a*****g
发帖数: 110 | 1 一般文件系统读写block
real world case下 block之间会有怎样的分布？
一般那些file system benchmark都用pure random的来做测试
但实际情况呢？
请大牛指教，谢啦～ | k**********g
发帖数: 989 | 2 sequential, random, row, column, burst, scheduled, mixed, replicate,
read/write ratio | a*****g
发帖数: 110 | 3 我想问的是block读写的spatial & temporal locality
会有怎样的分布呢？
【在 k**********g 的大作中提到】
: sequential, random, row, column, burst, scheduled, mixed, replicate,
: read/write ratio
| k**********g
发帖数: 989 | 4 Physical disks only need to evaluate pure random access performance (subject
to disk utilization ratio, e.g. only using the first 30% of disk sectors /
cylinders to reduce seek time)
This is because OS and database caching have already taken care of
predictable access patterns. Sequential access is usually very fast (near
the interface's top speed) that it is not a major concern.
If you are concerned with real world performance when used with a particular
database, this actually turns the whole question into a database
performance question. Because that performance is largely determined by how
the OS and database arrange its storage space on the disk.
To perform experiments, you would need to take a production database disk,
make a sector-level copy onto the test disk, then run experiments on that
test disk. |
|
