”,
http://en.wikipedia.org/wiki/Universal_Software_Radio_Periphe
ral.
[14]
L. Choong., “Multi
-channel IEEE 802.15.4 packet capture using
software defined radio”, UCLA Networked and Embedded
Sensing Lab, 2009.
[15]
Ettus Company LLC., “Daughterboard products“,
https://www.ettus.com/product/category/Daughterboards.
[16]
Universal Hardware Driver., “wiki”,
http://ettus-
apps.sourcerepo.com/redmine/ettus/projects/uhd/wiki.
[17] TUN/TAP Interface Tutorial.,
http://backreference.org/2010/03/26/tuntap-interface-tutorial/.
[18]
J.C. O’Sullivan, P. Di Francesco, U.K. Anyanwu, L. A. DaSilva,
and A. B. MacKenzie., “Multi
-hop MAC Implementations for
Affordable SDR Hardware”,
IEEE Symposia on New Frontiers
on Dynamic Spectrum Access Networks (DySPAN). poster
paper, pp. 632-636, 2011.
[19] A. Puschmann
,
M. Kalil and A. Mitschele-Thiel.,
“Implementation and Evaluation of a Practical SDR Testbed”
,
4th International Conference on Cognitive Radio and Advanced
Spectrum Management (CogART), 2011.
[20] P.D. Sutton, J. Lotze, H. Lahlou, S.A. Fahmy, K.E. Nolan, B.
Ozgul, T.W Rondeau, J. Noguera, L.E Doyle.,
“
Iris: An
Architecture for Cognitive Radio Networking Testbeds
”,
IEEE
Communications Magazine, vol. 48, 2010.
14
14
Do'stlaringiz bilan baham: |