Both tcmalloc_minimal and jemalloc outperform the native malloc
implemention on Windows (MSVCRT) in terms of committed memory
consumption (~-30%) and performance (e.g. far less page faults, ~-60%),
depending, of course, on the actual workload.
The longer the download queue, the bigger the impact ;)
On *nix the picture is a little different... tcmalloc usually still
outperforms the native malloc implementation, but not that significantly
than on Windows. jemalloc however is only marginally better than recent
native Linux implementations, while it is already used by some BSD as the
native allocator.
tcmalloc is part of gperftools and very mature and tested by now. It
doesn't work on OSX in the default configuration, however.
http://code.google.com/p/gperftools/
jemalloc is the default allocator at least on FreeBSD and NetBSD and
used in Firefox.
http://www.canonware.com/jemalloc/index.html
If we got IPv4 lookup response, we don't wait for IPv6 lookup
response. This is because DNS server may drop AAAA query and we have
to wait for the long time before timeout. We don't do the inverse,
because, based on todays deployment of DNS server, almost all of them
can respond A query just fine.
Currently, aria2 checks configured addresses at the startup. But there
are chances that interfaces are not setup at that moment. For example,
if aria2 is used as daemon, it may start before network interfaces
up. To workaround this, we check addresses again if both addresses are
not configured at the startup.
This change makes --save-session save only options specified for
download, more specifically, options in command-line, -i file and via
RPC. The other options from conf file and default values are not
saved. This will drastically decrease the size of session file.
The currently used URIs are inserted back into remaining URI list in
FileEntry::putBackRequest(), which overlaps to some of the URIs in
spentUris_. If we save spent URIs, each time save is performed, the
number of URIs are increased due to this overlap. This change fixes
this bug.
LibUV event will use the best available polling method on a system, kind
of like aria2 does already with the different *EventPoll
implementations.
However, libuv may support different/newer polling mechanisms; for
example on Windows it will use IO Completion Ports which are superior to
select() ;)