MemLiner is a managed runtime built for a memory-disaggregated cluster where each managed application runson one server and uses both local memory and remote memory located on another server. When launched on MemLiner,the process fetches data from the remote server via the paging system. MemLiner reduces the local-memory working set and improves the remote-memory prefetching by lining upthe memory accesses from application and GC threads. MemLiner is transparent to applications and can be integrated in any existing GC algorithms, such as G1 and Shenandoah. Please refer to our OSDI'22 paper, MemLiner, for more details.
Two sets of environment have been tested for MemLiner:
Ubuntu 18.04
Linux 5.4
OpenJDK 12.04
GCC 5.5
MLNX_OFED driver 4.9-2.2.4.0
or
CentOS 7.5 - 7.7
Linux 5.4
OpenJDK 12.04
GCC 5.5
MLNX_OFED driver 4.9-2.2.4.0 Among the requirements, the Linux version, OpenJDK version and MLNX-OFED driver version are guaranteed during the build & installation process below. Just make sure that the Linux distribution version and gcc version are correct.
Next we will use Ubuntu 18.04 as an example to show how to build and install the kernel. Please change the kernel on both CPU and memory server.
（1）Change the grub parameters
sudo vim /etc/default/grub
# Choose the bootup kernel version as 5.4.0
GRUB_DEFAULT="Advanced options for Ubuntu>Ubuntu, with Linux 5.4.0"
# Change the value of GRUB_CMDLINE_LINUX:
GRUB_CMDLINE_LINUX="nokaslr transparent_hugepage=madvise intel_pstate=disable idle=poll processor.max_cstate=0 intel_idle.max_cstate=0"
（2）Build the Kernel source code && install it. In case new kernel options are prompted during sudo ./build_kernel.sh build, press enter to use the default options.
# Change to the kernel folder:
cd MemLiner/Kernel
# In case new kernel options are prompted, press enter to use the default options.
sudo ./build_kernel.sh build
sudo ./build_kernel.sh install
sudo rebootPreparations:
MemLiner is only tested on MLNX_OFED-4.9-2.2.4.0. Download and unzip the package according to your system version, on both CPU and memory server.
Take Ubuntu 18.04 as an example:
# Download the MLNX OFED driver for the Ubuntu 18.04
wget https://content.mellanox.com/ofed/MLNX_OFED-4.9-2.2.4.0/MLNX_OFED_LINUX-4.9-2.2.4.0-ubuntu18.04-x86_64.tgz
tar xzf MLNX_OFED_LINUX-4.9-2.2.4.0-ubuntu18.04-x86_64.tgz
cd MLNX_OFED_LINUX-4.9-2.2.4.0-ubuntu18.04-x86_64
# Remove the incompatible libraries
sudo apt remove ibverbs-providers:amd64 librdmacm1:amd64 librdmacm-dev:amd64 libibverbs-dev:amd64 libopensm5a libosmvendor4 libosmcomp3 -y
# Install the MLNX OFED driver against the kernel 5.4.0
sudo ./mlnxofedinstall --add-kernel-support(1) Enable and start the openibd service
sudo systemctl enable openibd
sudo systemctl start openibd
# confirm the service is running and enabled:
sudo systemctl status openibd
# the log shown as:
● openibd.service - openibd - configure Mellanox devices
Loaded: loaded (/lib/systemd/system/openibd.service; enabled; vendor preset: enabled)
Active: active (exited) since Mon 2022-05-02 14:40:53 CST; 1min 24s ago
(2) Enable and start the opensmd service:
sudo systemctl enable opensmd
sudo systemctl start opensmd
# confirm the service status
sudo systemctl status opensmd
# the log shown as:
opensmd.service - LSB: Manage OpenSM
Loaded: loaded (/etc/init.d/opensmd; generated)
Active: active (running) since Mon 2022-05-02 14:53:39 CST; 10s ago
#
# Warning: you may encounter the problem:
#
opensmd.service is not a native service, redirecting to systemd-sysv-install.
Executing: /lib/systemd/systemd-sysv-install enable opensmd
update-rc.d: error: no runlevel symlinks to modify, aborting!
#
# Please refer to the **Question #2** in FAQ for how to solve this problem
#Check the InfiniBand information
# Get the InfiniBand information
ibstat
# the log shown as:
# Adapter's stat should be Active.
Port 1:
State: Active
Physical state: LinkUp
Rate: 100
Base lid: 3
LMC: 0
SM lid: 3
Capability mask: 0x2651e84a
Port GUID: 0x0c42a10300605e88
Link layer: InfiniBandInstall the necessary libraries
sudo yum install -y gtk2 atk cairo tcl tkDownload the MLNX OFED driver for CentOS 7.7
wget https://content.mellanox.com/ofed/MLNX_OFED-4.9-2.2.4.0/MLNX_OFED_LINUX-4.9-2.2.4.0-rhel7.7-x86_64.tgzAnd then, repeat the steps from 3.3.1 to 3.3.3.
The user needs to build the RemoteSwap on both the CPU server and memory servers.
(1) IP configuration
Assign memory server’s ip address to both the CPU server and memory servers. Take guest@zion-1.cs.ucla.edu(CPU server) and guest@zion-4.cs.ucla.edu(Memory server) as an example. Memory server’s InfiniBand IP address is 10.0.0.4: (InfiniBand IP of zion-12 is 10.0.0.12. IPs of IB on other servers can be printed with ifconfig ib0 | grep inet)
// (1) CPU server
// Replace the ${HOME}/Memliner/scripts/client/rswap_rdma.c:783: char ip[] = "memory.server.ib.ip";
// to:
char ip[] = "10.0.0.4";
// (2) Memory server
// Replace the ${HOME}/Memliner/scripts/server/rswap_server.cpp:61: const char *ip_str = "memory.server.ib.ip";
// to:
const char *ip_str = "10.0.0.4";(2) Available cores configuration for RemoteSwap server (memory server).
Replace the macro, ONLINE_CORES, defined in MemLiner/scripts/server/rswap_server.hpp to the number of cores of the CPU server (which can be printed by command line, nproc , on the CPU server.)
// ${HOME}/MemLiner/scripts/server/rswap_server.hpp:38:
#define ONLINE_CORES 16(1) Build the client end on the CPU server, e.g., guest@zion-1.cs.ucla.edu
cd ${HOME}/MemLiner/scripts/client
make clean && make(2) Build the server end on the memory server, e.g., guest@zion-4.cs.ucla.edu
cd ${HOME}/MemLiner/scripts/server
make clean && makeAnd then, please refer to Section 1.1 for how to connect the CPU server and the memory server.
Build MemLiner JDK. Please download and install jdk-12.0.2 and other dependent libraries to build the MemLiner (OpenJDK)
cd ${HOME}/MemLiner/JDK
./configure --with-boot-jdk=$HOME/jdk-12.0.2 --with-debug-level=release
make JOBS=32Warning: The CPU server and the memory server only need to be connected once, and the connection will persist until reboot or memory server process is killed. Before trying to connect, check whether they are already connected. See Question#1 in FAQ for instructions. If it is connected, directly go to section 1.2.
Launch the memory server:
# Log into memory server, e.g., guest@zion-4.cs.ucla.edu or guest@zion-12.cs.ucla.edu
# Let memory server run in background
tmux
# Launch the memory server and wait the connection from CPU serever
cd ${HOME}/MemLiner/scripts/server
./rswap-serverLaunch the CPU server:
# Log into the CPU server, e.g., guest@zion-1.cs.ucla.edu or guest@zion-3.cs.ucla.edu
cd ${HOME}/MemLiner/scripts/client
./manage_rswap_client.sh install
# Confirm the success of the RDMA connection between the CPU server and the memory server.
# Print the kernel log by:
dmesg | grep "frontswap module loaded"
# the output should be:
[Some timestamp] frontswap module loadedMemLiner relies on cgroup to limit the memory usage of applications running on the CPU server. we are using cgroup version 1 here.
Create cgroup on the CPU server:
# Create a memory cgroup, memctl, for the applications running on the CPU server.
# $USER is the username of the account. It's "guest" here.
sudo cgcreate -t $USER -a $USER -g memory:/memctl
# Please confirm that a memory cgroup directory is built as:
/sys/fs/cgroup/memory/memctlLimit local memory size on the CPU server. E.g., set local memory to 9GB:
# Please adjust the local memory size for different applications.
# e.g., for Spark with a 32GB Java heap. 25% local memory ratio means that we limit its CPU server memory size to 9GB.
# Application's working set includeds both Java heap and off-heap data, which is around 36GB in total.
echo 9g > /sys/fs/cgroup/memory/memctl/memory.limit_in_bytesStep 1: ensure that connection is already established.
# On CPU server
# Check RDMA connection between the CPU server and the memory server.
# search for connection built log by:
dmesg | grep "frontswap module loaded"
# If the output is:
[Some timestamp] frontswap module loaded
# Then the connection has been established.If there is no connection, connect according to instructions on section 1.1.
Step 2: ensure that connection is still working.
Warning: Press ctrl+b, then press d to leave a tmux session. Don’t kill the process running.
# On memory server
# try attaching all tmux sessions. E.g., attach to the most recent session
tmux a
# If a command is still running and there's output like below, the connection is still active.
[...]
handle_cqe, 2-sided RDMA message sent done ?
handle_cqe, REQUEST_CHUNKS, Send available Regions to CPU server.
send_regions , Send registered Java heap to CPU server, 12 chunks.
send_message, message size = 248
handle_cqe, 2-sided RDMA message sent done ?If the connection is not active, reboot both servers. After the reboot, connect according to instructions on section 1.1.
opensmd.service is not a native service, redirecting to systemd-sysv-install.
Executing: /lib/systemd/systemd-sysv-install enable opensmd
update-rc.d: error: no runlevel symlinks to modify, aborting!The original /etc/init.d/opensmd
8 ### BEGIN INIT INFO
9 # Provides: opensm
10 # Required-Start: $syslog openibd
11 # Required-Stop: $syslog openibd
12 # Default-Start: null
13 # Default-Stop: 0 1 6
14 # Description: Manage OpenSM
15 ### END INIT INFOChange the content in the line Default-start /etc/init.d/opensmd to :
12 # Default-Start: 2 3 4 5# Enable and strart the opensmd service
sudo update-rc.d opensmd remove -f
sudo systemctl enable opensmd
sudo systemctl start opensmd
# confirm the service status
sudo systemctl status opensmd
# The log shown as:
opensmd.service - LSB: Manage OpenSM
Loaded: loaded (/etc/init.d/opensmd; generated)
Active: active (running) since Mon 2022-05-02 14:53:39 CST; 10s ago