Semester Graphs

Fall 2009 (9/18/09 - 12/18/09, excluding weekends and student holidays) data is presented below, broken down by data point, or as a PDF here.

Login Breakdown

Because logins are provided on a per-weekly basis, this data set runs from 9/2/09 to 12/19/09 and does not exclude weekends or holidays.

This shows the breakdown of total Athena logins by machine type.

(click for larger version)

Occupancy Rate

A good metric of cluster use is the occupancy rate for any given cluster. Unfortunately, the average daily occupancy rate for any given cluster is skewed by the fact that the occupancy for most clusters is nearly 0 between the hours of midnight and 9:00am. If the occupancy rate is presented as a range between average daily high occupancy and average daily low occupancy, we get a much better representation of cluster usage.

The occupancy rate graph for Fall 2009 is presented here:

As you can see, many of the clusters exceed 80% occupancy during the day. Some clusters, however, are underutilized and perhaps some workstations in those clusters could be replaced with small group workspaces.

Instances of Full Capacity

Another good metric of cluster use is the number of days in which a cluster has at least one occurrence (5 minutes) of full capacity (that is, all the workstations are occupied). This tell us the relative popularity of a cluster.

The instances of peak capacity for Fall 2009 is presented here:

As you can see, 12-182 and 56-129 remain highly popular, as does E51-075, as it is the only cluster east of Ames St. Another interesting trend is that all the library clusters (Hayden, Rotch, Barker) experienced days with peak capacity, perhaps indicating the value of these clusters as quiet workspaces.

Data Sources

Login Reports

Starting in early 2005, we began collecting data on the number of logins Athena-wide. We capture both total logins and unique logins on a weekly basis. Unique logins are users who logged in at least once in the one-week period.

Accuracy: Information is collected by the lert server, which the default dotfiles contact at login time. It is possible for users to bypass this check, and we have no data on the number of users who do that, so it's possible these numbers are underreported.

Starting in 2006, logins were broken down between amongst cluster, quickstation, dialup, and "other" (private workstations). However, between January 2007 and March 2009, the "dialup" count was included in "other". See the raw data for details.

Data is reported in the sysd_stats meeting on a weekly basis. Every year, data will be collected in a spreadsheet for convenience and posted here. The Academic Year runs from Sept - August. Note that some academic years have 52 weeks and some have 53.

Cluster Usage Data

Starting in 1992, usage data was collected Athena-wide. The larvnet server snapshots cview data on an hourly basis throughout the day, and then averages those snapshots to create the average number of machines free per day. The minimum and maximum total number of machines are also recorded. (The total number of machines can decrease if a machine breaks or is taken out of service).

Starting in December 1998, usage data was broken down by cluster.

NOTE: Due to various outages or migrations of the larvnet and syslog servers over the years, some portions of the raw data had duplicate entries – that is, entries for the same day and cluster. These duplicate entries were pruned. In some cases, it was clear which entry to prune, either because the entries were exact duplicates of each other, or because the values were clearly erroneous (for example, showing 0 total machines in a cluster). In other cases, a more subjective method was used. Regardless, each spreadsheet lists the pruned data so it can be reincorporated if necessary.

Each Excel file has two sheets in it – one containing the actual data, and another listing the duplicate entries that were pruned from the raw data.