Re: Long Post - Re: How are OO databases doing

From: bob <bretlb_at_gemstone.com>
Date: 21 Feb 2006 09:27:31 -0800
Message-ID: <1140542851.734199.199740_at_z14g2000cwz.googlegroups.com>


Arturo Hernandez wrote:
  I also remember hearing about objects pulling other referenced objects along when they were not required just because the objects being loaded were pointing at them. Is that an issue now?

In the GemStone products we implement a multi level cache. The first level (the one closest to the disk) caches pages of objects. The second level of cache is the virtual machine object memory. When an object is referenced and not present in the VM object memory it is located in the page cache and copied into the VM memory. Because we extend the byte codes to understand our Persistent Object Memory ObjectIds (PomIds) we don't need to bring an entire graph into the virtual machine and objects are faulted in one at a time.

Other systems that use serialization would probably have the problem you describe since they must bring in the entire serialized graph of objects even if they only need one.

Bob Bretl
Principal Engineer, Facets Technologies
GemStone Systems Received on Tue Feb 21 2006 - 18:27:31 CET

Original text of this message