This is my post in O3 Hubei partner community, and the public number is also posted. Address:
https://cn.o3community.huawei.com/community/o3communityPortal.htm? source=circlePostDetail&sxz-lang=zh_CN&schoolId=1663500457860972546&activeIndex=7&subIndex=1&thirdIndex=1&circleId=SC202311280001&blogId=cd8eae9f-401d-4e12-8e7d-8d8a8c513b00&isLogin=1
i. Preface
II. Natural characteristics of full flash storage
third, what is the phenomenon in reality when writing about the cliff
assume that when a customer first uses full flash storage, it takes half an hour for a job that mainly writes data. As time goes by, the job gradually needs 1 hours, excluding other influences, this is the phenomenon of writing cliff. Generally speaking, it is "slowing down", and there is an obvious decline in writing performance on the graph. Writing cliffs is not a unique phenomenon of any manufacturer, but a natural problem caused by storage media based on electronics. IV, SSD the slower the disk is used4.1 suppose a full flash storage 4 number block, suppose 1 number block only 4 number page
4.2 create a file ceshi.txt(Orange), assuming occupied 6 number page
4.3 modify this ceshi.txt, assuming all 6 number page all changed. By Feature 2 new writing principle, in the new 6 number page(Yellow), original 6 number page marked as invalid "(Gray) 4.4 create a new file yewu.txt, suppose so 6 number page size. Only left now 4 number page yes, it is not enough. Can it be marked as "invalid" at this time? page clean it up for users. By Feature 3, press block erase, put the upper left corner block wipe off 4.5 erase it before writing yewu.txt(Green) 4.6 in fact, at this time, let's think about it, create yewu.txt is the time already created ceshi.txt one more step is required: What is the pre-write erasure procedure? Yes, that's right. As the Solid State Disks are continuously written, data is accumulated in new locations and the capacity is not enough. At this time, if we write again, we need to start another processing step, that is, cleaning up. Clean up blocks that were originally marked as "invalid". We call this process recycling. As you use SSD disks, the tasks that need to be recycled become heavier and heavier, thus slowing down. Five, not only recycling, but also relocation
4.4 create a new file yewu.txt, suppose so 6 number page size. Only left now 4 number page yes, it is not enough. Can it be marked as "invalid" at this time? page clean it up for users. By Feature 3, press block erase, put the upper left corner block wipe off
4.5 erase it before writing yewu.txt(Green)
4.6 in fact, at this time, let's think about it, create yewu.txt is the time already created ceshi.txt one more step is required: What is the pre-write erasure procedure? Yes, that's right. As the Solid State Disks are continuously written, data is accumulated in new locations and the capacity is not enough. At this time, if we write again, we need to start another processing step, that is, cleaning up. Clean up blocks that were originally marked as "invalid". We call this process recycling. As you use SSD disks, the tasks that need to be recycled become heavier and heavier, thus slowing down.
Five, not only recycling, but also relocation
after relocation:
5.2 create after relocation omg.txt(Blue)
6. Before writing data, it needs to be recycled and relocated. It is strange that it remains slow.
VII. Summary
second, the selection must be thorough and careful POC, keep running, keep writing, get scientific and reasonable data, "go through the motions" is not acceptable. Third, pay attention to the capacity usage of all-flash storage during maintenance. This is what I have been emphasizing with my team, capacity usage may only be space for business users, but it also indicates performance for our storage maintainers. We need to look deeper, think more forward, and be more comprehensive.
This article is reprinted from: Ashburton public account