Re: slowest tap tests - split or accelerate?

Hi,

On 2022-01-17 15:13:57 -0500, Robert Haas wrote:
> I guess there must be something explaining it, but I don't know what
> it could be. The client and the server are each running the checksum
> algorithm over the same data. If that's not the same speed then .... I
> don't get it. Unless, somehow, they're using different implementations
> of it?

I think that actually might be the issue. On linux a test a pg_verifybackup
was much faster than on windows (as in 10x). But if I disable openssl, it's
only 2x.

On the windows instance I *do* have openssl enabled. But I suspect something
is off and the windows buildsystem ends up with our hand-rolled implementation
on the client side, but not the server side. Which'd explain the times I'm
seeing: We have a fast CRC implementation, but the rest is pretty darn
unoptimized.

Greetings,

Andres Freund