- Infos im HLRS Wiki sind nicht rechtsverbindlich und ohne Gewähr -
- Information contained in the HLRS Wiki is not legally binding and HLRS is not responsible for any damages that might result from its use -

HPSS User Access: Difference between revisions

From HLRS Platforms
Jump to navigationJump to search
No edit summary
No edit summary
(3 intermediate revisions by 2 users not shown)
Line 1: Line 1:
= General =


Please be aware that HPSS is a Hierarchical Storage Management System (HSM) with tape storage included; certain characteristics differ from a disk usage.
When transferring a file to the system via (p)ftp it is stored on the disk cache as first step. Later, the data is migrated to two copies on tape.
When retrieving a file, it may happen that the file is not on the disk cache anymore and has to be recalled from tape storage. This can take a few minutes.


Currently, there are two methods to access the HPSS complex:
= Best Practices =
 
* Do not store large numbers of files.
* It is not recommended to recursively store directory structures.
* Better tar small files to one larger archive. Tar can also be used to keep the directory structures.
* HPSS is not meant as a backup system.
* Pftp is preferred since it automatically choses the right Class of Service (CoS) and is much faster due to parallel streaming of data.
 
= Access Methods =


== Parallel FTP ==
== Parallel FTP ==


On LAKI, HERMIT and HORNET, there is a Parallel FTP client available.
On LAKI and HAZELHEN, there is a Parallel FTP client available.


The advantage of Parallel FTP is a file transfer which makes use of several parallel
The advantage of Parallel FTP is a file transfer which makes use of several parallel
network connections or at least several parallel I/O streams.
network connections or at least several parallel I/O streams.


On HERMIT and HORNET, you have to load the module "hpss" first:  
On LAKI and HAZELHEN, you have to load the module "hpss" first:  


{{Command| command=module load hpss}}
{{Command| command=module load hpss}}


To start the client, you should specify the number of parallel streams. We recommend a setting of 2.
To start the client, you should specify the number of parallel streams. We recommend a setting of 4.


{{Command| command=pftp_client -w2 hpsscore 4021}}
{{Command| command=pftp_client -w 4 hpsscore 4021}}


When requested, please provide your username and password.
When requested, please provide your username and password.
The password is your general hww password (as you have it
The password is your general hww password (as you have it
for HERMIT for example).
for HAZELHEN for example).


{{Note| text=For technical reasons a password change  
{{Note| text=For technical reasons a password change  
is valid on HPSS with a one day delay.}}
is valid on HPSS with a one day delay.}}


Compared to the ftp access, there is no setting for the class of service
Compared to the ftp access, there is no setting for the class of service necessary as this is done automatically by the pftp_client.
necessary as this is done automatically by the pftp_client.


Use the parallel ''pput'' and ''pget'' commands instead of the normal ''put'' and ''get''.
Use the parallel ''pput'' and ''pget'' commands instead of the normal ''put'' and ''get''.
Line 56: Line 67:
The following values are possible:
The following values are possible:


112: size < 2GB
102: size < 500MB


122: 2GB < size < 8GB
122: 0.5GB < size < 8GB


132: size > 8GB
132: size > 8GB
Line 66: Line 77:
ftp prompt:
ftp prompt:
{{Command| command=site setcos <cos id>}}
{{Command| command=site setcos <cos id>}}
Default setting for the cos id is 112. The setting of the cos id is valid for all
The setting of the cos id is valid for all
following file transfers.
following file transfers.


Line 77: Line 88:
there is currently no access to HPSS available from outside the hww
there is currently no access to HPSS available from outside the hww
network. For a transfer of stored files to your home site,
network. For a transfer of stored files to your home site,
please move these files to a system within hww where you have scp access.
please move these files to a system within hww where you have scp or GridFTP access.
This is an intermediate solution, we plan to provide a method for a
 
direct transfer within the next months.
= Further Information =
 
The official user guide by IBM can be found here: http://www.hpss-collaboration.org/documents/hpss752/Users_Guide.pdf

Revision as of 15:22, 21 November 2018

General

Please be aware that HPSS is a Hierarchical Storage Management System (HSM) with tape storage included; certain characteristics differ from a disk usage. When transferring a file to the system via (p)ftp it is stored on the disk cache as first step. Later, the data is migrated to two copies on tape. When retrieving a file, it may happen that the file is not on the disk cache anymore and has to be recalled from tape storage. This can take a few minutes.

Best Practices

  • Do not store large numbers of files.
  • It is not recommended to recursively store directory structures.
  • Better tar small files to one larger archive. Tar can also be used to keep the directory structures.
  • HPSS is not meant as a backup system.
  • Pftp is preferred since it automatically choses the right Class of Service (CoS) and is much faster due to parallel streaming of data.

Access Methods

Parallel FTP

On LAKI and HAZELHEN, there is a Parallel FTP client available.

The advantage of Parallel FTP is a file transfer which makes use of several parallel network connections or at least several parallel I/O streams.

On LAKI and HAZELHEN, you have to load the module "hpss" first:

module load hpss


To start the client, you should specify the number of parallel streams. We recommend a setting of 4.

pftp_client -w 4 hpsscore 4021


When requested, please provide your username and password. The password is your general hww password (as you have it for HAZELHEN for example).

Note: For technical reasons a password change is valid on HPSS with a one day delay.


Compared to the ftp access, there is no setting for the class of service necessary as this is done automatically by the pftp_client.

Use the parallel pput and pget commands instead of the normal put and get.

More details of using ftp and pftp to access HPSS can be found in the official User's guide.

FTP

Initiate the ftp connection to HPSS using

ftp hpsscore.hww.de 4021


When requested, please provide your username and password. The password is your general hww password (as you have it for Hermit for example).

ftp will check the .netrc file in the user's home directory, where e.g. login information can be given and the the ftp command can be used in e.g. scripts without a password prompt (see ftp man pages). Be aware of security issues, if a password is stored in plain text in a .netrc file.

Note: For technical reasons a password change is valid on HPSS with a one day delay.


Class of Service

When putting files into HPSS you have to specify a so called class of service. The class of service depends on the file size. The following values are possible:

102: size < 500MB

122: 0.5GB < size < 8GB

132: size > 8GB

You have to specify the class of service for a file prior to moving the file into HPSS. Before the transfer, you have to specify at the ftp prompt:

site setcos <cos id>

The setting of the cos id is valid for all following file transfers.

Beside the setting of the class of service, this option works just like normal ftp.


Other Access Methods

Currently, there are no other access methods available. Especially, there is currently no access to HPSS available from outside the hww network. For a transfer of stored files to your home site, please move these files to a system within hww where you have scp or GridFTP access.

Further Information

The official user guide by IBM can be found here: http://www.hpss-collaboration.org/documents/hpss752/Users_Guide.pdf