NAME¶
h_ertt
—
Enhanced Round Trip Time Khelp module
SYNOPSIS¶
#include
<netinet/khelp/h_ertt.h>
DESCRIPTION¶
The
h_ertt
Khelp module works within the
khelp(9) framework to provide TCP with a
per-connection, low noise estimate of the instantaneous RTT. The
implementation attempts to be robust in the face of delayed acknowledgements,
TCP Segmentation Offload (TSO), receivers who manipulate TCP timestamps and
lack of the TCP timestamp option altogether.
TCP receivers using delayed acknowledgements either acknowledge every second
packet (reflecting the time stamp of the first) or use a timeout to trigger
the acknowledgement if no second packet arrives. If the heuristic used by
h_ertt
determines that the receiver is
using delayed acknowledgements, it measures the RTT using the second packet
(the one that triggers the acknowledgement). It does not measure the RTT if
the acknowledgement is for the first packet, since it cannot be accurately
determined.
When TSO is in use,
h_ertt
will momentarily
disable TSO whilst marking a packet to use for a new measurement. The process
has negligible impact on the connection.
h_ertt
associates the following struct with
each connection's TCP control block:
struct ertt {
TAILQ_HEAD(txseginfo_head, txseginfo) txsegi_q; /* Private. */
long bytes_tx_in_rtt; /* Private. */
long bytes_tx_in_marked_rtt;
unsigned long marked_snd_cwnd;
int rtt;
int maxrtt;
int minrtt;
int dlyack_rx; /* Private. */
int timestamp_errors; /* Private. */
int markedpkt_rtt; /* Private. */
uint32_t flags;
};
The fields marked as private should not be manipulated by any code outside of
the
h_ertt
implementation. The non-private
fields provide the following data:
- bytes_tx_in_marked_rtt
- The number of bytes transmitted in the
markedpkt_rtt.
- marked_snd_cwnd
- The value of cwnd for the marked rtt measurement.
- rtt
- The most recent RTT measurement.
- maxrtt
- The longest RTT measurement that has been taken.
- minrtt
- The shortest RTT measurement that has been taken.
- flags
- The ERTT_NEW_MEASUREMENT flag will be set by the implementation when a new
measurement is available. It is the responsibility of
h_ertt
consumers to unset the flag if
they wish to use it as a notification method for new measurements.
SEE ALSO¶
cc_chd(4),
cc_hd(4),
cc_vegas(4),
mod_cc(4),
hhook(9),
khelp(9)
ACKNOWLEDGEMENTS¶
Development and testing of this software were made possible in part by grants
from the FreeBSD Foundation and Cisco University Research Program Fund at
Community Foundation Silicon Valley.
HISTORY¶
The
h_ertt
module first appeared in
FreeBSD 9.0.
The module was first released in 2010 by David Hayes whilst working on the
NewTCP research project at Swinburne University of Technology's Centre for
Advanced Internet Architectures, Melbourne, Australia. More details are
available at:
http://caia.swin.edu.au/urp/newtcp/
AUTHORS¶
The
h_ertt
Khelp module and this manual page
were written by
David Hayes
⟨david.hayes@ieee.org⟩.
BUGS¶
The module maintains enhanced RTT estimates for all new TCP connections created
after the time at which the module was loaded. It might be beneficial to see
if it is possible to have the module only affect connections which actually
care about ERTT estimates.