Re^8: Serializing a large object

[daverave]

Thank you BrowserUk. I greatly appreciate your kind attention.

I guess I will stick with the original, space-consuming implementation. Does it makes sense to compress the store files? If so, can you think of any particular module/method that is most appropriate for compressing this kind of data (I usually use IO::Compress::Gzip but I know there are other options too.

Thanks again for your help,
Dave.

[BrowserUk]

Does it makes sense to compress the store files?

Yes & no. :(

• Yes.

  I generated a random set of 3,000 overlaps--positive & negative--with a maximum range of 10,000.

  The nstore'd file on disk was: 26/09/2010  15:26        60,783,878 fred.bin.

  gzipping that resulted in:     26/09/2010  15:26           423,984 fred.bin.gz.

  It'll certainly save you large amounts of disk space. But that's not your aim.

• No.

  The problem is that whilst you save time reading from disk. You send time decompressing. And in the end, much of the time spent retrieve()ing the data, is the time required to allocate the memory and reconstruct it.

It would certainly be worth you investigating the idea with your real-world datasets; and it will absolutely save huge amounts of disk space. But whether it will actually load faster will depend on many factors; you'll have to try for yourself with real data.