1 !(.3) "# $ - - PDF document

1

• !

"#

$ % $ ! $ !& $ #'(') $ *# $ ##+ $ %+#

%

$ , #

– Scalability

• Petabytes storage capacity, billions of file, file sizes (2**64), number
• f users

– Performance:

• Networking: separate data & control networks
• Disk, tape striping, Parallel IO

$

• –

Requirements for fast per file and aggregate data transfers – Requirements for large data stores – Requirements for large single logical name space – Requirements for distributed, secure access to data

%(.)

$ /0 – Initial proposal at 1992 – Initial release on 7/1/1996 – Collaborate effort from government-industry, institutions include Lawrence Livermore, Los Alamos, Sandia, Oak Ridge, NASA Langley Research Center, and IBM Government Systems, etc. $

• – Current version 6.2

– Available from IBM Global Services – Houston as a service

• ffering

– Used by national labs like SDSC, ORNL, LANL, LLNL for storing petabytes of data.

"#

$ % $ ! $ !& $ #'(') $ *# $ ##+ $ %+#

!(1223)

2

!(.3) "#

$ % $ ! $ !& $ #'(') $ *# $ ##+ $ %+#

!&(1223)

• #4#
• '#%

– Replacement of selected software components – Support Application Direct access to lower level services

• 4#
• #4#*
• *
• '%

!&(.3)

• 5+#4#

– Scalable data throughput – Scalable storage capacity and storage space management – Scalable robustness – Scalable name service – Scalable numbers of clients – Scalable deployment across geographical distances and multiple cooperating institutions – Scalable storage system management – Scalable security – Client roles in scalability

"#

$ % $ ! $ !& $ #'(') $ *# $ ##+ $ %+#

'#

$ +'

– Extend disk space – Back up disk files to tape – Managed permanent archive

$ !*### #

– Data Migrate down the hierarchy – Migrated files may be asynchronously purged from higher level (e.g. disk) to free up space

$ '#+##* #+

– Disk to Tape – Tape only – High Speed Disk to low cost disk to MAID to tape library to shelf tape

3

"#

$ % $ ! $ !& $ #'(') $ *# $ ##+ $ %+#

*#(1223)

4

*#(.3) "#

$ % $ ! $ !& $ #'(') $ *# $ ##+ $ %+#

##+(1)

$ 6#(/#)

– Logical sequence of bits that constitutes the data for a single file

$ #5#

– A unit of storage media on which HPSS stores data – For disk, this is an OS-level raw device (e.g., /dev/rhdisk5) – For tape, this is a cartridge (e.g., A00950)

$ 5#5#

– Logical Abstraction of physical volume used by Storage Server – Striped group of 1 or more Physical Volumes – Each Virtual Volume is assigned to a Storage Class

##+(.)

$

• –

An abstraction storage object which is mapped onto a virtual volume

$ #

– A set of characteristics and usages parameters to be associated with a particular grouping of HPSS virtual volumes

$ #*(")

– Defines a set of parameters associated with operational and performance characteristics of a bitfile

##+(7) "#

$ % $ ! $ !& $ #'(') $ *# $ ##+ $ %+#

, % , % , +

5

%+#(%1)

• +#

*

• *+

– Remote Procedure Calls (RPC) – Metadata System – Security Services – Concurrency Services (threads) – Transaction Management Service – Communication Services – Logging Services

%+#(%.)

$ %#+#

– Distributed Computing Environment provides RPC, security, thread, and time services – Encina/Structured File System (SFS) provide metadata management and transaction management services

$ 8*#*

– Encina/SFS is replaced by IBM DB2 based schema since release 5.1 – DCE is replaced by a set of industry standard RPC, security services and thread services since release 6.1.

%+#(%)

• %#%

– Client API following POSIX semantics – FTP (standard and parallel) – NFS – Parallel File System

• 8*#*

– None-DCE client APIs – Linux VFS

%+#(+1)

• +

– Mover

• '+

– Core Server: Name Service, Bitfile Server, Storage Service – Migration/Purge server – Storage System Management – Location Server – Logging service

• /+

– PVL – PVR

%+#(+.)

• *

– Name Server (NS) : Maps a human readable filename to an HPSS object identifier. – Bitfile server (BFS): Provides the abstraction of bit file to its clients – Storage server (SS): Provides a hierarchy of storage objects: storage segments,

virtual volumes and physical volumes

• '*

– Responsible for transfering data from a source device to a sink device

• #5# 4(5 )

– PVL manages all physical volumes

• #5#+(5)

– PVR manages all HPSS cartridges

• '(')

– SSM monitors and controls the available resources of the HPSS storage system.

6

9-