rev_read operation slowdown after large number of lines read about rev HOT 2 CLOSED

I cannot reproduce. What was your .py config? Are you using a Release build of Rev?

With 20,000 lines around 80 characters wide each:

real    0m3.999s
user    0m3.240s
sys     0m0.601s

With 50,000 lines around 80 characters wide each:

real    0m8.242s
user    0m7.434s
sys     0m0.652s

8.242 / 3.999 = 2.061

50000.0 / 20000.0 = 2.5

So the 50,000 line one is actually much faster per-line than the 20,000 line one.

That was with RevMem. Here it is for MemH 20,000 lines:

real    1m45.388s
user    1m45.139s
sys     0m0.084s

MemH 50,000 lines:

real    4m20.000s
user    4m19.723s
sys     0m0.100s

260.000 / 105.388 = 2.467, so like RevMem, MemH is faster per-line with 50,000 lines, but just marginally.

The above results are reproducible with SST 13.1.0.

from rev.

This test is not reproducible, and after communicating with Luke, he cannot reproduce it either, with a loop which simply reads and prints lines.

We suspect it may be something to do with his tuple parsing code, which this reduced test case has removed, rather than rev_read().

I asked for a profiler run to be done to identify the hotspots, but Luke is too busy with other tasks now.

from rev.