Re: [Gems-users] Collecting Ruby stats for TM

[Dan Gibson <degibson@xxxxxxxx>]

These aren't the transactions you're looking for.

The transaction stats you are looking at are stats for 'workload' transactions, not 'transactional memory' transactions. Admittedly, the stat is badly named, but was badly named before the TM craze, so it wasn't changed post-craze.

In the past, we have modified workloads to notify Ruby that progress is being made through special 'magic' instructions that signal 'transaction' boundaries (e.g., DB transaction boundaries, requests serviced in HTTP servers, etc.) or outer-loop iterations (in scientific workloads). Since your workload isn't executing any of these 'workload transactions', your profiling of this flavor of 'transaction' reads as zero. These transactions are totally unrelated to TM transactions.

Look around at the LogTM stats (I have no idea what they are named, sorry) to find the transactions you're looking for. 

Regards,
Dan

On Sun, Nov 29, 2009 at 10:31 PM, Polina Dudnik <pdudnik@xxxxxxxxx> wrote:

Turn on all the debugging and make sure that your profiling flags are also on. If you turn on the debugging you should see the print out that says execution is in begin transaction, end transaction and so forth. Make sure that transactional mode is enabled and you are using the correct protocol (MESI_CMP_filter_directory).

On Sun, Nov 29, 2009 at 10:32 PM, Jiří Šimša <jiri.simsa@xxxxxxxxx> wrote:

Hello,

I was able to get the deque microbenchmark to run on a 4 processor Simics checkpoint. Next, I tried collecting information about the use of transactions using Ruby as follows:

I started Simics with my checkpoint that has the binary of deque_TM loaded in and everything is ready to be run.
Next, I did the following in Simics:
simics> instruction-fetch-mode instruction-fetch-trace; dstc-disable; istc-disable; cpu-switch-time 1; load-module ruby
[cpu0 info] Note that on this cpu, instruction-fetch-trace is implemented using instruction-cache-access-trace with a suitable cache line size.
[cpu1 info] Note that on this cpu, instruction-fetch-trace is implemented using instruction-cache-access-trace with a suitable cache line size.
[cpu4 info] Note that on this cpu, instruction-fetch-trace is implemented using instruction-cache-access-trace with a suitable cache line size.
[cpu5 info] Note that on this cpu, instruction-fetch-trace is implemented using instruction-cache-access-trace with a suitable cache line size.
Turning D-STC off and flushing old data
Turning I-STC off and flushing old data
The switch time will change to 1 cycles (for CPU-0) once all processors have synchronized.
successful installation of the ruby timing model.
simics> ruby0.setparam g_NUM_PROCESSORS 4
simics> ruby0.init

Then I ran the benchmark.

Finally, I dumped the stats using ruby0.dump_short_stats. Producing:

simics> ruby0.dump-short-stats

SHORT Profiler Stats
--------------
Virtual_time_in_seconds: 878.31
Virtual_time_in_minutes: 14.6385
Virtual_time_in_hours:   0.243975
Virtual_time_in_days:    0.243975

Ruby_current_time: 24374620
Ruby_start_time: 1
Ruby_cycles: 24374619

Total_misses: 160366
total_misses: 160366 [ 9456 2310 1851 146749 ]
user_misses: 12822 [ 617 529 548 11128 ]
supervisor_misses: 147544 [ 8839 1781 1303 135621 ]

instruction_executed: 83469212 [ 24984755 25572397 25627341 7284719 ]
simics_cycles_executed: 389994915 [ 97498479 97498479 97498479 97499478 ]
cycles_per_instruction: 1.16808 [ 0.97558 0.953161 0.951118 3.34599 ]
misses_per_thousand_instructions: 1.92126 [ 0.378471 0.0903318 0.0722275 20.1448 ]

transactions_started: 0 [ 0 0 0 0 ]
transactions_ended: 0 [ 0 0 0 0 ]
instructions_per_transaction: 0 [ 0 0 0 0 ]
cycles_per_transaction: 0 [ 0 0 0 0 ]
misses_per_transaction: 0 [ 0 0 0 0 ]

...

Anyone could explain to me why is the information about transactions missing? Is there a simple check that I can run to see if the transactional memory subsystem is actually used? Thanks!

Best,

--
Jiri Simsa
Ph.D. Student
Carnegie Mellon University

_______________________________________________
Gems-users mailing list
Gems-users@xxxxxxxxxxx
https://lists.cs.wisc.edu/mailman/listinfo/gems-users
Use Google to search the GEMS Users mailing list by adding "site:https://lists.cs.wisc.edu/archive/gems-users/" to your search.

_______________________________________________
Gems-users mailing list
Gems-users@xxxxxxxxxxx
https://lists.cs.wisc.edu/mailman/listinfo/gems-users
Use Google to search the GEMS Users mailing list by adding "site:https://lists.cs.wisc.edu/archive/gems-users/" to your search.

-- 
http://www.cs.wisc.edu/~gibson [esc]:wq!