Bug 29959 - Investigate parallel reader scalability
Summary: Investigate parallel reader scalability
Status: NEW
Alias: None
Product: gdb
Classification: Unclassified
Component: symtab (show other bugs)
Version: HEAD
: P2 normal
Target Milestone: ---
Assignee: Not yet assigned to anyone
URL:
Keywords:
Depends on:
Blocks: 29366
  Show dependency treegraph
 
Reported: 2023-01-03 14:04 UTC by Tom Tromey
Modified: 2025-01-12 18:55 UTC (History)
0 users

See Also:
Host:
Target:
Build:
Last reconfirmed:


Attachments
screen shot from gnuplot (25.36 KB, image/png)
2023-01-03 14:04 UTC, Tom Tromey
Details
timing data (981 bytes, text/plain)
2023-01-03 14:05 UTC, Tom Tromey
Details

Note You need to log in before you can comment on or make changes to this bug.
Description Tom Tromey 2023-01-03 14:04:10 UTC
Created attachment 14548 [details]
screen shot from gnuplot

I was curious about the scalability of the parallel reader,
so I built on gcc112, which reports 160 processors.

I made a file like:

maint time 1
file /tmp/gdb

Then I ran:

for j in $(seq 0 180); do echo ======= $j ; ./gdb/gdb -nh -quiet -iex "maint set worker-threads $j" < S; done

I had it go over 160 just to see if that was interesting or not.

To my surprise, the results don't seem to improve past N=8 or so.
There are multiple possible explanations for this, but it would
be worthwhile to find out exactly why.
Comment 1 Tom Tromey 2023-01-03 14:05:24 UTC
Created attachment 14549 [details]
timing data
Comment 2 Tom Tromey 2025-01-12 18:55:36 UTC
One recent idea here is that the DWARF reader could just limit
itself to 8 jobs; then with the background reader enabled, this
would allow more parallel work on large machines.  That is,
try to take advantage of the lack of scalability.