Loading large amount of data in a hash

[shaezi]

Hi all, I am attempting to work with a hash, but the amount of data that I'm loading into it is causing an out of memory error. The data is over 800mb.
I tried using tie, like:

use GDBM_File;
my %data_parsed;
   tie %data_parsed, "GDBM_File", "$hashfil", O_RDWR|O_CREAT, 0666;

First of all I want to know if this is a good approach or not. Second, for the values of each key I'm referencing an array. When I try to get a value for example: $data_parsed{$hashkey}[$i]I get an error message like:Can't use string ("ARRAY(0x200e1dfc)") as an ARRAY ref while "strict refs" in use. So I was wondering what I was doing wrong. Any help would be appreciated. Thanks!

[maverick]

Right idea. You just have a few little details. You can't natively store references in dbms. You'll have to serialize the arrayrefs down to scalars before you can store them, using something like Storable. Which leads to the second issue. If memory serves, GDBM has a fixed size for those scalars...Berkely BTrees do not. Try something like:

use DB_File;
use Storable;
tie %data_parsed, "DB_File", "$hashfil", O_RDWR|O_CREAT, 0666;

# inside read data loop
$data_parsed{$key} = freeze($array_ref);

# inside use data loop
$array_ref = thaw($data_parsed{$key});

DB_File is slow to create, but fast to read. That may or may not be an issue.

HTH

Update

Ya know...after taking another glance at this, 800 is a lot of data. Plus, you have a key, and array combination. Perhaps it's time to move up to a full blown database like MySQL or Postgres?

/\/\averick
OmG! They killed tilly! You *bleep*!!

[strat]

Hello,

with big data in memory (e.g. hashes of hashes of lists), I prefer using perl 5.005_03 to perl 5.6 or 5.61, because it takes about 20% less memory and sometimes saves quite a lot of runtime.

In a case I'm just working with, the statistics are about like the following:

version	cpu-time	memory
5.005_03	~ 25 minutes	480 MB
5.6.1	~80 minutes	~600 MB

I've tested this behaviour under Solaris and Linux and Win2k.

Best regards,
perl -le "s==*F=e=>y~\*martinF~stronat~=>s~[^\w]~~g=>chop,print"

[perrin]

Instead of GDBM_File, use MLDBM, with GDBM underneath.