[802.3AZ] traces for our project
Based on feedback I've received over the past week, here are the
conditions I think we should use for collecting trace files:
At a minimum, we only need to capture enough of the frame to get the
time-stamp and frame size (frame capture length is 14 Bytes). If
someone is willing to share a trace file that includes more information,
e.g. TCP header (note that the minimum capture length would be 34 Bytes)
we will make it available, assuming we have the storage capacity to do so.
Trace files are to be submitted in the packet capture library (pcap)
file format.
Trace periods should be 1 hour long. We need enough 1-hour traces to
get a sense of the nature of the network utilization for a given
application. 24 consecutive hours would be ideal.
Since most of our discussion has been focused on data centers, I think
the following use cases would provide the most value:
High-performance compute clusters, e.g. visualization, complex modeling,
etc.
Networked storage
Database servers
Client switch aggregation ports
Server aggregation ports
Data-center switch aggregation ports
With respect to our discussions on AVB, traces from home and office PC
connections, video and audio streams, VOIP calls, and Video on Demand
would be useful.
Of course we could still use traces from enterprise routers, WAN/ISP
links, enterprise application servers e.g. web, e-mail, and video
hosting, but these have the lowest priority.
Even though some people in the task force have been using traces from
the last category, I think it would be useful for the task for to use
the same traces for simulations and comparisons.
Meta data on the packet captures will be provided in a text file and
will include:
* Date of collection
* Name and company of individual who collected the trace
* A description of the capture method, e.g tcpdump from linux
machine, pcap file from hardware analyzer, capture settings such
as capture length, and if there were any packets lost during the
capture. If the capture comes from a hardware analyzer,
information about the clock accuracy should be provided.
* Link speed and type, e.g. 1000BASE-T, 10BGASE-LR (since we can't
get 10GBASE-T production traffic yet), and MTU
* Type of devices at the ends of the link, e.g., router-switch,
desktop-switch, switch-switch, etc.
* General usage, such as department file server, desktop PC in
office, HPC cluster node, etc. This should also include a comment
regarding the measure workload and how it compares to the typical
work-lload for the measured system.
LBNL will host the files and make them available. I will send the
details for this as soon as we have it ready.
This is what I propose we do. If you have anything to add or change
please let me know by the end of the day Monday, May 5.
Regards,
Mike