[Dataset] cnu/cdma (v. 2007-09-26)

top

the initial version

@MISC{cnu-cdma-2007-09-26,
  author = {Youngseok Lee},
  title = {{CRAWDAD} data set cnu/cdma (v. 2007-09-26)}, 
  howpublished = {Downloaded from http://crawdad.cs.dartmouth.edu/cnu/cdma},
  month = sep,  
  year = 2007
}
					

We collected tcpdump data from a CDMA 1x EV-DO network in South Korea that provides 
high-speed "always on" Internet connectivity in a wide-area mobile environment.

Since cellular networks are equipped with wide coverage, mobility, and 
high data rates, the number of subscribers to the data service of 
cellular networks grows rapidly. The number of CDMA2000 1x EV-DO subscribers 
in South Korea was 7.7 million at the end of June 2004 according to 
the Ministry of Information and Communication in Korea (http://www.mic.go.kr).
Mobile Internet users with CDMA 1x EV-DO terminals use Internet applications 
such as web browsing, multimedia streaming, or email. Therefore, the efficient 
transport protocol is important to achieve the maximum throughput over 
the error-prone wireless link with uctuating bandwidth, large delay, and jitter.

in order to investigates the long-lived TCP bulk throughput over the CDMA 
1x EV-DO service, we collected TCP traffic data from a CDMA 1x EV-DO network 
in South Korea that provides high-speed "always on" Internet connectivity 
in a wide-area mobile environment 

This data set allows for measurement-based analysis for TCP performance 
that could be used for the correct model of the 3G wireless link characteristic 
and for the real-world simulation of TCP behavior over the 3G wireless network.

Code Division Multiplexing Access (CDMA) 1x EVolution-Data Only (EV-DO), 
which was finalized by the 3G Partnership Project 2 (3GPP2) and was published 
by the Telecommunication Industry Association (TIA) as Interim Standard (IS)-856, 
provides high-speed data service (2.4Mbps/153Kbps for downlink/uplink) 
with wide coverage and mobility. The CDMA 1x EV-DO service with high-speed 
"always on" connectivity in a wide-area mobile environment is being deployed 
throughout the world. In South Korea, since 2002, two carriers (KTF and SK Telecom) 
have deployed CDMA2000-based 1x EV-DO services which enable Video on Demand (VOD) 
and Multimedia Message Service (MMS).

The CDMA 1x EV-DO system addes a high-speed data solution for existing IS-95 
or CDMA 1x service providers while they maintain the compatibility with the 
frequency and RF modules. While the specification of 1x EV-DO is based 
on the High Data Rate (HDR) proposal from Qualcomm, Inc., it includes new features 
such as Incremental Redundancy (IR) Hybrid ARQ for improving system performance 
against the fast fading condition.

The 1x EV-DO system provides large service coverage as well as handover with
high bandwidth. In addition, it employs a time-shared downlink which serves
only one user at any instant in time-multiplexed manner. Therefore, the mobile
terminal calculates its Signal-to-Interference-plus-Noise Ratio (SINR) and determines
the highest data rates among available 11 data rates with the calculated
SINR at every slot.  Based on the periodic report of the data rates 
every 1.67ms (1 slot duration), the base station schedules slot allocation. 

The characteristics and the performance of cellular links have been widely 
studied in 2/3G networks (i.e., General Packet Radio Service (GPRS), IS-95A,
IS-95B, CDMA2000, and CDMA 1x EV-DO). It is well known that Forward Error
Correction (FEC) and link-layer retransmissions have been implemented to
defeat the challenging radio propagation environments. When a packet is transmitted,
the channel has to be allocated to each packet, which also causes the
variable delay. In CDMA 1x EV-DO networks, the peak rate of downlink is 
specified as 2.4Mbps and that of uplink as 153Kbps per user, and the maximum number
of data users is 16. Recently, 1x EVolution Data and Video (EV-DV), which
is being standardized, is expected to provide 3Mbps/1.5Mbps downlink/uplink
data rates by integrating voice and data channels.

A laptop was connected to the network of a CDMA 1x EV-DO carrier through 
the point-to-point protocol (PPP). TCP performance was tested with the Iperf 
tool which generates long-lived bulk traffic between the laptop and 
the Linux/FreeBSD machine. 

The packet traces were collected both at the server and at the client 
with tcpdump or windump.

[Traceset] cnu/cdma/tcpdump (v. 2007-09-26)

top

the initial version.

@MISC{cnu-cdma-tcpdump-2007-09-26,
  author = {Youngseok Lee},
  title = {{CRAWDAD} trace set cnu/cdma/tcpdump (v. 2007-09-26)}, 
  howpublished = {Downloaded from http://crawdad.cs.dartmouth.edu/cnu/cdma/tcpdump},
  month = sep,  
  year = 2007
}
					

We collected tcpdump data from a CDMA 1x EV-DO network in South Korea that provides 
high-speed "always on" Internet connectivity in a wide-area mobile environment.

We used a laptop with a CDMA 1x EV-DO USB modem made by CMOTECH 
and a Linux or FreeBSD machine for our experiments. The operating 
systems of the laptop for the experiment is Microsoft Windows XP 
home edition. The operating systems of Unix machines for the experiments 
are Linux kernel 2.4 and FreeBSD 4.9 release.

The laptop was connected to the network of a CDMA 1x EV-DO carrier through 
the point-to-point protocol (PPP). TCP performance has been tested with 
the Iperf tool which generates long-lived bulk traffic between the laptop 
and a Linux/FreeBSD machine. The Linux/FreeBSD machine was located 
at the research networks called "KOREN" which is connected to the network of 
a CDMA 1x EV-DO carrier via high-speed Korea Internet Exchange Points (KIX, KT-IX). 

Although the location is not the perfect place to monitor the CDMA 1x EV-DO 
link performance, it is assumed that the average TCP throughput of the CDMA 1x 
EV-DO subscribers could be approximately found from many runs of experiments. 

The Maximum Segment Size (MSS) was set to 1460 or 1448 bytes, SACK was 
enabled, and the duration of the test was 150 - 300 seconds. It is 
believed that the test duration is long enough to observe the TCP steady 
state performance, since the outstanding window has become stable within 
20 seconds in our measurements.

The packet traces have been collected both at the server and at the client with 
tcpdump or windump.

[Trace] cnu/cdma/tcpdump/1x_EV-DO_downlink (v. 2007-09-26)

top

the initial version

@MISC{cnu-cdma-tcpdump-1x_EV-DO_downlink-2007-09-26,
  author = {Youngseok Lee},
  title = {{CRAWDAD} trace cnu/cdma/tcpdump/1x_EV-DO_downlink (v. 2007-09-26)}, 
  howpublished = {Downloaded from http://crawdad.cs.dartmouth.edu/cnu/cdma/tcpdump/1x_EV-DO_downlink},
  month = sep,  
  year = 2007
}
					

Tcpdump trace colleced from CDMA 1x EV-DO Downlink.

This trace was collected from the downlink of a CDMA 1x EV-DO network.

A laptop was connected to the network of a CDMA 1x EV-DO carrier through 
the point-to-point protocol (PPP). TCP performance has been tested with 
the Iperf tool which generates long-lived bulk traffic between the laptop 
and a Linux/FreeBSD machine. The Linux/FreeBSD machine was located 
at the research networks called "KOREN" which is connected to the network of 
a CDMA 1x EV-DO carrier via high-speed Korea Internet Exchange Points (KIX, KT-IX).
By downlink traffic, we mean the traffic from the Linux/FreeBSD machine to 
the laptop.

Traces 1-6 and 8-9 below are sender-receiver pairs. For trace 7 below, 
the receiver trace was not captured because that the throughput and loss rate 
could be examined with the sender trace. 

At the TCP sender and receiver, we coud enable or disable the TCP timestamp option. 
Traces tagged with "TCP timestamp" below are captured after the TCP timestamp option 
enabled at both the sender and the receiver.

1. CDMA 1x EV-DO Downlink, 20050817 tcpdump data at the iperf sender (34 connections, 5-minute duration iperf, TCP timestamp)
2. CDMA 1x EV-DO Downlink, 20050817 tcpdump data at the iperf receiver (34 connections, 5-minute duration iperf, TCP timestamp)
3. CDMA 1x EV-DO Downlink, 20050817-2 tcpdump data at the iperf sender (16 connections, 5-minute duration iperf, TCP timestamp)
4. CDMA 1x EV-DO Downlink, 20050817-2 tcpdump data at the iperf receiver(16 connections, 5-minute duration iperf, TCP timestamp)
5. CDMA 1x EV-DO Downlink, 20050820 tcpdump data at the iperf sender (38 connections, 5-minute duration iperf, TCP timestamp)
6. CDMA 1x EV-DO Downlink, 20050820 tcpdump data at the iperf receiver(38 connections, 5-minute duration iperf, TCP timestamp)
7. CDMA 1x EV-DO Downlink, 20050828 tcpdump data at the iperf sender (30 connections, 5-minute duration iperf, TCP timestamp)
8. CDMA 1x EV-DO Downlink, 20050907 tcpdump data at the iperf sender (116 connections, 2.5-minute duration iperf, TCP timestamp)
9. CDMA 1x EV-DO Downlink, 20050907 tcpdump data at the iperf receiver (116 connections, 2.5-minute duration iperf, TCP timestamp)

tcpdump format

[Trace] cnu/cdma/tcpdump/1x_EV-DO_uplink (v. 2007-09-26)

top

the initial version

@MISC{cnu-cdma-tcpdump-1x_EV-DO_uplink-2007-09-26,
  author = {Youngseok Lee},
  title = {{CRAWDAD} trace cnu/cdma/tcpdump/1x_EV-DO_uplink (v. 2007-09-26)}, 
  howpublished = {Downloaded from http://crawdad.cs.dartmouth.edu/cnu/cdma/tcpdump/1x_EV-DO_uplink},
  month = sep,  
  year = 2007
}
					

Tcpdump trace colleced from CDMA 1x EV-DO Uplink.

This trace was collected from the uplink of a CDMA 1x EV-DO network.

A laptop was connected to the network of a CDMA 1x EV-DO carrier through 
the point-to-point protocol (PPP). TCP performance has been tested with 
the Iperf tool which generates long-lived bulk traffic between the laptop 
and a Linux/FreeBSD machine. The Linux/FreeBSD machine was located 
at the research networks called "KOREN" which is connected to the network of 
a CDMA 1x EV-DO carrier via high-speed Korea Internet Exchange Points (KIX, KT-IX).
By uplink traffic, we mean the traffic from the laptop to the Linux/FreeBSD machine.

Traces 1-2  below are sender-receiver pairs. For traces 3-4 below, the receiver trace was 
not captured because that the throughput and loss rate could be examined 
with the sender trace. 

At the TCP sender and receiver, we coud enable or disable the TCP timestamp option. 
Traces tagged with "TCP timestamp" below are captured after the TCP timestamp option 
enabled at both the sender and the receiver.

   1. CDMA 1x EV-DO Uplink, 20050821 tcpdump data at the iperf sender (35 connections, 5-minute duration iperf)
   2. CDMA 1x EV-DO Uplink, 20050821 tcpdump data at the iperf receiver (35 connections, 5-minute duration iperf)
   3. CDMA 1x EV-DO Uplink, 20050906 tcpdump data at the iperf sender (146connections, 2.5-minute duration iperf, TCP timestamp)
   4. CDMA 1x EV-DO Uplink, 20050907 tcpdump data at the iperf sender (110connections, 2.5-minute duration iperf, TCP timestamp)

tcpdump format

[Trace] cnu/cdma/tcpdump/1x_downlink (v. 2007-09-26)

top

the initial version

@MISC{cnu-cdma-tcpdump-1x_downlink-2007-09-26,
  author = {Youngseok Lee},
  title = {{CRAWDAD} trace cnu/cdma/tcpdump/1x_downlink (v. 2007-09-26)}, 
  howpublished = {Downloaded from http://crawdad.cs.dartmouth.edu/cnu/cdma/tcpdump/1x_downlink},
  month = sep,  
  year = 2007
}
					

Tcpdump trace colleced from CDMA 1x Downlink.

This trace was collected from the downlink (a Linux/FreeBSD machine to a mobile labtop) 
of a CDMA 1x network.

For traces 1-2 below, the receiver trace was not captured because that the throughput and 
loss rate could be examined with the sender trace. 

   1. CDMA 1x Downlink, 20050821 tcpdump data at the iperf sender (11 connections, 5-minute duration iperf)
   2. CDMA 1x Downlink, 20050830 tcpdump data at the iperf sender (38 connections, 5-minute duration iperf)

tcpdump format

[Trace] cnu/cdma/tcpdump/1x_EV-DO_downlink_ktx (v. 2007-09-26)

top

the initial version

@MISC{cnu-cdma-tcpdump-1x_EV-DO_downlink_ktx-2007-09-26,
  author = {Youngseok Lee},
  title = {{CRAWDAD} trace cnu/cdma/tcpdump/1x_EV-DO_downlink_ktx (v. 2007-09-26)}, 
  howpublished = {Downloaded from http://crawdad.cs.dartmouth.edu/cnu/cdma/tcpdump/1x_EV-DO_downlink_ktx},
  month = sep,  
  year = 2007
}
					

Tcpdump trace colleced from CDMA 1x EV-DO Downlink at a fast train, called KTX, moving from Seoul to Chonan (about 70Km) in Korea, in the speed of 295Km/hour.

This trace was collected from the downlink of a CDMA 1x EV-DO network 
when a mobile client (laptop) was in a fast train, called KTX.
During the measurement, the train was moving about 70 Km in the speed 
of 295 Km/hour from Seoul to Chonan in Korea. 

The laptop was connected to the network of a CDMA 1x EV-DO carrier through 
the point-to-point protocol (PPP). TCP performance has been tested with 
the Iperf tool which generates long-lived bulk traffic between the laptop 
and a Linux/FreeBSD machine. The Linux/FreeBSD machine was located 
at the research networks called "KOREN" which is connected to the network of 
a CDMA 1x EV-DO carrier via high-speed Korea Internet Exchange Points (KIX, KT-IX).
By downlink traffic, we mean the traffic from the laptop to the Linux/FreeBSD machine.

For trace 1 below, the receiver trace was not captured because that the throughput 
and loss rate could be examined with the sender trace.  3-minute trace was 
enough to capture the TCP steady-state behavior,  so the latter experiements 
was performed for 3 minutes to reduce the experiment time.

   1. CDMA 1x EV-DO Downlink at KTX express train from Seoul to Chonan, 20050830 tcpdump data at the iperf sender (2 connections, 16-minute duration iperf, 3-minute duration iperf)

tcpdump format

[Trace] cnu/cdma/tcpdump/1x_EV-DO_timestamp (v. 2007-09-26)

top

the initial version

@MISC{cnu-cdma-tcpdump-1x_EV-DO_timestamp-2007-09-26,
  author = {Youngseok Lee},
  title = {{CRAWDAD} trace cnu/cdma/tcpdump/1x_EV-DO_timestamp (v. 2007-09-26)}, 
  howpublished = {Downloaded from http://crawdad.cs.dartmouth.edu/cnu/cdma/tcpdump/1x_EV-DO_timestamp},
  month = sep,  
  year = 2007
}
					

Tcpdump trace colleced from CDMA 1x EV-DO Downlink with TCP timestamp option enabled.

At the TCP sender and receiver, we coud enable or disable the TCP timestamp option. 
To compare the typical packet loss patterns of TCP connections with/without 
the TCP timestamp option over CDMA 1x EV-DO downlink, we collected this trace
from the downlink of a CDMA 1x EV-DO networkwith TCP timestamp enabled.

A laptop was connected to the network of a CDMA 1x EV-DO carrier through 
the point-to-point protocol (PPP). TCP performance has been tested with 
the Iperf tool which generates long-lived bulk traffic between the laptop 
and a Linux/FreeBSD machine. The Linux/FreeBSD machine was located 
at the research networks called "KOREN" which is connected to the network of 
a CDMA 1x EV-DO carrier via high-speed Korea Internet Exchange Points (KIX, KT-IX).
By downlink traffic, we mean the traffic from the Linux/FreeBSD machine to 
the laptop.

For traces 1-4 below, the receiver trace was not captured because that 
the throughput and loss rate could be examined with the sender trace. 

   1. CDMA 1x EV-DO Downlink, 20060125 tcpdump data at the iperf sender (20 connections, 4.5 minute iperf, TCP timestamp)
   2. CDMA 1x EV-DO Downlink, 20060125 tcpdump data at the iperf sender (8 connections, 4.5 minute iperf, TCP timestamp)
   3. CDMA 1x EV-DO Downlink, 20060126 tcpdump data at the iperf sender (32 connections, 4.5 minute iperf, NO TCP timestamp)
   4. CDMA 1x EV-DO Downlink, 20060127 tcpdump data at the iperf sender (59 connections, 4.5 minute iperf, NO TCP timestamp)

tcpdump format

[Author] Youngseok Lee

top

[Paper] lee-tcp-cdma

top

This paper investigates the long-lived TCP bulk throughput over the CDMA 1x 
EV-DO service that provides high-speed \always on" Internet connectivity in a 
wide-area mobile environment. Although the peak rates of downlink/uplink are 
speci ed as 2.4 Mbps/153 Kbps, the user-experienced application-layer 
throughput has not been much reported and analyzed. In our experiment, it was 
shown that average TCP throughputs over downlink/uplink are 572.5/94.7Kbps and 
the average packet loss rates of 1x EV-DO downlink/uplink are 0.2/4.7%. The 
average end-to-end round-trip delay was 417.4ms with the variance of 14,995ms. 
Although the packet loss rate is low, bursty packet losses frequently occur 
because of packet corruption with TCP checksum failures, which result in TCP 
performance degradation by the retransmission timeout. Our study showed that 
this TCP checksum errors are related with the TCP/IP header compression 
algorithm at link layer protocols such as PPP. Our measurement-based analysis 
of TCP performance could be used for the correct model of the 3G wireless link 
characteristic and for the real-world simulation of TCP behavior over the 3G 
wireless network.