Add standalone netperf code directly within site_wifitest

Add standalone netperf code directly within site_wifitest

Create a version of netperf test directly within site_wifitest that:

  - Only runs if netperf is pre-installed on both client and server
  - Knows about opening a hole for the netperf control socket in the netbook firewall
  - Can start up the netserv on one end, and netperf on the other
  - Parses the output of netperf itself, and will make separate perf
    numbers for each netperf run within the test suite
  - A few fixes to the "ping" code so that each iteration of ping stats are displayed

BUG=none
TEST=Run MatFunc 034* -- running full MatFunc while I sleep

[Sam Leffler, 2010-08-31 17:40:12]

Yowza!  Great stuff.  May want to run multiple instances though perhaps it's
better to just get iperf setup the same way and do it with that as it's got
builtin support and/or finer-grained controls (e.g. to limit b/w)

http://codereview.chromium.org/3275009/diff/7001/8001
File server/site_wifitest.py (right):

http://codereview.chromium.org/3275009/diff/7001/8001#newcode139
server/site_wifitest.py:139: '/usr/sbin/netserver')
on the client these land in /usr/local (via chromeos-test)

http://codereview.chromium.org/3275009/diff/7001/8001#newcode545
server/site_wifitest.py:545: (self.iptables_cmd, proto, proto, src))
need a method to close the firewall hole too

http://codereview.chromium.org/3275009/diff/7001/8001#newcode559
server/site_wifitest.py:559: self.__firewall_open(self.client, 'udp', client_ip)
I think this belongs in the caller

http://codereview.chromium.org/3275009/diff/7001/8001#newcode568
server/site_wifitest.py:568: return None
I'd prefer to just fail as the full autotest use is a hack already and doesn't
work for our needs (e.g. because the server side cannot be controlled and may
have a different architecture which autotest can't handle).

http://codereview.chromium.org/3275009/diff/7001/8001#newcode579
server/site_wifitest.py:579: prefix = 'Netperf_%d_' % self.netperf_iter
Need to include 'test' and 'modem' in the prefix; otherwise I think keyvals will
collide between client+server

http://codereview.chromium.org/3275009/diff/7001/8001#newcode635
server/site_wifitest.py:635: self.kevals[prefix + 'Trasnfer_Rate'] =
float(lines[6].split()[5])
IIRC there's lots more useful data we can get from netperf like cost/packet but
this is a great start

[Paul Stewart, 2010-09-01 16:45:05]

Many changes, hopefully for the better.  PTAL.   Specific scrutiny should be
applied to my removal of the old neperf code and my treatment of firewall rule
creation and cleanup.

http://codereview.chromium.org/3275009/diff/7001/8001
File server/site_wifitest.py (right):

http://codereview.chromium.org/3275009/diff/7001/8001#newcode139
server/site_wifitest.py:139: '/usr/sbin/netserver')
On 2010/08/31 17:40:12, Sam Leffler wrote:
> on the client these land in /usr/local (via chromeos-test)

Done.

http://codereview.chromium.org/3275009/diff/7001/8001#newcode545
server/site_wifitest.py:545: (self.iptables_cmd, proto, proto, src))
On 2010/08/31 17:40:12, Sam Leffler wrote:
> need a method to close the firewall hole too

Done, but this opens a whole can of worms generally related to "cleanup" in
site_wifitest.  Cleanup usually appears only to be called after an error, and
not otherwise.   We need to chat about whether this should change, or be a
(non-default) option.  For now I've made it so that __run_netperf() cleans up
after itself with respect to iptables.

http://codereview.chromium.org/3275009/diff/7001/8001#newcode559
server/site_wifitest.py:559: self.__firewall_open(self.client, 'udp', client_ip)
On 2010/08/31 17:40:12, Sam Leffler wrote:
> I think this belongs in the caller

I'm not sure whether the caller (or rather the original autotest stuff) is/will
be sensitive to firewalls.  If it ever worked, it will have needed to be, since
it's presumably meant to work by itself.  Therefore my decision was to do this
only locally to the standalone version.

http://codereview.chromium.org/3275009/diff/7001/8001#newcode568
server/site_wifitest.py:568: return None
On 2010/08/31 17:40:12, Sam Leffler wrote:
> I'd prefer to just fail as the full autotest use is a hack already and doesn't
> work for our needs (e.g. because the server side cannot be controlled and may
> have a different architecture which autotest can't handle).

Done, in the sense that I've eliminated the old code, and changd this to an
error.TestFail()

http://codereview.chromium.org/3275009/diff/7001/8001#newcode579
server/site_wifitest.py:579: prefix = 'Netperf_%d_' % self.netperf_iter
On 2010/08/31 17:40:12, Sam Leffler wrote:
> Need to include 'test' and 'modem' in the prefix; otherwise I think keyvals
will
> collide between client+server

Done.

[Sam Leffler, 2010-09-01 19:21:12]

LGTM (maybe update cros-*.config files to record the rspro ip's though I know we
should be using your newer script)