These are a set of larger cloud-like workloads to measure the impact of malloc speedup. We use Xapian, an open-source search engine, and Masstree, a multicore key-value store. The versions provided here were used to produce all results in the Mallacc paper; the latest versions can be found at xapian.org and https://github.com/kohler/masstree-beta.
For information about the paper, please refer to:
Svilen Kanev, Sam (Likun) Xi, Gu-Yeon Wei, and David Brooks.
Mallacc: Accelerator Memory Allocation.
Architectural Support for Operating Systems and Programming Languages, April 2017.
PDF
Our system has toolchains installed in non-standard locations, so our build flow
relies on writing do_configure.sh scripts that execute configure with hard-coded
toolchain locations and other default options. All installable packages come
with such a script. You will need to edit each script to change the toolchain
locations to match your target system, or simply remove all those options if your
toolchains are located in the default directories (e.g. /usr/bin).
NOTE: Before running do_configure.sh, we also recommend you run
autoreconf -i from the top level directory of the source package to generate
the configure script for your system and autotools installation.
Then, execute the do_configure.sh script, then run make and make install
from within the build directory to build and install the package.
Other than the toolchain installation directories, do not remove any of the other CFLAGS/CXXFLAGS specified, like flags about the ABI, sized deletes, etc. These are important.
Detailed instructions can be found in Xapian's INSTALL. Here are some general instructions.
We install Xapian (and all its additional dependencies) to
/home/user/xapian/install. You can change this to suit your needs, but
remember that the entire Mallacc build flows assumes this installation
directory, so other do_configure.sh scripts will need to be changed.
Xapian's primary dependencies are zlib and libuuid. We used libuuid version
1.0.3, provided here. Use the do_configure.sh
script to install libuuid.
When building Xapian's core, you need to select whether to link it against a version of libtcmalloc with or without the Mallacc special instructions, which determines which directory the configure script will look for the tcmalloc installation. See do_configure.sh.
If you need to build Omega, you'll also need libmagic and libpcre. These libraries are also provided under deps. You must first install the Xapian core library and these dependencies before building Omega. Note that we did not use Omega itself as a malloc benchmark, so there is no option to link Omega with a special version of tcmalloc.
We provide three datasets for Xapian benchmarking, and each dataset has its own client driver. These are located under xapian/benchmarks.
- tiny_index: A sample program that runs queries over a very small search index.
- wiki_abstracts: Run queries over an index of Wikipedia page abstracts.
- wiki_pages: Run queries over an index of full Wikipedia page text.
To build them, just go to the appropriate folder and run make. By default, they
are linked with the baseline TCMalloc library (no magic instructions). To link
against the special build, instead run make MALLACC=1. This assumes that the
Mallacc enabled TCMalloc library has been built and installed to the right
place.
The Wikipedia datasets were built from Wikipedia dumps, downloaded in July 2016 from here. They were processed and indexed by Omega. The Xapian databases are availble for download from here:
Download them and extract the files to a location. To make building the search index more tractable and reduce individual file sizes, we divided the full wikipedia page index into eight shards. We use a ''stub database'' to dynamically aggregate them together when we execute queries. The abstracts database was small enough to not require sharding, but we use a stub database anyways for consistency. Therefore, you will need to update the stub database file to point to the place your databases are downloaded and extracted.
Go to the corresponding client driver (query_wiki_abstracts or
query_wiki_pages), open up the file called stub_database.db. Update the
paths to the appropriate search index. Each line should look like:
auto /path/to/database/default
Detailed instructions can be found in Masstree's README.
As was the case with Xapian, we provide a do_configure.sh script, which you
will need to edit to change toolchain paths.
In general, the steps are as follows:
cd masstree-beta
./bootstrap.sh
# Edit do_configure.sh.
./do_configure.sh
makeThis section will only discuss running workloads natively, not in simulation.
Go to the directory for the dataset you want to run (e.g. wiki_pages). Each dataset comes with at least one text file of queries to run. Select the one you want and execute the following command:
make run
The client driver will run the complete set of queries 5 times by default (this can be changed in common/run_benchmark.h). The driver will record how long each iteration takes. When all iterations are finished, it will print the recorded times to stdout, followed by the results of the queries. No printing will occur until all queries are finished.
We ran two of the built-in Masstree performance tests, wcol11 and same.
./mttest -j1 --no-notebook wcol1
./mttest -j1 --no-notebook -l 10000000 same