Allocation unit (cluster) is a file system measurement unit. When you format your HD in Windows, the default cluster is 4,096 bytes (NTFS). One cluster can have data from one file only.
If you for instance store, save a file that is 40,960 bytes big, it's stored in optimal circumstances in 10 clusters next to each other. Normally HD's fragment a bit, the file is stored in clusters not next to each other. Logically, a file that is stored to clusters next to each other is faster to access, so you have to defragment your HD every now and then. Defragmenting tries to move clusters from one file to next to each other. On the other hand, a very small cluster size means computer has more clusters to seek and search, increasing access time, so access time is not only depending on how fragmented the HD is.
As I mentioned, one cluster can only have data from one file. This means that cluster is reserved even if there is only a byte or two in it. Let's say you save a file that is 5,000 bytes big. It needs two clusters (NTSF default), one to fill it with first 4,096 bytes of that file and the second to put last 904 bytes. Both clusters are now reserved, and you can not save anything else in to these clusters. This is why if you check properties of a file or a folder, you can see two different values, size
and size on disk
This example file for instance is 40,059 bytes, filling 9 clusters full and one partially, so it needs 10 clusters * 4,096 bytes = 40,960 bytes disk space.
In bigger files and / or folders you can regain that "lost space" by compressing files, so the used space can in fact be smaller than the actual size. Here, in my USers folder for instance I've gained over two gigs by compressing some rarely used files: