問題文
A team confirms the direct storage path works for large reads but finds that small reads of a few kilobytes go through the ordinary path. Nothing appears broken. What accounts for this?
選択肢
- The direct path requires transfers to be a multiple of the page size, and small reads never satisfy that.
- The file system in use does not support the direct path for reads under a fixed size defined by the vendor, and the library falls back to the ordinary path for anything below it.
- A configurable size threshold sends operations at or below it through the ordinary path, since the direct path's setup cost does not pay off for tiny transfers.
- Small reads are always served from the accelerator's own cache and therefore never appear on either path, so what the team is observing must be the metadata operations that accompany the reads rather than the reads themselves.