2018-02-12 14:47:50 UTC
We have a wrapper code which uses libcurl to upload data from files.
Looking at the strace when it is run on a bare metal and a VM there is a
difference in the file read/send buffer size. On bare metal it uses 16 KB
and on VM it uses 8KB even though CURLOPT_BUFFERSIZE is not set explicitly.
Process on bare metal seems to use the expected default of 16 KB and on VM
it seems to be using a calculated optimal value. Is this the expected