gem5-users@gem5.org

The gem5 Users mailing list

View all threads

回复:Re: 回复:Re: Different simulation results on different computers with the same configuration

2
2497597
Tue, Sep 20, 2022 3:04 PM

The correct result is below:

average_packet_latency = 15.510408
average_packet_latency = 15.528615
average_packet_latency = 15.682214
average_packet_latency = 15.695504
average_packet_latency = 15.769957
average_packet_latency = 15.821728
average_packet_latency = 15.912262
average_packet_latency = 16.051925
average_packet_latency = 16.167249
average_packet_latency = 16.319634
average_packet_latency = 16.479105
average_packet_latency = 16.725313
average_packet_latency = 17.055812
average_packet_latency = 17.588959
average_packet_latency = 18.500431
average_packet_latency = 21.669417
average_packet_latency = 103.241365
average_packet_latency = 273.002675
average_packet_latency = 430.695013
average_packet_latency = 596.634683
average_packet_latency = 732.220679
average_packet_latency = 854.214438
average_packet_latency = 969.032975
average_packet_latency = 1087.468352
average_packet_latency = 1207.588344

1.gem5 version is v21.2.1.0

2.build command is ”scons build/NULL/gem5.opt -j 4"

3.run command is below and the "--injection-rate" is increasing with step=0.02

Command to Run:
./build/NULL/gem5.opt
configs/example/garnet_synth_traffic.py \
--network=garnet2.0 \
--num-cpus=64 \
--num-dirs=64 \
--topology=Mesh_XY \
--mesh-rows=8 \
--sim-cycles=10000 \
--inj-vnet=0 \
--injectionrate=0.02 \
--synthetic=uniform_random

I find some interesting phenomenons.
I think I figured out where the problem might be.
It might be the permission error and the build order.
I built "X86","ARM","NULL" on my old computer before,so I try to restore it all.
Then I find the miss folder "pycache" come out,I think I'm going to get close to the root of the problem.
But I failed.
I rebuild a same version gem5 on my new computer with the command"”scons build/Garnet_standalone/gem5.opt -j 4" under the root state.
Then I run the same script,I get the correct output.(I can see the average_packet_latency is increasing with the --injectionrate)
Although I get the desired output, I can't figure out why I get the wrong output.

------------------ 原始邮件 ------------------
发件人:                                                                                                                        "The gem5 Users mailing list"                                                                                    <gabriel.busnot@arteris.com>;
发送时间: 2022年9月20日(星期二) 晚上7:40
收件人: "gem5-users"<gem5-users@gem5.org>;

主题: [gem5-users] Re: 回复:Re: Different simulation results on different computers with the same configuration

Not sure to understand what you mean by “correct result".

Can you please provide us with:

gem5 version (commit SHA)

build command

run command

Can you please also double-check that you are using supported versions of python and gcc/clang. Also check that config.ini is the same in all cases.

Gabriel

The correct result is below: average_packet_latency = 15.510408 average_packet_latency = 15.528615 average_packet_latency = 15.682214 average_packet_latency = 15.695504 average_packet_latency = 15.769957 average_packet_latency = 15.821728 average_packet_latency = 15.912262 average_packet_latency = 16.051925 average_packet_latency = 16.167249 average_packet_latency = 16.319634 average_packet_latency = 16.479105 average_packet_latency = 16.725313 average_packet_latency = 17.055812 average_packet_latency = 17.588959 average_packet_latency = 18.500431 average_packet_latency = 21.669417 average_packet_latency = 103.241365 average_packet_latency = 273.002675 average_packet_latency = 430.695013 average_packet_latency = 596.634683 average_packet_latency = 732.220679 average_packet_latency = 854.214438 average_packet_latency = 969.032975 average_packet_latency = 1087.468352 average_packet_latency = 1207.588344 1.gem5 version is v21.2.1.0 2.build command is ”scons build/NULL/gem5.opt -j 4" 3.run command is below and the "--injection-rate" is increasing with step=0.02 Command to Run: ./build/NULL/gem5.opt configs/example/garnet_synth_traffic.py \ --network=garnet2.0 \ --num-cpus=64 \ --num-dirs=64 \ --topology=Mesh_XY \ --mesh-rows=8 \ --sim-cycles=10000 \ --inj-vnet=0 \ --injectionrate=0.02 \ --synthetic=uniform_random I find some interesting phenomenons. I think I figured out where the problem might be. It might be the permission error and the build order. I built "X86","ARM","NULL" on my old computer before,so I try to restore it all. Then I find the miss folder "__pycache__" come out,I think I'm going to get close to the root of the problem. But I failed. I rebuild a same version gem5 on my new computer with the command"”scons build/Garnet_standalone/gem5.opt -j 4" under the root state. Then I run the same script,I get the correct output.(I can see the average_packet_latency is increasing with the --injectionrate) Although I get the desired output, I can't figure out why I get the wrong output. ------------------ 原始邮件 ------------------ 发件人: "The gem5 Users mailing list" <gabriel.busnot@arteris.com&gt;; 发送时间:&nbsp;2022年9月20日(星期二) 晚上7:40 收件人:&nbsp;"gem5-users"<gem5-users@gem5.org&gt;; 主题:&nbsp;[gem5-users] Re: 回复:Re: Different simulation results on different computers with the same configuration Not sure to understand what you mean by “correct result". Can you please provide us with: gem5 version (commit SHA) build command run command Can you please also double-check that you are using supported versions of python and gcc/clang. Also check that config.ini is the same in all cases. Gabriel
JL
Jason Lowe-Power
Tue, Sep 20, 2022 3:40 PM

Hello,

The following two command produce very different gem5 binaries

scons build/Garnet_standalone/gem5.opt

The above command will use the Garnet_standalone "coherence" protocol. I
use quotes around coherence because that protocol is essentially an empty
protocol with no coherence. It's meant for testing networks and nothing
else.

scons build/NULL/gem5.opt

This command will use the MI_example protocol. This is a completely
different coherence protocol than Garnet_standalone. See
https://gem5.googlesource.com/public/gem5/+/refs/heads/stable/build_opts/NULL
for how the protocol is defined in the default build options.

You can specify the exact protocol you want when you build gem5 by using
the scons parameter --protocol

Cheers,
Jason

On Tue, Sep 20, 2022 at 8:08 AM 2497597 2497597@qq.com wrote:

The correct result is below:

average_packet_latency = 15.510408
average_packet_latency = 15.528615
average_packet_latency = 15.682214
average_packet_latency = 15.695504
average_packet_latency = 15.769957
average_packet_latency = 15.821728
average_packet_latency = 15.912262
average_packet_latency = 16.051925
average_packet_latency = 16.167249
average_packet_latency = 16.319634
average_packet_latency = 16.479105
average_packet_latency = 16.725313
average_packet_latency = 17.055812
average_packet_latency = 17.588959
average_packet_latency = 18.500431
average_packet_latency = 21.669417
average_packet_latency = 103.241365
average_packet_latency = 273.002675
average_packet_latency = 430.695013
average_packet_latency = 596.634683
average_packet_latency = 732.220679
average_packet_latency = 854.214438
average_packet_latency = 969.032975
average_packet_latency = 1087.468352
average_packet_latency = 1207.588344

1.gem5 version is v21.2.1.0

2.build command is ”scons build/NULL/gem5.opt -j 4"

3.run command is below and the "--injection-rate" is increasing with
step=0.02

Command to Run:
./build/NULL/gem5.opt
configs/example/garnet_synth_traffic.py
--network=garnet2.0
--num-cpus=64
--num-dirs=64
--topology=Mesh_XY
--mesh-rows=8
--sim-cycles=10000
--inj-vnet=0
--injectionrate=0.02
--synthetic=uniform_random

I find some interesting phenomenons.
I think I figured out where the problem might be.
It might be the permission error and the build order.
I built "X86","ARM","NULL" on my old computer before,so I try to restore
it all.
Then I find the miss folder "pycache" come out,I think I'm going to
get close to the root of the problem.
But I failed.
I rebuild a same version gem5 on my new computer with the command"”scons
build/Garnet_standalone/gem5.opt -j 4" under the root state.
Then I run the same script,I get the correct output.(I can see the
average_packet_latency is increasing with the --injectionrate)
Although I get the desired output, I can't figure out why I get the wrong
output.

------------------ 原始邮件 ------------------
发件人: "The gem5 Users mailing list" gabriel.busnot@arteris.com;
发送时间: 2022年9月20日(星期二) 晚上7:40
收件人: "gem5-users"gem5-users@gem5.org;
主题: [gem5-users] Re: 回复:Re: Different simulation results on different
computers with the same configuration

Not sure to understand what you mean by “correct result".

Can you please provide us with:

1.

gem5 version (commit SHA)
2.

build command
3.

run command

Can you please also double-check that you are using supported versions of
python and gcc/clang. Also check that config.ini is the same in all cases.

Gabriel


gem5-users mailing list -- gem5-users@gem5.org
To unsubscribe send an email to gem5-users-leave@gem5.org

Hello, The following two command produce very different gem5 binaries scons build/Garnet_standalone/gem5.opt The above command will use the Garnet_standalone "coherence" protocol. I use quotes around coherence because that protocol is essentially an empty protocol with no coherence. It's meant for testing networks and nothing else. scons build/NULL/gem5.opt This command will use the MI_example protocol. This is a completely different coherence protocol than Garnet_standalone. See https://gem5.googlesource.com/public/gem5/+/refs/heads/stable/build_opts/NULL for how the protocol is defined in the default build options. You can specify the exact protocol you want when you build gem5 by using the scons parameter `--protocol` Cheers, Jason On Tue, Sep 20, 2022 at 8:08 AM 2497597 <2497597@qq.com> wrote: > The correct result is below: > > average_packet_latency = 15.510408 > average_packet_latency = 15.528615 > average_packet_latency = 15.682214 > average_packet_latency = 15.695504 > average_packet_latency = 15.769957 > average_packet_latency = 15.821728 > average_packet_latency = 15.912262 > average_packet_latency = 16.051925 > average_packet_latency = 16.167249 > average_packet_latency = 16.319634 > average_packet_latency = 16.479105 > average_packet_latency = 16.725313 > average_packet_latency = 17.055812 > average_packet_latency = 17.588959 > average_packet_latency = 18.500431 > average_packet_latency = 21.669417 > average_packet_latency = 103.241365 > average_packet_latency = 273.002675 > average_packet_latency = 430.695013 > average_packet_latency = 596.634683 > average_packet_latency = 732.220679 > average_packet_latency = 854.214438 > average_packet_latency = 969.032975 > average_packet_latency = 1087.468352 > average_packet_latency = 1207.588344 > > 1.gem5 version is v21.2.1.0 > > 2.build command is ”scons build/NULL/gem5.opt -j 4" > > 3.run command is below and the "--injection-rate" is increasing with > step=0.02 > > Command to Run: > ./build/NULL/gem5.opt > configs/example/garnet_synth_traffic.py \ > --network=garnet2.0 \ > --num-cpus=64 \ > --num-dirs=64 \ > --topology=Mesh_XY \ > --mesh-rows=8 \ > --sim-cycles=10000 \ > --inj-vnet=0 \ > --injectionrate=0.02 \ > --synthetic=uniform_random > > I find some interesting phenomenons. > I think I figured out where the problem might be. > It might be the permission error and the build order. > I built "X86","ARM","NULL" on my old computer before,so I try to restore > it all. > Then I find the miss folder "__pycache__" come out,I think I'm going to > get close to the root of the problem. > But I failed. > I rebuild a same version gem5 on my new computer with the command"”scons > build/Garnet_standalone/gem5.opt -j 4" under the root state. > Then I run the same script,I get the correct output.(I can see the > average_packet_latency is increasing with the --injectionrate) > Although I get the desired output, I can't figure out why I get the wrong > output. > > ------------------ 原始邮件 ------------------ > *发件人:* "The gem5 Users mailing list" <gabriel.busnot@arteris.com>; > *发送时间:* 2022年9月20日(星期二) 晚上7:40 > *收件人:* "gem5-users"<gem5-users@gem5.org>; > *主题:* [gem5-users] Re: 回复:Re: Different simulation results on different > computers with the same configuration > > Not sure to understand what you mean by “correct result". > > Can you please provide us with: > > 1. > > gem5 version (commit SHA) > 2. > > build command > 3. > > run command > > > Can you please also double-check that you are using supported versions of > python and gcc/clang. Also check that config.ini is the same in all cases. > > Gabriel > _______________________________________________ > gem5-users mailing list -- gem5-users@gem5.org > To unsubscribe send an email to gem5-users-leave@gem5.org >
2
2497597
Wed, Sep 21, 2022 1:42 AM

Thanks!!!!
I believe this is the reason, so I can't wait to test it out.

I rebuild the gem5 by using command "scons build/NULL/gem5.opt PROTOCOL=Garnet_standalone -j 4" and run the same script as before.
As expected, I got the correct output!

I'm so happy to have your help, I'll take this mistake to heart.

------------------ 原始邮件 ------------------
发件人:                                                                                                                        "The gem5 Users mailing list"                                                                                    <jason@lowepower.com>;
发送时间: 2022年9月20日(星期二) 晚上11:40
收件人: "The gem5 Users mailing list"<gem5-users@gem5.org>;
抄送: "Eliot&nbsp,Moss"<moss@cs.umass.edu>;
主题: [gem5-users] Re: 回复:Re: 回复:Re: Different simulation results on different computers with the same configuration

Hello,

The following two command produce very different gem5 binaries

scons build/Garnet_standalone/gem5.opt 

The above command will use the Garnet_standalone "coherence" protocol. I use quotes around coherence because that protocol is essentially an empty protocol with no coherence. It's meant for testing networks and nothing else.

scons build/NULL/gem5.opt

This command will use the MI_example protocol. This is a completely different coherence protocol than Garnet_standalone. See https://gem5.googlesource.com/public/gem5/+/refs/heads/stable/build_opts/NULL for how the protocol is defined in the default build options.

You can specify the exact protocol you want when you build gem5 by using the scons parameter --protocol 

Cheers,
Jason

On Tue, Sep 20, 2022 at 8:08 AM 2497597 <2497597@qq.com> wrote:

The correct result is below:

average_packet_latency = 15.510408
average_packet_latency = 15.528615
average_packet_latency = 15.682214
average_packet_latency = 15.695504
average_packet_latency = 15.769957
average_packet_latency = 15.821728
average_packet_latency = 15.912262
average_packet_latency = 16.051925
average_packet_latency = 16.167249
average_packet_latency = 16.319634
average_packet_latency = 16.479105
average_packet_latency = 16.725313
average_packet_latency = 17.055812
average_packet_latency = 17.588959
average_packet_latency = 18.500431
average_packet_latency = 21.669417
average_packet_latency = 103.241365
average_packet_latency = 273.002675
average_packet_latency = 430.695013
average_packet_latency = 596.634683
average_packet_latency = 732.220679
average_packet_latency = 854.214438
average_packet_latency = 969.032975
average_packet_latency = 1087.468352
average_packet_latency = 1207.588344

1.gem5 version is v21.2.1.0

2.build command is ”scons build/NULL/gem5.opt -j 4"

3.run command is below and the "--injection-rate" is increasing with step=0.02

Command to Run:
./build/NULL/gem5.opt
configs/example/garnet_synth_traffic.py \
--network=garnet2.0 \
--num-cpus=64 \
--num-dirs=64 \
--topology=Mesh_XY \
--mesh-rows=8 \
--sim-cycles=10000 \
--inj-vnet=0 \
--injectionrate=0.02 \
--synthetic=uniform_random

I find some interesting phenomenons.
I think I figured out where the problem might be.
It might be the permission error and the build order.
I built "X86","ARM","NULL" on my old computer before,so I try to restore it all.
Then I find the miss folder "pycache" come out,I think I'm going to get close to the root of the problem.
But I failed.
I rebuild a same version gem5 on my new computer with the command"”scons build/Garnet_standalone/gem5.opt -j 4" under the root state.
Then I run the same script,I get the correct output.(I can see the average_packet_latency is increasing with the --injectionrate)
Although I get the desired output, I can't figure out why I get the wrong output.

------------------ 原始邮件 ------------------
发件人:                                                                                                                        "The gem5 Users mailing list"                                                                                    <gabriel.busnot@arteris.com>;
发送时间: 2022年9月20日(星期二) 晚上7:40
收件人: "gem5-users"<gem5-users@gem5.org>;

主题: [gem5-users] Re: 回复:Re: Different simulation results on different computers with the same configuration

Not sure to understand what you mean by “correct result".

Can you please provide us with:

gem5 version (commit SHA)

build command

run command

Can you please also double-check that you are using supported versions of python and gcc/clang. Also check that config.ini is the same in all cases.

Gabriel


gem5-users mailing list -- gem5-users@gem5.org
To unsubscribe send an email to gem5-users-leave@gem5.org

Thanks!!!! I believe this is the reason, so I can't wait to test it out. I rebuild the gem5 by using command "scons build/NULL/gem5.opt&nbsp;PROTOCOL=Garnet_standalone&nbsp;-j 4" and run the same script as before. As expected, I got the correct output! I'm so happy to have your help, I'll take this mistake to heart. ------------------&nbsp;原始邮件&nbsp;------------------ 发件人: "The gem5 Users mailing list" <jason@lowepower.com&gt;; 发送时间:&nbsp;2022年9月20日(星期二) 晚上11:40 收件人:&nbsp;"The gem5 Users mailing list"<gem5-users@gem5.org&gt;; 抄送:&nbsp;"Eliot&amp;nbsp,Moss"<moss@cs.umass.edu&gt;; 主题:&nbsp;[gem5-users] Re: 回复:Re: 回复:Re: Different simulation results on different computers with the same configuration Hello, The following two command produce very different gem5 binaries scons build/Garnet_standalone/gem5.opt&nbsp; The above command will use the Garnet_standalone "coherence" protocol. I use quotes around coherence because that protocol is essentially an empty protocol with no coherence. It's meant for testing networks and nothing else. scons build/NULL/gem5.opt This command will use the MI_example protocol. This is a completely different coherence protocol than Garnet_standalone. See&nbsp;https://gem5.googlesource.com/public/gem5/+/refs/heads/stable/build_opts/NULL for how the protocol is defined in the default build options. You can specify&nbsp;the exact protocol&nbsp;you want when you build gem5 by using the scons parameter `--protocol`&nbsp; Cheers, Jason On Tue, Sep 20, 2022 at 8:08 AM 2497597 <2497597@qq.com&gt; wrote: The correct result is below: average_packet_latency = 15.510408 average_packet_latency = 15.528615 average_packet_latency = 15.682214 average_packet_latency = 15.695504 average_packet_latency = 15.769957 average_packet_latency = 15.821728 average_packet_latency = 15.912262 average_packet_latency = 16.051925 average_packet_latency = 16.167249 average_packet_latency = 16.319634 average_packet_latency = 16.479105 average_packet_latency = 16.725313 average_packet_latency = 17.055812 average_packet_latency = 17.588959 average_packet_latency = 18.500431 average_packet_latency = 21.669417 average_packet_latency = 103.241365 average_packet_latency = 273.002675 average_packet_latency = 430.695013 average_packet_latency = 596.634683 average_packet_latency = 732.220679 average_packet_latency = 854.214438 average_packet_latency = 969.032975 average_packet_latency = 1087.468352 average_packet_latency = 1207.588344 1.gem5 version is v21.2.1.0 2.build command is ”scons build/NULL/gem5.opt -j 4" 3.run command is below and the "--injection-rate" is increasing with step=0.02 Command to Run: ./build/NULL/gem5.opt configs/example/garnet_synth_traffic.py \ --network=garnet2.0 \ --num-cpus=64 \ --num-dirs=64 \ --topology=Mesh_XY \ --mesh-rows=8 \ --sim-cycles=10000 \ --inj-vnet=0 \ --injectionrate=0.02 \ --synthetic=uniform_random I find some interesting phenomenons. I think I figured out where the problem might be. It might be the permission error and the build order. I built "X86","ARM","NULL" on my old computer before,so I try to restore it all. Then I find the miss folder "__pycache__" come out,I think I'm going to get close to the root of the problem. But I failed. I rebuild a same version gem5 on my new computer with the command"”scons build/Garnet_standalone/gem5.opt -j 4" under the root state. Then I run the same script,I get the correct output.(I can see the average_packet_latency is increasing with the --injectionrate) Although I get the desired output, I can't figure out why I get the wrong output. ------------------ 原始邮件 ------------------ 发件人: "The gem5 Users mailing list" <gabriel.busnot@arteris.com&gt;; 发送时间:&nbsp;2022年9月20日(星期二) 晚上7:40 收件人:&nbsp;"gem5-users"<gem5-users@gem5.org&gt;; 主题:&nbsp;[gem5-users] Re: 回复:Re: Different simulation results on different computers with the same configuration Not sure to understand what you mean by “correct result". Can you please provide us with: gem5 version (commit SHA) build command run command Can you please also double-check that you are using supported versions of python and gcc/clang. Also check that config.ini is the same in all cases. Gabriel _______________________________________________ gem5-users mailing list -- gem5-users@gem5.org To unsubscribe send an email to gem5-users-leave@gem5.org