Re: [Gems-users] Notary paper results

[Luke Yen <lyen@xxxxxxxxxxx>]

Hi Ricardo,

   I've been busy writing my dissertation, so I haven't gotten a chance to 
respond to your email.
   Here are some major params I used:

       MESI_CMP_filter_directory, no virtualization support enabled.  I 
bound all threads to separate processors, and used a 16p system.
      The H3 results uses 2 hash functions for Parallel signatures (e.g., 
H3_2048_2_Parallel)
      My input to Labyrinth is random-x32-y32-z3-n64.txt. I attached it 
to this email.
       I used eager conflict detection and eager version management with 
"magic" waiting after aborts (set ENABLE_MAGIC_WAITING to true and 
XACT_NO_BACKOFF to true) in the 
Ruby config param.  Some of the STAMP workloads perform worse if exponential backoff 
is used.
       The only possible change I might of made to Labyrinth is to pad 
data structures to avoid false sharing between threads.  This is done by 
running the smallest imperfect signature (e.g., 64 bit signatures) and 
working backwards from the stall stats by PC to see where the stalls occur 
in the code.
    Let me know if these params get it closer to my results.

   Luke

On Tue, 9 Dec 2008, Ricardo Quislant del Barrio wrote:

> Hi all,
> I'm trying to reproduce the results you obtained for Labyrinth benchmark
> in "Notary" paper
> and I get the following:
>
> -Perfect filter: 202 millions of Ruby_cycles
> -H3_2048_4_Regular: 256 millions of Ruby_cycles
> -H3_1024_4_Regular: 55 millions of Ruby_cycles
> -H3_512_4_Regular: 95 millions of Ruby_cycles
> -H3_256_4_Regular: 24 millions of Ruby_cycles
> -H3_128_4_Regular: 29 millions of Ruby_cycles
> -H3_64_4_Regular: 27 millions of Ruby_cycles
>
> Ruby parameters:
>    - protocol: MESI_CMP_filter_directory, logtm_se_virtualization
>    - random_seed: 213
>    - labyrinth input: -t15 -i inputs/random-x32-y32-z3-n16.txt
>
> I'm not using the same number of hash functions neither the same input
> to labyrinth (I guess you're using "random-x48-y48-z3-n64.txt") but I
> was expecting consistent results like yours (not 10x more ruby_cycles
> for Perfect_filter than for H3_64_4_Regular).
>
> I've posted about that but you didn't advice me how to get consistent
> results.
> Could you give me some clues on how to get results like yours (protocol,
> paramaters, labyrinth modifications,...) ?
>
> Thank you very much,
> Ricardo
> _______________________________________________
> 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.
# Dimensions (x, y, z)
d  32 32 3

# Paths: Sources (x, y, z) -> Destinations (x, y, z)
p    27  24 1     8  16 1
p    25   9 1    18  29 1
p     9  24 1     8  29 2
p    25  28 0    23  28 2
p    15   3 1    19  29 2
p    15  27 0    25  17 0
p    23  12 2    21   0 1
p    27   7 0    27   6 1
p     7  30 2    14   2 0
p    16  29 0    17  22 1
p    26  17 2    19  18 2
p    19  12 1     9   6 0
p    19  21 1     2  24 2
p    29  26 2    29  17 1
p    22   8 2    27  28 1
p    30  18 1    21  31 2
p    25   2 1    16  20 2
p     7  23 0     7  25 0
p    26   3 0    22   1 1
p    29  17 2     0  20 1
p    18  12 1    31   1 0
p    30   5 0     6  25 2
p     0  13 0     8   7 1
p    11   5 1     1   3 2
p     6  11 2    26  29 0
p    21  30 0    21  27 1
p     8  19 1     5  15 1
p    18  16 0    11  26 2
p    17   0 2    10   1 0
p     7  30 1     9  11 2
p    20  19 2    12  13 2
p     0   6 1     7  20 1
p    28  18 1    12  22 1
p    21   1 1     8   5 1
p    15  17 2    28  15 0
p    14  25 2    25   6 2
p    20   2 2    31  12 2
p    10   6 2     0  26 1
p     3   3 1    27   8 2
p     3  27 1     2   8 1
p    25  27 0    16  20 1
p    27   8 0     1  21 1
p    30  30 2     1  23 2
p    20  22 2    20  11 1
p     6  18 0     4  10 2
p    22  10 1     1   5 2
p     9  12 1     9  15 1
p     1   5 1     7  12 2
p    13   3 2    15  26 2
p    10  15 2    13   9 2
p    28  29 1    12  31 1
p     2   2 2     1   0 1
p    16  14 1    18  21 1
p    11  31 0    24  13 1
p     2  27 2    28  14 2
p     3  12 0     1  30 0
p     4   6 1    17   4 2
p    31   4 1    21  28 1
p    29  10 1    15  21 0
p    19   7 1    30  28 2
p     1   4 0    25  26 1
p    22  25 0     2  27 0
p     7   1 0    27  10 0
p     4  20 2    16  28 1