Check out the new USENIX Web site.
...\title1
This work was supported by the U.S. Dept. of Energy's Laboratory-Directed Research & Development Program and the Los Alamos Computer Science Institute through Los Alamos National Laboratory contract W-7405-ENG-36. Any opinions, findings, and conclusions, or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of DOE, Los Alamos National Laboratory, or the Los Alamos Computer Science Institute. This paper is Los Alamos Unclassified Report (LA-UR) 01-5065.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
... stream,2
The sockid field is, in fact, the run-time value of the pointer to the kernel's status information for the specific connection.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
... event.3
A negative value in the size field reflects the error code returned by the function causing the event.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
... file4
The major and minor numbers for this device are system specific and can be discovered by inspecting /proc/net/magnet.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
... archive.5
Since magnet.copy is called while magnet-read is not running, any traffic produced by the data archiving will not be captured by MAGNeT. This behavior can easily be changed by re-ordering the commands in magnet.cron.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
... possible.6
The command used was ``netperf -P 0 -c {local CPU index} -C {remote CPU index} -H {hostname}''
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
... small. 7
Of course, with 1000-Mbps networks, the time taken to send a minimal TCP packet is reduced by an order of magnitude. In this case, MAGNeT takes significantly longer to record a packet than the transmission time. However, it is likely that Gigabit networks will be used with much faster machines than our test machines, thus reducing this discrepancy.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
... paper,8
By utilizing the data field of the instrumentation record, MAGNeT is able to capture much more detailed traces but still not actual packet data.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
... available.9
Since MAGNeT never collects actual data (only traffic patterns), the privacy of each individual machine is maintained.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
... run-time. 10
Dynamic CPU clock changing is also an issue with many aspects of the Linux kernel. For instance, short delay loops in the kernel rely on $x$ number of loops taking $y$ amount of time to compute. This relation is also affected by CPU clock-rate changes.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.