Here are some notes about when to acquire space, and how that pertains to covert channels.

Most algorithms acquire an amount of space that depends entirely on their input and not at all on the cost or availability of space. On the other hand there are often alternative algorithms that provide dramatic tradeoffs between space and time, or space and other valuables. Bloom filters trade off space against false positive rates.

Few algorithms do anything useful when they fail to get the space they ask for. (See text by David Goldberg here.) When an algorithm knows that it will need some particular amount of space it is usually wise to acquire or at least reserve that space at the beginning lest it spend time building intermediate values that are lost upon the rejection of a request for space late in the calculation. On the other hand asking for space too soon is an extra cost. Futures in space might be indicated. This is all painfully familiar to people who run factories.

There are categories of data such as web caches that are expendable. Such data can be reproduced at some cost when the value of the space that they occupy exceeds the expected cost of reproducing them. Such estimates can be automated.

The above disjointed comments bear on the problem of granting space in a confined realm. We must presume that there are those in the realm who wish to buy and sell space so as to affect the availability of space outside the realm and thus to smuggle out secrets that they are privy to. Such a signal path is called a covert channel.