基于数据相关性的多片闪存系统垃圾回收器

A data affinity based garbage collector for multi-bank flash-memory storage system

研制了一种基于数据相关性的垃圾回收器.通过更新频率计数器链表机制,检测出具有相关性的数据.垃圾回收器对相关性数据进行迁移,集中到同一个片内,有效地延长了片处于关闭状态的时间,有利于抵消同步损失,达到节能的目的.使用能量权函数来衡量处于不同能量状态的片的回收代价,最小化对处于休眠态的片进行回收的次数.回收块选择算法采用加权代价函数的方法,考虑了能量权和迁移权,使得回收耗能最小.实验结果显示:规格化能量减少了大约15%,规格化延时增加了6.3%;其乘积比代价-收益策略减少了9.6%.该垃圾回收器能够用较小的性能损失获得较大的节能比例.

A novel garbage collector based on data affinity was designed.Data with the same update frequencies was detected by update-counter list and clustered into the same bank during data migration.The shutdown time of banks was lengthened greatly.It is highly advantageous to compensate the resynchronization penalty to save energy.Moreover,using energy coefficients to evaluate the cost of recycling a bank being in different energy state.Recycled-object selection algorithm considered energy and migration coefficien...