out-of order data…

$cat /proc/sys/net/ipv4/tcp_reordering
tcp_reordering - INTEGER
Maximal reordering of packets in a TCP stream.
Default: 3
  1921:         tp->reordering = sysctl_tcp_reordering;
nms.csail.mit.edu/~kandula/data/tcp-mult.tgz - Unknown - C

   271:         fprintf (script, "# Set TCP Re-Ordering value in kernel to '5'\n");
   272:         fprintf (script, "if [ -e /proc/sys/net/ipv4/tcp_reordering ]; then\n"
   273:         "  echo 5 > /proc/sys/net/ipv4/tcp_reordering\nfi\n\n");
packetstormsecurity.nl/.../firewall/firestarter/firestarter-0.9.0.tar.gz - GPL - C

 64934: .LC2778:
 64935:         .string "NET_TCP_REORDERING"
 64936: .LC873:
de.samba.org/samba/ftp/cifs-cvs/cifs-1.13-2.6-bad.tar.gz - Unknown - Assembly

• Load splitting: To balance the load among the multiple
paths, different packets of the same stream take different
routes leading to different delays causing reordering.
Problems caused by reordering are handled at different
levels in TCP/IP suite. TCP allows adjustment of ‘dupthresh’
parameter, i.e., the number of duplicate ACKs to be allowed
before classifying a following non-acknowledged packet as
lost [3]. This parameter (also called tcp_reordering in Linux
implementations) allows the reordering to occur to a certain
extent without affecting the throughput. At application level,
the out-of-sequence packets are buffered until they can be
played back in sequence. An increase in out-of-order delivery
however consumes more resources and also affects the end-to-
end performance. Consequently, certain techniques attempt to
reduce reordering at intermediate nodes, i.e., at IP level.

A Comparative Analysis of Packet Reordering Metrics*
Nischal M. Piratla1, 2
Anura P. Jayasumana1 Abhijit Bare1
Computer Networking Research Laboratory, Colorado State University, Fort Collins, CO 80523, USA
Deutsche Telekom Laboratories, Ernst-Reuter-Platz 7, D-10587 Berlin, Germany