Internet Backplane Protocol API and Applications
Download
Report
Transcript Internet Backplane Protocol API and Applications
Mobile Management of
Network Files
Alex Bassi
Micah Beck
Terry Moore
Computer Science Department
University of Tennessee
1
Logistical Networking
• Technology for shared network storage that
can scale in terms of
– the size of the user community,
– the aggregate quantity of storage that can be
allocated, and
– the breadth of distribution of service nodes
across network borders
• Parallels layered IP networking stack
2
Bottom-up Design Philosophy
• The lowest globally accessible network
layer in the network storage stack should
– Enable scalable Internet-style resource sharing
– Expose underlying storage resources in order to
maximize freedom at higher levels
• Standard network storage systems fails in
some degree to satisfy one or both criteria.
3
Internet Backplane Protocol
(IBP)
• Servers that make
allocation of primitive
“byte arrays” available
to clients
• Byte arrays are not
blocks (more abstract)
– Network capabilities
(primitive security)
– Variable extents
• Byte arrays are not
files (weaker
semantics)
– Size & duration are
limited
– “Volatile” allocations
– Best effort reliability
and availability
– No directory structure,
accounting
– No caching, replication
4
The Logistical Networking Project
at the University of Tennessee
• Primary UT Investigators
• UT Graduate Students
– Micah Beck
– Wo Ling
– James S. Plank
– Yong
• UT Research Staff
– Xiang
– Alex Bassi
– Erica Fuentes
– Terry Moore
– Anthony Burton
• Funding
– DOE Next Generation Internet & SciDAC
– NSF Next Generation Software
Logistical Computation and Internetworking (LoCI) project w/ Birman,
Dongarra, Wolski
5
Collaborating Projects
• Network Weather
Service (UCSB)
– Rich Wolski
– Martin Swany
• NetSolve (UT ICL)
– Jack Dongarra
– Michelle Miller
• L-Bone & LoFS (UT)
– James S. Plank
• Application Level
Scheduling (UCSD)
– Fran Birman
– Henri Casanova
– Graziano Obertelli
• Tamanoir
(Université Claude
Bernard Lyon 1 )
– Laurant Lefavre
6
Building on IBP
• Many applications assume file semantics
– Unbounded size & duration
– High reliability & availability
– Caching & replication
• In a layered architecture, these are
implemented through aggregation and
additional intelligence at the next level
7
Enter the exNode
• In the Unix file system, blocks are
aggregated into files using the intermediate
node data structure, or inode
• We implement file semantics on top of IBP
using the external node (exNode) data
structure
– Aggregation of IBP byte array allocations
– Additional metadata to enable management
8
ExNode vs inode
IBP Allocations
the network
local system
capabilities
exNode
inode
user
kernel
block addresses
disk blocks
9
The exNode is a Set
• A set of mappings and
associated metadata
• A mapping associates
a data extent and a
temporal extent with a
(set of) storage
resource(s).
• Minimal container API
– constructor &
destructor
– add mapping
– enumerate
• Lookup functions
– get byte extent
– get event list
10
The exNode is Mobile Code!
(files are structured processes)
• Mappings are location-independent
• XML serialization enables interoperability
• The state of file system processes can be
captured in an abstract, portable form
• Mobile files can be freed of directory,
operating system, accounting boundaries
11
Example: IBP Mail
SMTP
sender
exNode
receiver
IBP
write
IBP
read
IBP copy
12
File System Services
•
•
•
•
Indirection to preserve integrity
Temporal and data extent aggregation
Reliability through RAID, replication
Performance through locality
13
Active File Management
• Active Services
Time
– Active Probing to
Maintain Fault
Tolerance
– Lease Renewal
– Defragmentation
– Asynchronous Transfer
Management
allocate on
depot 1
renew on
depot 1
text
fragment to
depot 2 & 3
renew on
depot 3
Defragmentation
14
Conclusions
• The Internet Backplane Protocol provides a
fabric for network state management
• Scalable and ubiquitous sharing are enabled
by weak semantics & minimal accounting
• The exNode & serialization provide a
mobile abstraction of storage aggregation
• Active file services can span OS & FS
boundaries (Log Structured Logistical FS)
15