Closed jti-lanl closed 5 years ago
It's been some time since I looked into this problem, but I believe that the rewrite of the ne_read() logic included in pull request #18 should avoid it. I'll close this issue after the merge of that request is complete.
Pull request has been merged. Closing this issue.
[Developers are already aware of this. Just capturing it here, so it won't slip through the cracks.]
For all blocks except the first, in each stripe, it appears that there is a 1MB read, followed by a seek back to zero, then another 1MB read.
Read performance is already good, even with this issue. Fixing it should make performance that much better.