Difference between revisions of "Diskruimte"

From Cncz
Jump to navigation Jump to search
(nieuwe prijs EC8+3 en uitleg)
Line 108: Line 108:
 
| align="left" | ''[prijs][price] per TB per [jaar][year] [zonder][without] backup''
 
| align="left" | ''[prijs][price] per TB per [jaar][year] [zonder][without] backup''
 
|-
 
|-
| align="left" | Erasure coding 8+3
+
| align="left" | Erasure coding 8+3 (*)
 
| align="left" | [goedkoop][cheap]
 
| align="left" | [goedkoop][cheap]
| align="left" | € 45
+
| align="left" | € 50 (was 45)
 
|-
 
|-
| align="left" | Erasure coding 5+7
+
| align="left" | Erasure coding 5+4
 +
| align="left" | [goedkoop + extra redundantie][cheap + additional redundancy]
 +
| align="left" | € 60
 +
|-
 +
| align="left" | Erasure coding 5+7 (** obsolete, use EC5+4)
 
| align="left" | [goedkoop + extra redundantie][cheap + additional redundancy]
 
| align="left" | [goedkoop + extra redundantie][cheap + additional redundancy]
 
| align="left" | € 80
 
| align="left" | € 80
Line 120: Line 124:
 
| align="left" | € 100
 
| align="left" | € 100
 
|-
 
|-
| align="left" | 4 copy
+
| align="left" | 4 copy (**)
 
| align="left" | [snellere r+w + extra redundantie][faster r+w + additional redundancy]
 
| align="left" | [snellere r+w + extra redundantie][faster r+w + additional redundancy]
 
| align="left" | € 135
 
| align="left" | € 135
 
|}
 
|}
 +
 +
[nl]
 +
'''* EC8+3 prijs''' is van 45 naar 50 euro gegaan per 1 januari 2022, dit is vanwege het gebruik, waarbij blijkt dat er toch erg veel kleine files opgeslagen zijn, wat in Ceph tot flinke overhead zorgt.
 +
[/nl]
 +
[en]
 +
'''* EC8+3 price''' is raised from 45 to 50 euro as of January 1st 2022, due to the way the storage is used with large amounts of very small files. This creates a significant overhead in Ceph.
 +
[/en]
 +
 +
[nl]
 +
'''** EC57 en 4copy'''. Vanaf het moment dat we een derde locatie in gebruik namen was het mogelijk om een nieuwe goedkopere pool (EC5+4) te maken die ook beschikbaar bleef bij het uitvallen van een heel datacenter. Hetzelfde geldt voor de 4copy pool, want de 3copy pool blijft nu ook beschikbaar bij het uitvallen van een heel datacenter. De 4copy pool heeft wel het voordeel van een extra kopie, maar dat maakt niet uit bij het uitvallen van een heel datacenter.
 +
[/nl]
 +
[en]
 +
'''** EC57 and 4copy''' When we added another location (in Forum datacenter) it was possible to create a new cheaper pool (EC5+4) that is still available when a whole datacenter breaks. The same holds for the 4 copy pool, because now the 3copy pool will remain active when one datacenter breaks. Of course, the 4copy pool will have the additional benefit of an extra copy, but this has no advantage when a whole datacenter breaks.
 +
[/en]
 +
  
 
[nl]
 
[nl]

Revision as of 13:36, 17 December 2021

Diskspace

The diskspace on C&CZ servers can be used from all kind of C&CZ servers and personal computers, but also from other PCs or even from home with WinSCP or VPN. Almost all disks that are managed by C&CZ, are being backed up regularly, in order to be able to restore data in case of small or large calamities.

Home directories

Every user with a Science login has or is entitled to an amount of disc space of a few Gigabytes on a server. This disc space is called the "home-directory" on Unix/Linux computers and the "H- or U-drive" on Windows-computers. The location of this homedirectory (which server) can be viewed on the Do-It-Yourself website.

Naming

Username guest204
SMB (Windows/...) name \\home1.science.ru.nl\guest204

or
\\home2.science.ru.nl\guest204
check on DIY

URL (Apple/Linux/Android/...) name smb://home1.science.ru.nl/guest204

or
smb://home2.science.ru.nl/guest204
check on DIY

NFS (C&CZ beheerd Linux) name /home/guest204

Access rights

Long ago the (Unix) home directory of a user, except for a few protected areas, was readable for all users of the server. Today the home directory of a user can be accessed by the user himself. The user can change the access rights. C&CZ checks for home directories that are writable by other users.

Access through NFS

Mounting a home (U:) drive on Linux via NFS/Kerberos.

Functionality and costs of network shares

RAID server shares

Diskspace for groups/institutions/projects: there are a few fileservers with RAID storage with partitions that can be rented for a period of 3 years. The price for new discs or a new 3 year extension of an older disc is per July 2018 for FNWI departments:

size incl. backup without backup
ca. 200 GB € 40 per year € 10 per year
ca. 400 GB € 80 per year € 20 per year

Although even the cheapest version is much more expensive than buying 1 disk for 1 PC, it often makes sense, because of the reliability (redundant disks, backup, support contract) and security (stable server). One or more folders on such a partition can be mapped as a network drive on Windows PCs or NFS-mounted on Unix/Linux hosts. The ability to read and/or write files on these folders can be limited to a group of logins. That group can be managed by the department on the Do-It-Yourself website.
C&CZ has service contracts for these servers and has spares on site, so a failure can be resolved quite fast. Because the disks are part of a RAID set, the failure of 1 single disk or even 2 disks, will not give a disruption of service for users. The partitions are backed up (daily and incremental). Even in the case when the whole server room is lost, data can (eventually) be restored.

Ceph Storage

Starting November 2019 we can provide almost unlimited storage for the Faculty of Science using our Ceph storage cluster. The way Ceph works there is a tradeoff for performance and redundancy. Also it is possible to improve redundancy above single server RAID-6 level, with the additional redundancy options. NB Ceph volumes have no backups, the volumes tend to be too large to backup.

Ceph Erasure coding has a high overhead for smaller files, the prices mentioned below are based on the optimal storage overhead, which can be approximated when files stored are at least 4 megabytes or larger.

Pool why price per TB per year without backup
Erasure coding 8+3 (*) cheap € 50 (was 45)
Erasure coding 5+4 cheap + additional redundancy € 60
Erasure coding 5+7 (** obsolete, use EC5+4) cheap + additional redundancy € 80
3 copy faster r+w € 100
4 copy (**) faster r+w + additional redundancy € 135

* EC8+3 price is raised from 45 to 50 euro as of January 1st 2022, due to the way the storage is used with large amounts of very small files. This creates a significant overhead in Ceph.

** EC57 and 4copy When we added another location (in Forum datacenter) it was possible to create a new cheaper pool (EC5+4) that is still available when a whole datacenter breaks. The same holds for the 4 copy pool, because now the 3copy pool will remain active when one datacenter breaks. Of course, the 4copy pool will have the additional benefit of an extra copy, but this has no advantage when a whole datacenter breaks.


The Ceph storage can be used as Windows/Samba share, NFS share or S3 object store. Object store differs fundamentally from a normal filesystem, so data stored in a Windows or NFS share cannot be accessed using the S3 protocol.

The performance properties of Ceph are different from traditional single server storage; write speed usually exceeds read speed and lots of small files is killing for throughput, even worse than on traditional storage.

NB, 1 TB is 1.000.000.000.000 bytes

Naming

Volume name sharename
SMB (Windows/..) name \\sharename-srv.science.ru.nl\sharename
URL (Apple/Linux/Android/...) name smb://sharename-srv.science.ru.nl/sharename
NFS (C&CZ-beheerd Linux) name /vol/sharename

Access rights

Most of the shared disks can be read and written by a specific group of users. The owners of this group can administer on the Do-It-Yourself website which accounts are a member of this group.

Requests

A request for one or more network discs should contain:

  • requested name of the disc(s)
  • requested size (max ca. 500GB with backup)
  • possibly requested backup schedules to lower the price (Daily/Monthly/Yearly)
  • Science loginname of an owner
  • possibly Science loginname of a member
  • charge account (kostenplaats) or project code for the costs in the first three years.

Temporary shared diskspace

Every now and then you want to send one or more large files (more than a few tens of MBs) to someone else within the Faculty, mail is unsuited for those large files. To make this easy, one can use a network share, where one can store large files temporarily in order to have someone else copy the files from this location. Note that this is explicitly meant for temporary storage, we do not make backups of this share, every day we remove files older than 21 days old. When copying files to this share, make sure the file timestamps are updated. Some copy programs (like rsync) maintain the original timestamps and older files will be deleted. To update timestamps, you can use the following command:

find . -exec touch {} +

This share can also be used to store temporary files only readable for yourself by using a different name for the share. Note that also in this case, old files will be removed.

Please create a subdirectory with your name first, and put your files in that directory.

For files totaling less than 250GB, also Surfdrive is an alternative. For sending files up to 500GB SURFfilesender can be used.

Naming

Volume name temp
SMB (Windows/..) name \\temp-srv.science.ru.nl\share

or
\\temp-srv.science.ru.nl\onlyme

URL (Apple/Linux/Android/...) name smb://temp-srv.science.ru.nl/share

or
smb://temp-srv.science.ru.nl/onlyme

NFS (C&CZ-beheerd Linux) name /vol/temp

Access rights

  • Readable by all users: share
  • Only readable for the owner: onlyme

Lijn.png