User`s guide

D2.1.4 IST-033576

1 Introduction

The current prototype of Linux-XOS consists of two major components: the ker-

nel checkpoint/restart mechanisms and the node-level VO-support mechanisms.

Checkpointing in Linux-XOS is conceptually seperated in three layers: grid

checkpointer, system checkpointer and kernel checkpointer. This document mainly

describes the lower layer, the kernel checkpointer, as functionality for the sys-

tem and grid checkpointer. System and grid checkpointer are being implemented

in the framework deﬁned by the Application Execution Management work of

WP3.3. The current document presents guidelines to install and use the kernel

checkpointer functionalites. It also present the BLCR checkpoint/restart API.

The node-level VO support component described in section 3 consists of the

following modules: a PAM extension (a dynamic library pam_xos.so), a NSS

extension (a dynamic library libnss_xos.so), an Account Mapping Service

(AMS, a daemon program xos_amsd), and several utilities for testing. The

PAM extension is used to dynamically map VO user credentials into local user

accounts. The NSS extension is used for extracting user/group mapping informa-

tion via standard naming lookup APIs. AMS plays a role of managing runtime

mappings which serves as the backend for PAM/NSS extensions. Based on these

components, the widely used login tool, OpenSSH, is modiﬁed to allow VO users

to interactively access a remote XtreemOS node. Section 3 details a step-by-step

procedure of installation, conﬁguration and testing of all relevant components on

a fresh Linux environment.

2 Kernel Checkpointer

The kernel checkpoint/restart component consists of three kernel modules, an

API/library and three binaries. The library is used to implement user callbacks for

the checkpointing (as described in [2]). The binaries are the following: cr_run,

cr_checkpoint and cr_restart. They are respectively used to preload libcr and

libpthread if the application has not been linked against BLCR library, checkpoint

a tree of processes, and restart a previously checkpointed tree of processes. The

kernel modules implement the checkpointing logic: blcr.ko implements coordina-

tion between callbacks and the checkpoint of each process/thread, blcr_vmadump.ko

is used to do the underlying dump, and blcr_imports.ko is used to export unex-

ported kernel symbols.

5/49 XtreemOS–Integrated Project