In the MDI system dataset names were constructed as a 5-part name consisting of a project or program name, a reduction level name, a series name, a series index number, and a version number. The series name contained user readable indicators of the blocking of time into datasets and standardized series numbers were referenced to a common epoch (1 January 1993, 0 UT). Thus the dataset name for hour 0 of full disk velocity images for July 14 1996 is:
prog:mdi,level:lev1.5,series:fd_V_01h[30960]After requesting the location of that dataset by a call to a service (e.g. the "peq" program) the user could learn that the dataset is now in the directory "/PDS20/D6362810". If that dataset is not used for a few weeks its online copy will be deleted so the next time it is again requested it will be staged to disk and will appear in another directory. But the user never needs to see the actual storage working directory since the user refers to the data by its descriptive name. The velocity image for say the 10th minute of that hour will be in a file (e.g. 0009.fits) in that working directory. Again the actual file name is not seen by the user since she has used an SSSC provided API function to open the file for the requested minute number.
The JSOC naming system is derived from the MDI system but has some key differences. The concept of a multipart name is eliminated to reflect the MDI actual experience. A new top level name will identify the JSOC data server as a whole to allow a common format for access to e.g. the existing MDI data. The existing name parts simply become user viewable parts of the dataseries name. Thus the same dataset in the example above might be:
jsoc:/mdi_fd_V_lev1.5/[t_rec>=1996.07.14_00,t_rec<1996.07.14_01]
or
jsoc:/mdi_fd_V_01h_lev1.5/[sn=30960]
where the syntax here is still TBR.
When the user asks for this dataset in the JSOC system a function provided by the JSOC catalog API will generate a list of all the datarecords in the dataset. That list will include the working directory (storage unit) and file name and optionally slice within a file for each datarecord. The user will not normally need to see this information since she will simply use the "open_record" JSOC API function call.
[I don't formally define ``dataset'' anywhere, it is perhaps just a datarecord. A ``virtual dataset'' can be simply a record from a series where the records only contain a set of links and no data segments.]