公开(公告)号：KR100325688B1

公开(公告)日：2002-02-25

申请号：KR1019990051585

申请日：1999-11-19

Applicant: 한국전자통신연구원

Inventor： 배명남 ,  한미경 ,  박유미 ,  최완 ,  전경표 ,  김상욱 ,  김윤호

IPC: G06F5/00

Abstract: 본발명은리프노드(Leaf Node)에연결된오버플로우노드(Overflow Node)를사용함으로써, 디렉토리의분할을지연하는확장해슁의디렉토리분할시점제어방법및 그를이용한엔트리삽입, 삭제, 검색방법을제공하는데그 목적이있다. 본발명에따르면, 해쉬인덱스(Hash Index) 구조의확장해슁(Extendible Hashing) 디렉토리(Directory) 분할시점제어방법에있어서, 오버플로우(Overflow)된리프노드(Leaf Node)가새롭게발생하면, 전체리프노드수에대한오버플로우된리프노드수의비율을계산하여상기계산값이미리정하여진임계값을초과하는지여부를판단하는제 1 단계와; 상기제 1 단계에서의판단결과, 임계값을초과하면, 디렉토리의분할을수행하고, 임계값을초과하지아니하면, 리프노드뒤에새로운오버플로우노드를연결시키는제 2 단계를포함하여이루어진것을특징으로하는확장해슁디렉토리분할시점제어방법이제공된다.

公开(公告)号：KR1020000033403A

公开(公告)日：2000-06-15

申请号：KR1019980050248

申请日：1998-11-23

Applicant: 한국전자통신연구원

Inventor： 이승선 ,  이형호 ,  김상욱 ,  김윤호 ,  장연정

IPC: G06F17/30

Abstract: PURPOSE: A mutual exclusion method of a lock waiting information is provided to detect a deadlock occurred when a lock waiting information is commonly accessed by a deadlock detector and transactions in a DBMS by using a transaction table and the lock waiting information. The method can make the lock waiting information just by a static assignment of many transaction elements by setting properly the number of the static elements according to the system characteristics. CONSTITUTION: A mutual exclusion method of a lock waiting information comprises steps of using a latch on the lock waiting information pool as the mutual exclusion means, initializing the system, performing a lock waiting process in the case that a waiting lock is obtained by other transaction when a transaction requests a lock, performing a lock acquisition process, performing a lock return process, and performing a deadlock detection process with one of the above processes.

Abstract translation: 目的：提供锁等待信息的互斥方法，以通过使用事务表和锁等待信息来检测死锁检测器和DBMS中的事务共同访问锁等待信息时发生的死锁。 该方法可以通过根据系统特性适当地设置静态元素的数量，通过许多事务元素的静态分配来使锁等待信息。 构成：锁等待信息的互斥方法包括在锁等待信息池上使用锁存器作为互斥装置的步骤，初始化系统，在通过其他事务获得等待锁的情况下执行锁等待处理 当事务请求锁定时，执行锁获取处理，执行锁返回处理，并且执行上述处理之一的死锁检测处理。

公开(公告)号：KR100317229B1

公开(公告)日：2002-02-28

申请号：KR1019980050248

申请日：1998-11-23

Applicant: 한국전자통신연구원

Inventor： 이승선 ,  이형호 ,  김상욱 ,  김윤호 ,  장연정

IPC: G06F17/30

Abstract: 본 발명에서는 데이타베이스 관리 시스템내에서 교착 상태 검출기와 트랜잭션들이 공통으로 액세스하는 락 대기 정보의 상호 배제를 효과적으로 지원하는 새로운 방법에 관한 것이다. 본 발명은, 시스템 초기화, 트랜잭션의 락 대기 처리, 트랜잭션의 락 획득 처리 및 트랜잭션의 락 반환 처리를 각각 독립적으로 수행하되, 상기 각 트랜잭션 처리와 교착상태 검출기의 처리를 동시에 수행함으로써, 교착 상태 검출기와 트랜잭션들 간의 상호 간섭을 최소화하여서, 결국 동시성을 극대화할 수 있는 것이다.

公开(公告)号：KR1020010047384A

公开(公告)日：2001-06-15

申请号：KR1019990051585

申请日：1999-11-19

Applicant: 한국전자통신연구원

Inventor： 배명남 ,  한미경 ,  박유미 ,  최완 ,  전경표 ,  김상욱 ,  김윤호

IPC: G06F5/00

Abstract: PURPOSE: A method for controlling an extendable hashing directory dividing time of a hash index structure and a method for inserting, deleting and searching an entry using the method are provided to delay a directory separation by using an overflow node being connected to a leaf node. CONSTITUTION: It is decided whether a directory generated as an entry is inserted into an extendable hash at first is "NULL"(S301). If the entry is not "NULL"(S302), new directory is created and an index of a directory made by a bit as much as the directory size is calculated using a hashing function(S303). A bucket being connected to the calculated directory index becomes a home bucket(S304). It is judged whether an overflow bucket exists in the home bucket(S305). If the entry is "NULL"(S302), the above (S303) stage is repeated. If an overflow bucket exists in the home bucket, it is judged whether the last bucket of the overflow bucket being connected by a chain may insert an entry(S306). If the last bucket of the overflow bucket can not insert an entry, new bucket is allocated and connected to the last bucket of the overflow bucket(S307), and the entry is inserted into the last bucket and the process is complected(S308). If an overflow bucket does not exist in the home bucket, it is judged whether the home bucket being connected to the calculated directory index may insert an entry(S309). If the home bucket can insert an entry, an entry is inserted into the home bucket, and the process is completed(S310). If the home bucket can not insert an entry, it is detected whether a local depth is less than a global depth(S311). If a local depth is less than a global depth, two buckets are allocated and the entry in the home bucket is re-hashed(S312), it is decided whether the connected home entry may insert an entry(S313). If the home entry may insert an entry, an entry is inserted into the home bucket(S314), and the process is completed. If the home entry can not insert an entry, new bucket is allocated and an entry is inserted and connected to the home bucket(S315).

Abstract translation: 目的：提供一种用于控制散列索引结构的可扩展散列目录分割时间的方法以及使用该方法插入，删除和搜索条目的方法，以通过使用连接到叶节点的溢出节点来延迟目录分离。 构成：决定是否将作为条目生成的目录首先插入到可扩展散列中是“NULL”（S301）。 如果条目不是“NULL”（S302），则创建新的目录，并且使用散列函数计算由目录大小的位数所产生的目录的索引（S303）。 连接到所计算的目录索引的桶成为家庭桶（S304）。 判断在家用桶中是否存在溢出桶（S305）。 如果条目为“NULL”（S302），则重复上述（S303）阶段。 如果家庭桶中存在溢出桶，则判断通过链接连接的溢出桶的最后一桶是否可以插入条目（S306）。 如果溢出桶的最后一个桶不能插入一个条目，则新的桶被分配并连接到溢出桶的最后一个桶（S307），该条目被插入到最后一个存储桶中，并且处理被复制（S308）。 如果家庭桶中不存在溢出桶，则判断连接到所计算的目录索引的家庭桶是否可以插入条目（S309）。 如果家庭桶可以插入条目，则将入口插入到家庭桶中，并且处理完成（S310）。 如果家庭桶不能插入条目，则检测本地深度是否小于全局深度（S311）。 如果本地深度小于全局深度，则分配两个桶，并重新进行家庭桶中的条目（S312），则确定连接的家庭条目是否可以插入条目（S313）。 如果家庭条目可以插入条目，则将入口插入到家庭桶（S314）中，并且处理完成。 如果家庭条目不能插入条目，则分配新的桶并且插入条目并连接到家庭桶（S315）。