Welcome

This is the source code for LogECMem, which is tested on Ubuntu 16.04.

Preparation

These are the required libraries that users need to download separately.

• gcc (v5.18.7), g++ (v5.4.0)
• make (v4.1), cmake (v3.5.1), autogen (v5.18.7), autoconf (v2.69), automake (v1.14),
• yasm (v1.3.0), nasm (v2.11)
• libtool (v2.4.6)
• boost libraries (libboost-all-dev) (v1.58)
• libevent (libevent-dev) (v2.0.21) $ sudo apt-get install gcc g++ make cmake autogen autoconf automake yasm nasm libtool libboost-all-dev libevent-dev

Users can install the following library manually: Intel®-storage-acceleration-library (ISA-l) (v2.14.0).

$ tar -zxvf isa-l-2.14.0.tar.gz
$ cd isa-l-2.14.0
$ sh autogen.sh
$ ./configure; make; sudo make install

LogECMem Installation

Memcached Servers (v1.4.25)

Users can use apt-get to install Memcached instances in servers.

$ sudo apt-get install memcached

For standalone setup, users can use bash cls.sh to re-set memcached instances with IPs and Ports; For distributed setup, users can use multiple nodes with different IPs to run memcached instances.

LogECMem Proxy

Users use source code to install libmemcached (extented from v1.0.18) in the proxy.

$ cd libmemcached-1.0.18
$ sh configure; make; sudo make install
$ export LD_LIBRARY_PATH=/usr/local/lib:$LD_LIBRARY_PATH  #(if it ocuurs library path issue)

Users can setup passwardless SSH login, and use SSH root@node bash cls.sh to re-set memcached instances in the proxy.

User use g++ to compile update.cpp and repair.cpp.

$ bash make.sh

Workloads

Users can use the provided workloads (ycsb_set.txt and ycsb_test.txt) in each directory to do demo tests, and further more workloads via YCSB with basic parameter in https://github.com/brianfrankcooper/YCSB/wiki/Running-a-Workload.

Benchmarks

1.Update latency and Memory overhead Users can configure N, K, workload dir, and server IP parameters, and gain all three In-palace, Full-stripe and LogECMem results. "./update [1|2|3] dir N K IP > /dev/null", 1|2|3 indicates In-place|Full-stripe|LogECMem respectively, N indicates the number of all data and parity chunks, K indicates the number of all data chunks, dir indicates the path of workloads and IP indicates the DRAM node's IP. Note that users can configure more IPs and Ports in update.cpp and run.sh for distributed setup.

$ cd update
$ bash run.sh

2.Multiple chunks failures performance Users can configure N, K, workload dir and server IP parameters, and gain all schemes' repair performance. "./repair dir N K > /dev/null", N indicates the number of all data and parity chunks, K indicates the number of all data chunks, dir indicates the path of workloads and IP indicates the DRAM node's IP. Note that users can configure more in repair.cpp and run.sh for distributed setup.

$ cd repair
$ bash run.sh