Re: [gem5-users] Parsec 2.1 + Garnet 2.0: Too Slow!!!

[F. A. Faisal]

Now, I tried the splash-2 benchmarks and even the parsec 2.1(Alpha +
Mesi_two_level) in both cases, I can get results within very short time if
I ignore the Garent network.

When I run the  fft (splash2) with garnet options, I found that program get
hanged after line- "SMP starting up secondaries" and nothing after that.

In case of parsec 2.1 (other than blackscholes), all the giving the "INFO:
task vips:894 blocked for more than 120 seconds" and running for long long
time and can't getting out of loop.

Please help me.

Thanks in advance.

F. A. Faisal

On Fri, Jun 23, 2017 at 1:51 PM, F. A. Faisal  wrote:

> I could obtain the blackscholes performance within 3hrs using
> MESI_Two_Level. However, MOESI_CMP_directory is very slow and requires more
> than 3hrs before starting the  ROI section.
>
> I recompiled the program as you mentioned and I found the system is
> showing some INFO message through the telnet localhost (However, I
> couldn't find such INFO message in case of blackscholes) and still running
> for about 24 hrs-
>
> Linux version: Scientific Linux release 6.1 (Carbon)
>
>
> # scons FULL_SYSTEM=1 build/ALPHA/gem5.fast RUBY=true
> PROTOCOL=MESI_Two_Level
>
> # ./build/ALPHA/gem5.fast configs/example/fs.py --cpu-type=timing
> --script=script/vips_16c_simsmall.rcS --mem-size=1GB --num-cpus=16 --ruby
> --mesh-rows=4 --network=garnet2.0 --topology=Mesh_XY  --caches --l2cache
> --l2_size=16MB --num-dirs=16 --num-l2=16 --l1d_size=64kB --l1i_size=64kB
> --l1d_assoc=2 --l1i_assoc=2 --disk-image=/home1/xxx/gem5/
> system/alpha/m5_system_2.0b3/disks/linux-parsec-2-1-m5.img
>
> INFO: task vips:894 blocked for more than 120 seconds.
>
> "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message.
>
> vips  D  0   894871
>
> fc003f073da8 0001 fc36e078 fc003f3ce740
>
>003b9ac9 0003 fc35a16c 
>
>fc36ddc0 fc86bf40 fc81bf40 fc35a0f0
>
>fc819040 fc003f063ea8 fc63 fc811b80
>
>fc8113b8 fc3111d0 fc3111e8 fc003f05fd40
>
>fc003f05fce8 fc821760  
>
> Trace:
>
> [] watchdog+0x2b8/0x320
>
> [] kthread+0x7c/0xd0
>
> [] watchdog+0x0/0x320
>
> [] kthread+0x0/0xd0
>
> [] kernel_thread+0x10/0x90
> [] kernel_thread+0x28/0x90
>
>  Thanks in advance.
>
> F. A. Faisal
>
> On Wed, Jun 21, 2017 at 10:35 PM, Krishna, Tushar 
> wrote:
>
>> Hi Faisal,
>> I see you are running a 256 threaded vips on the 16 core system. Maybe
>> try a 16 threaded version?
>> Also, can you try running without garnet -- I'm curious to see how long
>> the benchmarks take in that case.
>>
>> Cheers,
>> Tushar
>>
>>
>> On Jun 21, 2017, at 3:07 AM, F. A. Faisal  wrote:
>>
>> Dear All,
>>
>> I tried to simulate the Parsec 2.1 benchmarks with garnet 2.0.
>> I found that other than Blackscholes, rest of benchmarks are taking
>> several days (and still executing) and even I tried with increasing the
>>  num of benchmarks, which actually don't have too much effect.
>> However, the actual executions time should be in few hours.
>> Please help me to increase the execution rate and how to know, how many
>> percentage of benchmark program is being finished.
>>
>> My simlutation conditions-
>>
>> # scons FULL_SYSTEM=1 build/ALPHA/gem5.fast RUBY=true
>> PROTOCOL=MESI_Two_Level
>>
>> # ./build/ALPHA/gem5.fast configs/example/fs.py --cpu-type=timing
>> --script=script/vips_256c_simsmall.rcS  --num-cpus=16 --mem-size=1GB
>> --ruby --mesh-rows=4 --network=garnet2.0 --topology=Mesh_XY  --caches
>> --l2cache --l2_size=1MB --num-dirs=16 --num-l2=16 --l1d_size=32kB
>> --l1i_size=32kB --l1d_assoc=2 --l1i_assoc=2 --disk-image=/home1/faisal/gem
>> 5/system/alpha/m5_system_2.0b3/disks/linux-parsec-2-1-m5.img
>>
>> Thanks in advance...
>>
>> Best regards,
>>
>> F. A. Faisal
>>
>> ___
>> gem5-users mailing list
>> gem5-users@gem5.org
>> http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users
>>
>>
>> ___
>> gem5-users mailing list
>> gem5-users@gem5.org
>> http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users
>>
>
>
___
gem5-users mailing list
gem5-users@gem5.org
http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users

Re: [gem5-users] Parsec 2.1 + Garnet 2.0: Too Slow!!!

[F. A. Faisal]

I could obtain the blackscholes performance within 3hrs using
MESI_Two_Level. However, MOESI_CMP_directory is very slow and requires more
than 3hrs before starting the  ROI section.

I recompiled the program as you mentioned and I found the system is showing
some INFO message through the telnet localhost (However, I couldn't find
such INFO message in case of blackscholes) and still running for about 24
hrs-

Linux version: Scientific Linux release 6.1 (Carbon)


# scons FULL_SYSTEM=1 build/ALPHA/gem5.fast RUBY=true
PROTOCOL=MESI_Two_Level

# ./build/ALPHA/gem5.fast configs/example/fs.py --cpu-type=timing
--script=script/vips_16c_simsmall.rcS --mem-size=1GB --num-cpus=16 --ruby
--mesh-rows=4 --network=garnet2.0 --topology=Mesh_XY  --caches --l2cache
--l2_size=16MB --num-dirs=16 --num-l2=16 --l1d_size=64kB --l1i_size=64kB
--l1d_assoc=2 --l1i_assoc=2
--disk-image=/home1/xxx/gem5/system/alpha/m5_system_2.0b3/disks/linux-parsec-2-1-m5.img

INFO: task vips:894 blocked for more than 120 seconds.

"echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message.

vips  D  0   894871

fc003f073da8 0001 fc36e078 fc003f3ce740

   003b9ac9 0003 fc35a16c 

   fc36ddc0 fc86bf40 fc81bf40 fc35a0f0

   fc819040 fc003f063ea8 fc63 fc811b80

   fc8113b8 fc3111d0 fc3111e8 fc003f05fd40

   fc003f05fce8 fc821760  

Trace:

[] watchdog+0x2b8/0x320

[] kthread+0x7c/0xd0

[] watchdog+0x0/0x320

[] kthread+0x0/0xd0

[] kernel_thread+0x10/0x90
[] kernel_thread+0x28/0x90

 Thanks in advance.

F. A. Faisal

On Wed, Jun 21, 2017 at 10:35 PM, Krishna, Tushar 
wrote:

> Hi Faisal,
> I see you are running a 256 threaded vips on the 16 core system. Maybe try
> a 16 threaded version?
> Also, can you try running without garnet -- I'm curious to see how long
> the benchmarks take in that case.
>
> Cheers,
> Tushar
>
>
> On Jun 21, 2017, at 3:07 AM, F. A. Faisal  wrote:
>
> Dear All,
>
> I tried to simulate the Parsec 2.1 benchmarks with garnet 2.0.
> I found that other than Blackscholes, rest of benchmarks are taking
> several days (and still executing) and even I tried with increasing the
>  num of benchmarks, which actually don't have too much effect.
> However, the actual executions time should be in few hours.
> Please help me to increase the execution rate and how to know, how many
> percentage of benchmark program is being finished.
>
> My simlutation conditions-
>
> # scons FULL_SYSTEM=1 build/ALPHA/gem5.fast RUBY=true
> PROTOCOL=MESI_Two_Level
>
> # ./build/ALPHA/gem5.fast configs/example/fs.py --cpu-type=timing
> --script=script/vips_256c_simsmall.rcS  --num-cpus=16 --mem-size=1GB
> --ruby --mesh-rows=4 --network=garnet2.0 --topology=Mesh_XY  --caches
> --l2cache --l2_size=1MB --num-dirs=16 --num-l2=16 --l1d_size=32kB
> --l1i_size=32kB --l1d_assoc=2 --l1i_assoc=2 --disk-image=/home1/faisal/
> gem5/system/alpha/m5_system_2.0b3/disks/linux-parsec-2-1-m5.img
>
> Thanks in advance...
>
> Best regards,
>
> F. A. Faisal
>
> ___
> gem5-users mailing list
> gem5-users@gem5.org
> http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users
>
>
> ___
> gem5-users mailing list
> gem5-users@gem5.org
> http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users
>
___
gem5-users mailing list
gem5-users@gem5.org
http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users

Re: [gem5-users] Parsec 2.1 + Garnet 2.0: Too Slow!!!

[Krishna, Tushar]

Hi Faisal,
I see you are running a 256 threaded vips on the 16 core system. Maybe try a 16 
threaded version?
Also, can you try running without garnet -- I'm curious to see how long the 
benchmarks take in that case.

Cheers,
Tushar


On Jun 21, 2017, at 3:07 AM, F. A. Faisal 
> wrote:

Dear All,

I tried to simulate the Parsec 2.1 benchmarks with garnet 2.0.
I found that other than Blackscholes, rest of benchmarks are taking several 
days (and still executing) and even I tried with increasing the  num 
of benchmarks, which actually don't have too much effect.
However, the actual executions time should be in few hours.
Please help me to increase the execution rate and how to know, how many 
percentage of benchmark program is being finished.

My simlutation conditions-

# scons FULL_SYSTEM=1 build/ALPHA/gem5.fast RUBY=true PROTOCOL=MESI_Two_Level

# ./build/ALPHA/gem5.fast configs/example/fs.py --cpu-type=timing 
--script=script/vips_256c_simsmall.rcS  --num-cpus=16 --mem-size=1GB --ruby 
--mesh-rows=4 --network=garnet2.0 --topology=Mesh_XY  --caches --l2cache 
--l2_size=1MB --num-dirs=16 --num-l2=16 --l1d_size=32kB --l1i_size=32kB 
--l1d_assoc=2 --l1i_assoc=2 
--disk-image=/home1/faisal/gem5/system/alpha/m5_system_2.0b3/disks/linux-parsec-2-1-m5.img

Thanks in advance...

Best regards,

F. A. Faisal
___
gem5-users mailing list
gem5-users@gem5.org
http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users
___
gem5-users mailing list
gem5-users@gem5.org
http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users