Skip to content

Latest commit

 

History

History

experiments

Folders and files

NameName
Last commit message
Last commit date

parent directory

..
base-scripts
base-scripts
 
 
failover
failover
 
 
herd
herd
 
 
kvstore
kvstore
 
 
stress
stress
 
 
README.md
README.md
 
 

README.md

uKharon's experiments

This directory explains how to automatically run uKharon's experiments

Configuring the scripts

The file base-scripts/config.sh needs to be configured before running the experiments. Several variables need to be set as explained below: Note that the deployment scripts need not run from a deployment machine. You can instead run them e.g., from your laptop.

ROOT_DIR

It defines the root directory where the scripts deploy the precompiled binaries in the deployment machines. The default value is the home directory of the user used to access the deployment machines.

machine1
machine2
machine3
machine4
machine5
machine6
machine7
machine8

These variables define the symbolic names to access the deployment machines over ssh. To setup these symbolic names you need to edit ~/.ssh/config in the machine you run the deployment scripts from (e.g., your laptop). You can learn how to do this here.

It is important to be able to access the deployment machines without password. Thus, you need to install ssh keys. You can learn how to do this here. If your key contains a passphrase, you can rely on ssh-agent, also explained in the above previous link.

For example, the following entry in ~/.ssh/config

Host delta1
  HostName lpdquatro1.epfl.ch
  User user
  Port 22
  IdentityFile ~/.ssh/id_rsa
  ServerAliveInterval 120

Allows you to access the machine by merely typing ssh delta1.

machine1hostname
machine2hostname
machine3hostname
machine4hostname
machine5hostname
machine6hostname
machine7hostname
machine8hostname

These variables define the Fully Qualified Domain Names (FQDN) of these machines. Every machine should be able to access every other machine using the corresponding FQDN. You can learn how to setup the FQDNs here.

REGISTRY_MACHINE

It defines which machine is going to run the memcached server needed to exchange QP information when setting up RDMA connections. Make sure that memcached is installed in the declared machine and that its port 11211 is open.

UKHARON_MCGROUP
UKHARON_KERNELMCGROUP

Set these variables with the information retrieved when building ukharon-build.

UKHARON_SYNCKILLERMCGROUP

Set this variable to another InfiniBand multicast group. It is used by the sync_killer to trigger SIGKILLs remotely and thus simulate coordinated failures.

UKHARON_HAVE_SUDO_ACCESS
UKHARON_SUDO_ASKS_PASS
UKHARON_SUDO_PASS

Sudo access is necessary to achieve optimal performance. These variables refer to the deployment machines. If you have sudo access in the deployement machines, set the first variable to true. If when issuing a command with sudo you need to type your password, set the second variable to true and put the password in the third variable.

UKHARON_CPUNODEBIND
UKHARON_CPUMEMBIND

Set the variables, which refer to the deployment machines, to achieve optimal performance. In a multi-socket machine, set UKHARON_CPUNODEBIND to the socket that is closer to the RDMA NIC and UKHARON_CPUMEMBIND to the memory that is closer to this socket. The instructions in mu-build explain how to retrieve this information.

UKHARON_MEMSTRESS_CORES

These are the cores (in the deployment machines) that are used by stress-ng for memory stressing. These cores must not interfer with the cores selected in ukharon-build.

Running the experiments

  • To run the experiment of figure 3, go to the stress directory
  • To run the experiment of figure 4, go to the herd directory
  • To run the experiment of table 2, go to the failover directory
  • To run the experiment of figure 5, go to the kvstore directory