Spectral-line data sets can be very large, and you may find, even with a Gbyte all to yourself, that you are short of disk space. There are a few things you can do to alleviate this problem.
Firstly, you should seriously consider averaging the data in time. For many purposes, the 15 s samples that the ATCA usually offers by default (although you may have set this to something longer when you observed) are much too short; see the discussion in § 6 to see how to do this with UVAVG and what penalty, if any, you pay. Secondly, you can also save space by compressing each complex visibility into half as many bits. See the discussion of the adverb douvcomp in ATLOD\ in § 4 for what you sacrifice in doing this. This compression is available in many AIPS tasks, and there is a specific task, UVCMP, to compress or uncompress data. Thirdly, you can discard channels that are of no further use to you. For example, when you run SPLIT, you might decide to drop channels at the beginning and end of the spectrum which are particularly noisy (because of the multiplication by the bandpass correction). However, if you are going to do visibility based continuum subtraction (see § 13.3.3), then you should not do this, as you need to retain as many line-free channels as possible.
Naturally, you would want to do these things as early in the processing chain as is practicable. For example, you might run UVAVG straight away after ATLOD, or, if there are large amounts of corrupt data, after you have edited the data and run SPLIT.