from sandbox to playground: dynamic virtual environments in the grid
DESCRIPTION
From Sandbox to Playground: Dynamic Virtual Environments in the Grid. Kate Keahey [email protected] Argonne National Laboratory Karl Doering University of California, Riverside Ian Foster Argonne National Laboratory. Realizing the Grid Vision. Quality of Service - PowerPoint PPT PresentationTRANSCRIPT
From Sandbox to Playground:
Dynamic Virtual Environments in the Grid
Kate [email protected]
Argonne National LaboratoryKarl Doering
University of California, Riverside Ian Foster
Argonne National Laboratory
Grid 2004 Kate Keahey
Realizing the Grid Vision
Quality of Service Protocol, agreement,
advance reservation The ability to enforce
what was agreed on
Quality of Life Being able to find the
right configuration on the Grid
Grid 2004 Kate Keahey
Quality of Service
Some form of control over remote nodes Enforcement of multiple qualities
CPU, disk, memory, network traffic… More than per-process enforcement
Process group: a master process starts other processes
Dynamically modifiable to reflect changing policies and state in the Grid
Not just quality of service Quality of Protection, etc… QoX
Grid 2004 Kate Keahey
Quality of Life
The right node configuration is hard to find Operating system and architectural differences
Different Linux distributions 64 bit vs 32 bit
Library signature and versioning The ability to customize a remote execution
environment Effortless configuration of remote nodes
Subject to policies Quality of Life for multiple groups of Grid users
Avoiding maintenance nightmare, etc.
Grid 2004 Kate Keahey
We Need a Sandbox A configurable execution environment, container
Virtualizes Grid Node Configuration Sandbox = Dynamic Virtual Environment (DVE)
We need to be able to create and manage it Quota, termination, etc.
requirements available technology
solutions
How can DVEs be implemented? Relevance to our needs, quality of solution, etc.
Grid 2004 Kate Keahey
DVE: Interfaces Implemented as Grid Services
OGSI, WSRF Factory
Creates and configures a DVE in implementation-specific way
e.g., dynamic account, deploys a VM Writes/configures access and management policy
E.g., modify the GT3 gridmapfile
DVE Service Interface providing DVE management
E.g., explicit or soft-state termination (implies policy updates) Access policy management
Allows for inspecting and modifying DVE properties E.g., hardware properties such as quota or software configuration
Grid 2004 Kate Keahey
DVE Implementations: Requirements What is a “container”? General
Not require users to e.g., use a specific language Non-invasive
Proof-carrying code, etc. Strong protection environment
Otherwise users won’t trust sites and sites won’t trust users Isolate users from each other
Fine-grain enforcement Configurable architecture, software, environment
Configurable environment throughout the software stack Application software/libraries/licenses
Potentially: execution state Allow migration
Grid 2004 Kate Keahey
DVEs and the Globus Toolkit
C
lien
t
(1) DN
(4) GSH
local DVEimplementationsetuid
(3)
gridmapfile
(5)
GRAM(6) Request+GSH
(2)
DVESservice PEP
DVEFactory Service PEP
Grid 2004 Kate Keahey
DVE Implementations Unix accounts
Pros: efficient, ubiquitous Cons: very limited enforcement Enforcement properties can be improved if used in
conjunction with other technologies setrlimit, DSRT, chroot, chown, and others
Sandboxes VServer: protection, sharing and fine-grain
enforcement Pros: efficient, fine-grain enforcement, typically very
lightweight Cons: limited state enforcement, configuration flexibility Adjustments needed to fully leverage fine-grain
enforcement
Grid 2004 Kate Keahey
DVE Implementations (cntd) Virtual Machines
VMware (not evaluated, but very promising: Xen) Pros:
Flexibility (run linux on linux, 32 on 64-bit, etc.) Enhanced security, audit forensics, etc. Great user state management Freezing/migration Customized environment A promising distribution/deployment tool
Cons: Potential for being less efficient (emulation) Potential for resource overhead Poor implementation of sharing, relatively little enforcement (but can
be combined with other technologies for enforcement) Maturity issues
The potential is excellent, but needs more work
Grid 2004 Kate Keahey
The Need for Speed
0
0.2
0.4
0.6
0.8
1
1.2
110100jt 110105jt 110109jt
UNIX acctVserverVMware
Comparison using the Fusion EFIT application
Grid 2004 Kate Keahey
Other efficiency concerns
Startup time
Resource usage overhead Memory use: VMware: 24MB + 1 MB per 32 MB
memory allocated Disk use: large for VMware
Table 1: DVE create/destroy times
Linux VServer VMware
Create 100 ms 360 ms 14-52 sec
Destroy 70 ms 200 ms 3-38 sec
Grid 2004 Kate Keahey
Enforcement Capabilities
Unix account VServer VMware
CPU usage (sec) Via setrlimit() Not at present, but could be added
Not enforced
CPU usage (%) Not enforced Limited: no VServer can starve another
Not in VMware Workstation
Disk space usage Dynamically(per-user
quotas)
Dynamically (per context quotas)
Statically (virtual disks)
Memory usage No Not at present, but could be added
Statically
Network usage No Dynamically Dynamically
Grid 2004 Kate Keahey
DVE Comparison Dynamic Accounts
Adduser versus pooled accounts A limited but one that is here to stay… at least for now
VServer Interesting: sharing and efficiency
VMware No sharing Least efficient Migration, flexibility, etc.
General criteria Efficiency: very acceptable, also see Xen Enforcement: uneven, needs more research Virtual Machines lead as far as configurability and user state representation Sharing
Potential for replication One VM per machine model?
Grid 2004 Kate Keahey
Implementation Status
Prototype available (GT 3.2) Karl Doering:
http://www-unix.mcs.anl.gov/~keahey/DS/DynamicSessions.htm
GT4 Implementation adduser versus account pools Better policy handling
Virtual machines and other implementations Work in progress SC04 poster:
P05: “Quality of Life in the Grids: VMs Meet Bioinformatics Applications”, with T. Freeman and D. Galron
Grid 2004 Kate Keahey
From Sandbox to Workspace
Virtual Workspaces VWs are represented by an ontology description
Virtual resource characteristics, software stack, etc. Potentially integrating community policy They can be copied, etc.
They can be implemented using different technologies
They can be customized by the user Deployed, managed and terminated in
implementation-specific way Entails some changes to the architecture
Grid 2004 Kate Keahey
Virtual Workspaces in the Grids
Clie
nt
request
VW EPR
inspect and manage
deploy & suspend
use existing VW Create VW
VW Factory
VW Repository
VW Manager
create new VW
ResourceVW
start program
Grid 2004 Kate Keahey
From Sandbox to Playground How will this affect interactions in the Grid?
Other than add many new capabilities A larger role for the virtual organization
Account screening process: resource owner -> virtual organization Should a VO be a legal entity? Needs new privileges if takes on more responsibility
Administration of VWs VW repository and other services, potentially VW certification
Sharing between VWs More policies
Changes to many Grid services May depend on the implementation we use Security, networking, potentially others
Top-down model for building a Grid Define a Grid in terms of requirements
Grid 2004 Kate Keahey
Conclusions For Grids to scale we need a way to create and manage
remote environments in the dynamically and effortlessly Implementations will vary
Virtual is the new Real! VMs present a very compelling solution…
Efficiency, flexibility, migration, etc. …and introduce some new challenges
New services, different models of sharing, security, etc.
A growing role for Virtual Organizations Policy, Policy, Policy…
Policy of resource owners, VOs, users… Using WS-Agreement to negotiate virtual workspaces? Have we exchanged one problem for another?
www.mcs.anl.gov/~keahey