Exnode specification: Difference between revisions

From ReddNet
Jump to navigation Jump to search
No edit summary
 
(29 intermediate revisions by the same user not shown)
Line 2: Line 2:


[[Example LoRS exnode xml file]]
[[Example LoRS exnode xml file]]
* 1 namespace: exnode
* 6 elements: exnode, mapping, metadata, read, write, manage
* element metadata uses 2 attibutes: name, type
** 5 values used for attribute "name:" alloc_length, alloc_offset, e2e_blocksize, exnode_offset, logical_length
** 1 value used for attribute "type:" integer


[[Example L-Store exnode xml file]]
[[Example L-Store exnode xml file -- parity off]]
 
[[Example L-Store exnode xml file -- with parity blocks]]
*original implementation -- being modified
 
== Schema ==
 
The exnode xml is fairly unstructured and the functionality of a schema is minimal.
 
[[Simplified xml file]]
*(removed xmlns to view structure more clearly)
 
[[Simple schema]]
*In this example schema
**The schema shows that only the element "mapping" has required elements.
**The element metadata's use of attributes is unrestricted.
**The only re-used type is the IBP_capability (read,write,manage)
*extra-schema assumptions
**The exnode mapping order corresponds to the file block order
*L-Store additions
**element "function" (a tmp space holder to bypass a java bug)
**erasure metadata attribute "name" with a value of "parity slice size"
**element "parity" is the mapping for the parity block if parity slice size != -1
***the parity mapping immediately follows the data mapping


== Erasure codings ==
== Erasure codings ==
*L-Store writes extra erasure allocations for allocation recovery.
*L-Store writes data in its original format (possibly encrypted, etc), just like LoRS.
**erasure allocations are also stored in the exnode file
**These blocks can be read directly by LoRS.
**erasure allocations are only useful to the data-management area and not the data-client
*L-Store also writes extra erasure blocks to recover lost blocks.
**erasure allocations constitute an extension of the base (LoRS) exnode concept (or schema)
**the XOR-ing to generate erasure block is calculated on the depots
***nfu operation stored on the depot
**erasure blocks are also stored in the exnode file
**erasure blocks are only useful to the data-management area and not the data-client (see diagram)
**erasure blocks constitute an extension of the base (LoRS) exnode concept (or schema)
*L-Store can write a LoRS compatible exnode file that excludes the erasure allocations.
*L-Store can write a LoRS compatible exnode file that excludes the erasure allocations.
**LoRS compatible (the base schema)
**LoRS compatible (the base schema)
**LoRS does not have access to L-Store's capability to recover an allocation
**LoRS does not have access to L-Store's capability to recover a block
*The L-Store data-client (Lstcp) has no allocation recovery capability.
*The L-Store data-client (Lstcp) has no block recovery capability.
*data recovery, data warming, etc all data-management tasks, not data-client task.
*data recovery, data warming, etc all data-management tasks, not data-client task.


== Data client API & interfaces ==
== Data client API & interfaces ==
*command line tools for file management
*command line tools for file management
**list files, create directories, etc
**file movement
**file replication, augment, trim
*command line tools for namespace management
**list files
**create directories, etc
*LoRS API
*LoRS API
**uses the base exnode (LoRS)
**uses the base exnode schema (LoRS)
**reads local exnode files
**reads local exnode files
**talks to LoDN with a combined http:/LoDN protocol
**talks to LoDN with a combined http:/LoDN protocol
**C/C++ Posix and posix-like API
**C/C++ Posix and posix-like API
*Lstcp API (not developed yet)
*Lstcp API (not developed yet)
**does not use exnodes
**does not parse exnodes?
**talks to L-Server with the L-Store protocol
**talks to L-Server with the L-Store protocol
**receives IBP commands individually from L-Server?
***sends these commands directly to IBP depots?
*typical C/C++ API needs (proposed)
*typical C/C++ API needs (proposed)
**simple offsets and lengths
**simple offsets and lengths
**non-blocked (vector) reads
**non-blocked (vector) reads
== Schema ==
The exnode xml is fairly unstructured and the functionality of a schema is minimal.
[[Simplified xml file]]
*(removed xmlns to view structure more clearly)
[[Simple schema]]
*The schema shows that only the element mapping has much structure.
*The element metadata's use of attributes is completely open.
*The only re-used type is the IBP_capability (read,write,manage)

Latest revision as of 11:49, 29 January 2008

The exnode specification is implied by current usage in L-Store and LoRS.

Example LoRS exnode xml file

  • 1 namespace: exnode
  • 6 elements: exnode, mapping, metadata, read, write, manage
  • element metadata uses 2 attibutes: name, type
    • 5 values used for attribute "name:" alloc_length, alloc_offset, e2e_blocksize, exnode_offset, logical_length
    • 1 value used for attribute "type:" integer

Example L-Store exnode xml file -- parity off

Example L-Store exnode xml file -- with parity blocks

  • original implementation -- being modified

Schema

The exnode xml is fairly unstructured and the functionality of a schema is minimal.

Simplified xml file

  • (removed xmlns to view structure more clearly)

Simple schema

  • In this example schema
    • The schema shows that only the element "mapping" has required elements.
    • The element metadata's use of attributes is unrestricted.
    • The only re-used type is the IBP_capability (read,write,manage)
  • extra-schema assumptions
    • The exnode mapping order corresponds to the file block order
  • L-Store additions
    • element "function" (a tmp space holder to bypass a java bug)
    • erasure metadata attribute "name" with a value of "parity slice size"
    • element "parity" is the mapping for the parity block if parity slice size != -1
      • the parity mapping immediately follows the data mapping

Erasure codings

  • L-Store writes data in its original format (possibly encrypted, etc), just like LoRS.
    • These blocks can be read directly by LoRS.
  • L-Store also writes extra erasure blocks to recover lost blocks.
    • the XOR-ing to generate erasure block is calculated on the depots
      • nfu operation stored on the depot
    • erasure blocks are also stored in the exnode file
    • erasure blocks are only useful to the data-management area and not the data-client (see diagram)
    • erasure blocks constitute an extension of the base (LoRS) exnode concept (or schema)
  • L-Store can write a LoRS compatible exnode file that excludes the erasure allocations.
    • LoRS compatible (the base schema)
    • LoRS does not have access to L-Store's capability to recover a block
  • The L-Store data-client (Lstcp) has no block recovery capability.
  • data recovery, data warming, etc all data-management tasks, not data-client task.

Data client API & interfaces

  • command line tools for file management
    • file movement
    • file replication, augment, trim
  • command line tools for namespace management
    • list files
    • create directories, etc
  • LoRS API
    • uses the base exnode schema (LoRS)
    • reads local exnode files
    • talks to LoDN with a combined http:/LoDN protocol
    • C/C++ Posix and posix-like API
  • Lstcp API (not developed yet)
    • does not parse exnodes?
    • talks to L-Server with the L-Store protocol
    • receives IBP commands individually from L-Server?
      • sends these commands directly to IBP depots?
  • typical C/C++ API needs (proposed)
    • simple offsets and lengths
    • non-blocked (vector) reads