Przeglądaj źródła

Add README

master
haavee 5 miesięcy temu
rodzic
commit
fdbe15924c
1 zmienionych plików z 217 dodań i 0 usunięć
  1. +217
    -0
      README.md

+ 217
- 0
README.md Wyświetl plik

@@ -0,0 +1,217 @@
# VBS\_FS

The vbsfs package consists of three utilities for dealing with scattered
recorded VLBI data on FlexBuff, and since svn version r45 of July 2016, Mark6:

|Tool|Use for|
|:--- | :--- |
|**vbs_ls**| lists all(some) vbs recordings on the system|
|**vbs_rm**| remove (a) recording(s) |
|**vbs_fs**| a FUSE virtual file system, presents each scattered recording as a single file, for use with standard UNIX tools|

As of version r45 (27 July 2016) all utilities have been completely
rewritten for use on either FlexBuff or Mark6 hardware and deal
transparently with FlexBuff (`vbs`) and/or Mark6 recordings (`dplane`) even
when mixed recording formats are present on the disks.

Installation instructions can be found later on in this document.


## Basic operation (r45 and later)

All three tools scan a set of disks for known recordings. The command line
arguments have been synchronized such that all tools share the disk
selection mechanism and some helpful options:

--help print detailed help for the utility
--version print the actual version of the utility
-6 add Mark6 mountpoints to search path [/mnt/disks/*/*]
-v add FlexBuff mountpoints to search path [/mnt/disk*]
-R <path> add pattern <path> to search path
-I <pattern> index recording(s) matching <pattern> (r53 and later)

If the search path is not set from the command line, the FlexBuff pattern will be used.

Also new since r45 is that in `vbs_ls` and `vbs_fs` the owner, group and
permissions metadata reflect the actual on disk owner, group and permissions.


## `vbs_ls` and `vbs_rm`


vbs_ls and vbs_rm try to mimic their famous UNIX equivalents `ls(1)` and
`rm(1)`. As such `vbs_ls` and `vbs_rm` support many of the more pedestrian options
that `ls(1)` and `rm(1)` offer.

As an example:

$> vbs_ls -lrth

produces output equivalent to

$> ls -lrth


`vbs_ls` has an added `-T` option which yields ***totals per pattern given on
the command line***. Use this to e.g. find out the recorded size per
experiment:

$> vbs_ls -lT gm082_ys_* n16l1_o8_*

or per station:

$> vbs_ls -lT *_ys_* *_o8_*

Note that on some shells it is necessary to escape the asterisk or else
the shell will expand the file names (or set the "noglob" option in your
shell if it supports that)

Use
```bash
$> vbs_{fs,ls,rm} --help
```
for detailed information on command line options.

## Output format of `vbs_ls`
Because `vbs_ls` now supports both `dplane` and `vbs` recordings, the output
has been modified to reflect this. In the example below are shown some test
recordings made on one of JIVE's FlexBuffs in both `dplane` and `vbs` format:

verkout@flexbuf1:~$ vbs_ls -l haavee*
Found 4 recordings in 90 chunks, 57.68G
-rw-r--r-- jops flexbuf 16797931200 Jun 22 10:27 haavee_m6_no0001
-rw-r--r-- jops flexbuf 20981685600 Jun 22 10:28 haavee_m6_no0002
drw-r--r-- jops flexbuf 13689999360 Jun 22 10:27 haavee_vbs_no0001
drw-r--r-- jops flexbuf 10468823040 Jun 22 10:27 haavee_vbs_no0001a

Mark6/`dplane` recordings get type `-` (file) and EVN FlexBuff `vbs`
recordings are listed as `d`, for directories. On disk that is also the
distinction.

A file type of `+` means that the named recording was found in both vbs and
dplane format. This could happen if a recording by the exact same name was
made on two different disk packs and then mounted at the same time.

Or in this (synthetic) case - where `vbs_ls` is asked to aggregate (`-T`) all
recordings matching the pattern `haavee*` - which fits both the `vbs` and
`dplane` recordings shown previously:

verkout@flexbuf1:~$ vbs_ls -lTh haavee*
Found 3 recordings in 4503 chunks, 1.08T
+rw-r--r-- jops flexbuf 55.43G Jun 22 10:28 haavee* 90.00% recovered [81/90]


# vbs_fs

`vbs_fs` is the FUSE file system. It will gather in memory the scattered
recordings based on block sequence numbers. Basic usage:

$> mkdir /path/to/mountpoint
$> vbs_fs [options] /path/to/mountpoint

From this point on all recognized recordings found in the search path are presented as
regular files in `/path/to/mountpoint`. Since r45 `vbs_fs` reassembles Mark6
`dplane` recordings as well.


### Short use case

This is a description of how to run and, after use, unmount/stop the
`vbs_fs` file system.

Running `vbs_fs`:

# Create a mountpoint for the FUSE file system
$> mkdir /path/to/mountpoint

$> vbs_fs [-f] [-6] [-o allow_other] /path/to/mountpoint

Where:
-f run in foreground, do not daemonize (can ^C it)
-6 monitor Mark6 mountpoints for vbs recordings
(Only available since 0.4.1)
-o allow_other other users may access the mounted files too

Run `vbs_fs -h` to show all options. Since r45 use `vbs_fs --help`
to show all options and also refer to 'common command line options'
above.

`vbs_fs` r45 and later gained extra command line options, affecting startup
time and allow the user to balance memory usage with performance:

```bash
--quick perform a quick scan of the disks. Not all metadata
(size, access time) will be correct; a full scan is
expensive, which is the backward compatible default.

This option is mostly useful if it is known beforehand
which recording is to be opened (for e.g. correlation)
and a full scan of the disks is a waste of everyones
time.

The recording sizes will be set to 0 and a full scan of a
recording is done at the moment a recording is first
opened. This means that before a recording can be read
from, it must be opened at least once to trigger the
scan.

The use of touch(1) is useful for that.

-n <int> The readahead amount. `vbs_fs` is optimized for sequential
access and does readahead of `<n>` blocks. Use this option
to either limit the readahead (0 is a valid value and
also the default) or e.g. to the number of disks - this
keeps all disks spinning all the time.

Setting the read ahead amount to number of disks it
should be possible to reach read-back speeds of 1GB/s or
more; subject to number of disks, CPUs and memory
available - YMMV.
```

Unmounting of the mounted `vbs_fs` file system depends on how `vbs_fs` was
run:

```
* if started with the `-f` flag ('run in foreground') you can just `^C`
the program and it will shut down correctly and unmount

* otherwise it was daemonized - `vbs_fs` runs in the background. In such
case the FUSE file system has to be unmounted. Use:

$> fusermount -u /path/to/mountpoint

to unmount the file system. The `vbs_fs` daemon program will exit.
```


# Installation and compilation


- Make sure libfuse-dev, make and g++ are installed, for the FUSE header file(s) and libraries.

- Clone this repository and enter the directory
```bash
$> git clone https://code.jive.eu/verkout/vbs_fs.git
$> cd vbs_fs

# Optionally select any branch (if any)
$> git checkout <branch>
```

- Configure and build:
```bash
# 'make install' will install `vbs_fs`, `vbs_ls` and `vbs_rm`
# in PREFIX/bin. Use `./configure --help` to see defaults.
$> ./configure [--prefix=PREFIX]

# build and optionally install, if you want to do that right now
$> make [install]
```

- After configuring, the Makefile (created by the configure step) has two other useful make targets:

|Target|Use for|
|:--- | :--- |
|**clean**| cleans the object files and binary/ies locally, does not remove the installed binary/ies|
|**uninstall**| This target is only available since 0.4.2 and - as the name suggests - does an uninstall of the installed binary/ies and configuration file(s)|

Ładowanie…
Anuluj
Zapisz