We provide a virtualbox image where data center applications used for i-spy and ripple evaluation are pre-installed and set up to run. Please download the VMDK file from here. For instructions on how to import VMDK files, see this blog. We tested this virtualbox image with 8 processor cores and 16GB RAM. The username/password for the virtual machine are osboxes.org. Data center application traces used for ispy and ripple can be downloaded from here. These traces can be simulated using the Intel PT frontend of zsim.
Update: We also released traces of four more apps here
This guide assumes that you are using ubuntu Linux distribution.
Three HHVM applications (Drupal, Mediawiki, Wordpress): https://github.com/facebookarchive/oss-performance
Once you have installed HHVM (we use HHVM 5 built with gcc, the default ubuntu software manager one did not work for us, we had to follow this guide, https://docs.hhvm.com/hhvm/installation/building-from-source#ubuntu-15.04-vivid) and oss-performance suite along with their prerequisites, you can run the benchmarks using the following commands:
export HHVM=/path/to/hhvm
pushd .
cd /path/to/oss-performance
for i in drupal7 wordpress mediawiki;
do
$HHVM perf.php --$i --hhvm=$(echo $HHVM)
done
Alternatively, you can use the virtualbox image where HHVM workloads are pre-installed and set up to run. The oss-performance benchmarks are in a folder of the same name that you can cd into right from ~. You'll have to run the one line shell command from time_wait.md before you get started running benchmarks (if you don't it will tell you to do so). Also, the first run of a benchmark on a given startup will want to login to mysql to set things up- the credentials are root and root. So, to run the wordpress benchmark, for example, you would boot up the machine, cd into oss-performance, cat time_wait.md to figure out the one-liner shell command to run, then run hhvm perf.php --wordpress --hhvm=$(which hhvm).
Three DaCapo applications (Cassandra, Kafka, Tomcat): http://dacapobench.org
Once you have installed jvm (we used GraalVM, installation guide, https://www.graalvm.org/docs/getting-started-with-graalvm/linux/) and dacapo (We download download the file from https://sourceforge.net/projects/dacapobench/files/evaluation/dacapo-evaluation-git%2B309e1fa-java8.jar/download), you can run the benchmarks using the following commands:
for i in cassandra kafka tomcat;
do
java -jar /path/to/dacapo-evaluation-git+309e1fa-java8.jar $i -n 10
done
Two Renaissance applications (Finagle-Chirper and Finagle-HTTP): https://renaissance.dev Once you have installed jvm (again, we use GraalVM) and renaissance (we download the file from https://github.com/renaissance-benchmarks/renaissance/releases/download/v0.10.0/renaissance-gpl-0.10.0.jar), you can run the benchmarks using the following commands:
for i in chirper http;
do
java -jar /path/to/renaissance.jar finagle-$i -r 10
done
- Get the rocket-chip repo
git clone https://github.com/freechipsproject/rocket-chip.git
- Move to the same commit I was at:
cd rocket-chip
git checkout aca2f0c3b9fb81f42e4d1
- Instantiate the appropriate version of riscv-fesvr
git submodule update --init riscv-tools
cd riscv-tools
git submodule update --init riscv-tools riscv-fesvr
- Create a RISCV directory (to hold relevant software)
mkdir riscv
export RISCV=path to riscv
- Build & install riscv-fesvr
cd rocket-chip/riscv-tools/riscv-fesvr
mkdir build
cd build
../configure --prefix=$RISCV --target=riscv64-unknown-elf
make install
- Download and expand the tarball from Google Drive (https://drive.google.com/file/d/1WnrYhhYTGUMpndvoNpnzJOjgxdTOBHyL/view?usp=sharing) into rocket-chip (overwriting emulator)
Hopefully make will be able to reuse things, or at least choose to not invoke Chisel to regenerate the verilog files. I would try the binaries right away, and progressively try deleting more things (and rebuilding them for your system) until it works. The instructions below are for the single-core simulator because it is the fastest, but once you have a working methodology, change DefaultConfig to DefaultConfigN8 or DefaultConfigN16 to get more cores.
- Try executing the single-core simulator that is there:
cd emulator
./emulator-freechips.rocketchip.system-DefaultConfig +cycle-count ./dhrystone-head.riscv
- (If the above doesn't work), try to encourage make to re-link it all, and try #7 again
rm emulator-freechips.rocketchip.system-DefaultConfig
make CONFIG=DefaultConfig
- (If the above doesn't work), try rebuilding all .o files, and try #8 again
rm generated-src/freechips.rocketchip.system.DefaultConfig/*.o
- (If the above doesn't work), try rebuilding verilator (and re-running verilator) by executing #8 again
rm -rf verilator
rm -rf generated-src/freechips.rocketchip.system.DefaultConfig
When you start compiling large binaries, you can turn on parallel builds with -j:
make MAKEFLAGS="$MAKEFLAGS -j4" CONFIG=DefaultConfigN8
If you do need to run the Scala to generate new designs, make sure to install version 8 of the JVM. If when emitting a large design you run out of memory, you can ask for more:
make JVM_MEMORY=8G MAKEFLAGS="$MAKEFLAGS -j4" CONFIG=DefaultConfigN8
@inproceedings{khan2020ispy,
author = {Khan, Tanvir Ahmed and Sriraman, Akshitha and Devietti, Joseph and Pokam, Gilles and Litz, Heiner and Kasikci, Baris},
title = {I-SPY: Context-Driven Conditional Instruction Prefetching with Coalescing},
booktitle = {Proceedings of the 53rd IEEE/ACM International Symposium on Microarchitecture (MICRO)},
series = {MICRO 2020},
year = {2020},
publisher = {IEEE},
url = {https://doi.org/10.1109/MICRO50266.2020.00024},
doi = {10.1109/MICRO50266.2020.00024},
}
@inproceedings{khan2021ripple,
author = {Khan, Tanvir Ahmed and Zhang, Dexin and Sriraman, Akshitha and Devietti, Joseph and Pokam, Gilles and Litz, Heiner and Kasikci, Baris},
title = {Ripple: Profile-Guided Instruction Cache Replacement for Data Center Applications},
booktitle = {Proceedings of the 48th International Symposium on Computer Architecture (ISCA)},
series = {ISCA 2021},
year = {2021},
month = jun,
}