On Thu, 2017-12-21 at 14:47 +0100, Joerg Sonnenberger wrote:
> On Thu, Dec 21, 2017 at 11:29:45AM +0100, zyx wrote:
> The time is not spend in the read loop, but inside PKCS7_final. So
> changing the buffer size won't make any big difference, I think.
Hi,
okay. Cory didn't mention where the most time is spent, and due to
common issues with too many particularly slow read/write operations
using somewhat small buffers, I thought it's related.
> openssl smime -binary is significantly faster, i.e. for a 200MB PDF I
> have here, it needs less than 0.6s. I've cancelled podofosign after
> 2min.
Nice, that proves it's doable, only podofosign doesn't do it
efficiently. The slowness can be due to allocating/reallocating so
large buffer too.
> One obvious issue is that podofosign is not using the streaming
> interface, but keeping all the data in memory at once.
I agree with you, reading into the memory wasn't the best choice. Using
the streaming API should lower the memory requirement as well. If you'd
have a patch for it, then it'll be great.
Thanks and bye,
zyx
|