2008-11-21 fgmailbox 发件人： fgmailbox 发送时间： 2008-11-21 11:41:59 收件人： 49259BE6.8010704; joaquin 抄送： 主题： Re: [Boost-users] [multiindex] Why not to add B+Tree index to multiIndex? To: "boost-users@lists.boost.org" Message-ID: <49259BE6.8010704@tid.es> Content-Type: text/plain; charset=GB2312 fgmailbox escribi??:

we can use multiindex as a memory db,but the multiindex only support hash index and Order index(RB TREE),why not to add b+ Tree Index to multiindex to make it much faster .

I'm no expert in B+ trees, but I understand that these structures are more effective than regular binary trees when secondary storage (i.e. hard disk) is used, which is not the case for an in-memory container like multi_index_container. Have you any reference on the performance of B+ trees in in-memory scenarios? Joaqu??n M L??pez Mun~oz Telef??nica, Investigaci??n y Desarrollo Hi joaquin: I found b+ Tree code in a GNU Database Source Code of [firebird] Database code BePlusTree< > template class In [tree.h],your can find this source code in http://sourceforge.net template < typename Value, typename Key = Value, typename Allocator = MallocAllocator, typename KeyOfValue = DefaultKeyValue<Value>, typename Cmp = DefaultComparator<Key>, int LeafCount = LEAF_PAGE_SIZE / sizeof(Value), nt NodeCount = NODE_PAGE_SIZE / sizeof(void*)

class BePlusTree {....} the Allocator can be a memory allocator or a disk file allocator,so this template can use as a memory database index ,and I have use the template to test BePlusTree and stl::set to test performance, the BePlusTree is faster than set in the condition of large scale data query. 2008-11-21 fgmailbox