#609 支持数据集缓存加速

Open
created 1 year ago by linfj · 1 comments
linfj commented 1 year ago
利用SSD/HDD等对章鱼的数据集进行缓存加速
linfj added the
enhancement
label 1 year ago
linfj self-assigned this 1 year ago
denglei was assigned by linfj 1 year ago
linfj added this to the v4.3.2 milestone 11 months ago
linfj commented 11 months ago
Owner
数据集缓存主要是在任务较多或者多机任务频繁数据通信导致网络带宽不够的情况下能够加速数据集的访问,模拟这种场景可通过WonderShaper工具限制节点带宽https://www.51cto.com/article/717545.html 由于节点闲置硬盘资源有限,目前数据集缓存开启的入口放在管理端,暂时只支持缓存用户数据集不支持缓存预置数据集。开启数据集缓存时,根据实际情况填写参数,由于个别节点并没有足够的磁盘空间用来缓存,因此一般需要填写节点标签选择特定的一些节点而不是所有节点。 ![image](/attachments/487aab74-8eb5-4924-8131-288b4675e18b) 数据集开启缓存后,通过以下命令查看是否已经缓存到节点,以及查看所填写的目录是否有数据。 ![image](/attachments/c5148375-8fbd-41b6-8124-cf1eb90403ed) 创建训练任务时,缓存数据集节点的亲和性调度策略分数较高,是否调度到缓存节点还取决于其他调度策略分数综合 ![image](/attachments/639cecaf-ef87-4f25-8fce-9e11eba91a61)
linfj added the
开发完成
label 11 months ago
Sign in to join this conversation.
No Milestone
No Assignees
1 Participants
Notifications
Due Date

No due date set.

Dependencies

This issue currently doesn't have any dependencies.

Loading…
There is no content yet.