Ed Porter wrote:
I found that performance began to fail miserably as the blob size increased
above 500 bytes (has anyone else experienced this problem?).
I just ran a test where I do "count" INSERTs of a BLOB of different sizes.
Here's the result:
size=100 count=100 time=26374 microseconds per iteration
size=200 count=100 time=24480 microseconds per iteration
size=400 count=100 time=31388 microseconds per iteration
size=1000 count=100 time=37764 microseconds per iteration
size=2000 count=100 time=56178 microseconds per iteration
size=4000 count=100 time=85590 microseconds per iteration
size=8000 count=100 time=137930 microseconds per iteration
size=10000 count=100 time=168661 microseconds per iteration
size=20000 count=100 time=288281 microseconds per iteration
size=40000 count=100 time=429003 microseconds per iteration
size=100000 count=100 time=877089 microseconds per iteration
size=100 count=10000 time=1128732 microseconds per iteration
size=200 count=5000 time=824365 microseconds per iteration
size=400 count=2500 time=515504 microseconds per iteration
size=1000 count=1000 time=269727 microseconds per iteration
size=2000 count=500 time=243470 microseconds per iteration
size=4000 count=250 time=206263 microseconds per iteration
size=8000 count=125 time=170581 microseconds per iteration
size=10000 count=100 time=164367 microseconds per iteration
size=20000 count=50 time=156277 microseconds per iteration
size=40000 count=25 time=150406 microseconds per iteration
size=100000 count=10 time=151819 microseconds per iteration
In the first group of tests, I hold the count constant and
just increase the blob size. Doing 100 INSERTs of a 100K
blob is only 33 times slower even though it is inserting
100 times more data.
In the second set of tests, the total number of bytes inserted
is held constant. As the blob size increases the number of
INSERTs decreases. In this test we see that it is much faster
to insert a few large blobs that to insert many small blobs.
I doesn't look to me like performances is "miserable" for
larger blobs. How did you run your tests?
The results above are on SQLite version 3.0.6 using the
Tcl bindings.
--
D. Richard Hipp -- [EMAIL PROTECTED] -- 704.948.4565
