Re: Sharing objects

[Perrin Harkins <pe...@elem.com>]

On Wed, 2003-07-23 at 18:21, Aleksandr Guidrevitch wrote:
> What are common patterns of sharing data beetween apache processes,
> for example I'd like to share some indexes. Also, I'd like to avoid
> complex sycronyzation process (currently IPC::Sahreable seem to be the
> right thing)

No, IPC::Shareable is slow.  You are better off with one of these:
MLDBM::Sync
Cache::Mmap
BerkeleyDB (with native locking)
Cache::FileCache
IPC::MM

- Perrin

Re: Sharing objects

[Perrin Harkins <pe...@elem.com>]

Aleksandr Guidrevitch wrote:
> Actually I think to use Cache::FileCache as the storage backend.
> But I need to have Cache keys to be sorted by various criteria.
> I strive to avoid re-reading Cache::* keys and sort them each time,
> but to share somehow sorted lists beetween apache processes (as they 
> could be huge). However, having an extra key like keys_sorted_by_* in 
> Cache::FileCache will probably solve the problem.

It could, but if that's a large list it could be slow.  Cache::FileCache 
will serialize each cache value with Storable, and in this case the 
value will be one big array.

A different approach would be store a separate BerkeleyDB index as a 
BTree with a custom sorting function, or share your array as a 
BerkeleyDB recno database (serial records accessed like an array).

- Perrin

Re: Sharing objects

[Aleksandr Guidrevitch <pi...@tut.by>]

Hello Perrin

> No, IPC::Shareable is slow.  You are better off with one of these:
> MLDBM::Sync
> Cache::Mmap
> BerkeleyDB (with native locking)
> Cache::FileCache
Actually I think to use Cache::FileCache as the storage backend.
But I need to have Cache keys to be sorted by various criteria.
I strive to avoid re-reading Cache::* keys and sort them each time,
but to share somehow sorted lists beetween apache processes (as they 
could be huge). However, having an extra key like keys_sorted_by_* in 
Cache::FileCache will probably solve the problem.

Thanks

> IPC::MM

Alex