Conversation
adilger
left a comment
There was a problem hiding this comment.
Should wait on landing until testing has confirmed this fixes the issue.
| @@ -44,6 +44,12 @@ static double run(void){ | |||
| u_argv_push(argv, "-s=10000000"); | |||
There was a problem hiding this comment.
As previously discussed, it would be good to add a separate patch that sets -s=(file_size / (10 * 4096)) (i.e. 10% of file size) so that the random reads don't have too many cache hits. The sqrt(file_size / 4096) is the chance of one collision with random reads, and we could work out the math so that the random read phase exits when the chance of repeated collisions is too high. 10% of the file size is just a random guess.
| echo It will output OK at the end if builds succeed | ||
| echo | ||
| IOR_HASH=8ab8f69b32b919 | ||
| IOR_HASH=4a34dfc |
There was a problem hiding this comment.
A slightly longer hash would be preferred (10-12 chars). At only 7 chars this is likely to collide after only 16k commits in IOR. While not an issue today (about 1k commits in IOR), it is better to think about this for the future.
There was a problem hiding this comment.
Agreed, I do use the hash proposed by GitHub to id the version.
Supporting singe shared file and shifting random read away from previously written node.