etl
etl copied to clipboard
MinRTT missing from a large fraction of Web100 data 2009-2017
Diagnose root cause (which may be a bug in the underlying TCP implementation) and consider workarounds.
See OkRTT in mlab-sandbox.inspector.inspect_public_unified
@mattmathis is this still happening?
90+% of the tests 2009-2012 report suspect MinRTT 70+% of the tests 2013-2015 report suspect MinRTT 35% of 2016 and 14% of 2017 report suspect MinRTT Where suspect MinRTT means either 0 (e.g. missing) or greater than 500 mS (unlikely).
One possible workaround would be to scan all web100 snaps and report the smallest non-zero value for MinRTT. Since the actual MinRTT is almost always seen at the beginning and the zero bug was caused by a bogus measurement later in the connection, most of the MinRTT measurements from the beginning of the test would be valid.