Developer Libraries » IOCipher

When using the thread API and having get_sqlfs() get the sqlfs_t pointer each time its needed, this adds a measureable overhead, my guess is that its about 50% slower this way. Would it possible to cache the sqlfs_t* at the start of each function that needs it?

Here's the merged results of two runs of c_perf.test at only 8192 size, but you can see the general trend:

reads without transactions ------------------------------

• read 2097152 bytes in 256 8192 byte chunks in 0.081638 seconds
• read 2097152 bytes in 256 8192 byte chunks in 0.046230 seconds
  reads with transactions ------------------------------
• read 2097152 bytes in 256 8192 byte chunks in 0.062073 seconds
• read 2097152 bytes in 256 8192 byte chunks in 0.040215 seconds
  writes without transactions ------------------------------
• wrote 2097152 bytes in 256 8192 byte chunks in 0.099259 seconds
• wrote 2097152 bytes in 256 8192 byte chunks in 0.105794 seconds
  writes with transactions ------------------------------
• wrote 2097152 bytes in 256 8192 byte chunks in 4.260554 seconds
• wrote 2097152 bytes in 256 8192 byte chunks in 0.066501 seconds

Running tests using the thread API, i.e. sqlfs == 0:
reads without transactions ------------------------------

• read 2097152 bytes in 256 8192 byte chunks in 0.103902 seconds
• read 2097152 bytes in 256 8192 byte chunks in 0.051937 seconds
  reads with transactions ------------------------------
• read 2097152 bytes in 256 8192 byte chunks in 0.092716 seconds
• read 2097152 bytes in 256 8192 byte chunks in 0.044551 seconds
  writes without transactions ------------------------------
• wrote 2097152 bytes in 256 8192 byte chunks in 0.123683 seconds
• wrote 2097152 bytes in 256 8192 byte chunks in 0.095375 seconds
  writes with transactions ------------------------------
• wrote 2097152 bytes in 256 8192 byte chunks in 1.096097 seconds
• wrote 2097152 bytes in 256 8192 byte chunks in 2.242698 seconds