CSC Getting Started Guide

Storing, Uploading, and Downloading Files

The CSC computing environment has been configured to make using all of our resources as easy as possible. All of our computer systems share two large storage volumes called /home and /quicksand. That is, every machine – Hemisphere, Occam, Toaster, and all of the compute nodes – access the same storage space for /home and /quicksand, so you can move easily between machines without transferring files. Every user has a directory under both /home and /quicksand, but it’s important to place your files in the correct location to achieve the best I/O performance. In general, important data such as software and irreplaceable results should be stored on /home, and all compute jobs should be run on the higher performance storage provided by /quicksand.

Saving Files on /home and /quicksand

User home directories are located in /home. The /home partition is provided by a single RAID array directly attached to a server, and all of the other systems access /home using NFS. Thus, the /home partition provides reliable but low-performance storage. In addition to /home, all users have a directory under /quicksand to use as working storage for running jobs. The /quicksand partition is provided by a series of high-performance parallel filesystem servers. Quicksand is intended for working storage only: it provides a large quantity of temporary storage while jobs are in progress.

Volume Name Capacity Default Quota
/home 1.2 TB 2 GB
/quicksand 3.2 TB 1000 GB

It is very important to select the correct filesystem for your data. The /home partition is intended as stable and reliable long term storage, and we backup the contents of /home to tape occasionally. To keep the backup size manageable, home directories have a quota in the 2 – 10 GB range. The /home filesystem, however, is not intended to support the massive amount of I/O generated by jobs running on clusters. All compute jobs should be run out of /quicksand instead.

The /quicksand partition is provided by a Lustre cluster named /sandstorm. While Lustre provides reliable storage, the parallel filesystem technology requires that multiple servers remain running for the filesystem to operate properly. Thus, the high-capacity and high-performance storage provided by /quicksand can be slightly more volatile than /home. Important data should be moved to /home for safekeeping. The /quicksand partition enforces a 1000 GB quota for all users to prevent a single user from monopolizing all of the space.

Do not run your I/O intensive jobs out of /home! Run them out of /quicksand instead. Jobs running out of /home that severely impact the performance of the NFS servers may be deleted by the administration staff without notification.

Backups

While we make every effort to store user data in as reliable fashion as possible, and even occasionally backup data for disaster recovery purposes, it is important to note that CSC systems are considered research machines and we are not responsible for lost data. Do not store your only copy of irreplaceable data on CSC systems. If your data is very important, you should store a backup copy somewhere else.

All of the files on /home are backed up to a spare disk array twice per week. These files are stored in a secure partition accessible by the owner and the system administrators. If you delete an important file, we may be able to help recover a previous version from the backups depending on timing. Every 10 days, the backup files are archived to tape and stored for disaster recovery.

The files on /quicksand are considered volatile and are not backed up. In addition, we reserve the right to remove old files from /quicksand using an automated scrubbing process. Please use /quicksand for your job output, but you may want to move important results to /home or another computer system.

Transferring Files

To transfer files to or from the CSC filesystems, use the secure copy program ‘scp’ and connect to hemisphere.cs.colorado.edu. This server has high-bandwidth connections to the Internet as well as the internal systems providing the storage, so transfers initiated with fileserver have the best possible performance. While all CSC systems such as the Hemisphere and Occam head nodes have access to all filesystems, file transfers initiated with front-end nodes may perform much more slowly than transfers initiated with the designated fileserver system.

‹ prev | 1 2 3 4 5 6 7 8 | next ›