Documentation Source Text

Check-in [3563b95114]
Login

Many hyperlinks are disabled.
Use anonymous login to enable hyperlinks.

Overview
Comment:More tweaks to the 35% faster document.
Timelines: family | ancestors | descendants | both | trunk
Files: files | file ages | folders
SHA3-256: 3563b95114b93f967d7bbe43af011ac812365b3c6738b5227b62232214ced9af
User & Date: drh 2017-05-04 20:23:00
Context
2017-05-05
16:55
Updates to the 35% faster document. check-in: ac27ab7f6f user: drh tags: trunk
2017-05-04
20:23
More tweaks to the 35% faster document. check-in: 3563b95114 user: drh tags: trunk
20:14
Fix typos and make improvements to the 35% faster document. check-in: 4764f16dc0 user: drh tags: trunk
Changes
Hide Diffs Unified Diffs Ignore Whitespace Patch

Changes to pages/fasterthanfs.in.

99
100
101
102
103
104
105
106



107
108
109
110
111
112
113
relative to direct filesystem reads.

<p>When constructing the test data, trying varying the size of the blob.
The performance advantage will shift toward direct filesystem reads as
the size of blobs increase, since the cost of invoking open() and close()
will be amortized over more bytes transferred using read().  The break-even
point, the point where it becomes faster to read directly from the filesystem,
will vary from one system to another.




<p>The --blob-api option causes database reads to occur using the
[sqlite3_blob_open()], [sqlite3_blob_reopen()], and [sqlite3_blob_read()]
interfaces instead of using SQL statements.  Without the --blob-api
option, a separate SQL statement is run to read each blob and
the performance of reading from the database is approximately
the same as the performance from reading directly from files.







|
>
>
>







99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
relative to direct filesystem reads.

<p>When constructing the test data, trying varying the size of the blob.
The performance advantage will shift toward direct filesystem reads as
the size of blobs increase, since the cost of invoking open() and close()
will be amortized over more bytes transferred using read().  The break-even
point, the point where it becomes faster to read directly from the filesystem,
will vary from one system to another.  In the other direction, reducing the
blob size provide more advantage to database reads.  With a 5 KB blob size,
reading from the database is twice as fast and uses 60% less space than
blobs stored as individual files.

<p>The --blob-api option causes database reads to occur using the
[sqlite3_blob_open()], [sqlite3_blob_reopen()], and [sqlite3_blob_read()]
interfaces instead of using SQL statements.  Without the --blob-api
option, a separate SQL statement is run to read each blob and
the performance of reading from the database is approximately
the same as the performance from reading directly from files.