[Battlemesh] Open Tasks

[Gui Iribarren]

On 21/02/15 21:44, Simon Wunderlich wrote:
> Hey Daniel,
> 
>> Some random thoughts which came up since last year:
>> [...]
>>  * Can we use a more portable routing protocol for the management
>>    network?
>>    batman-adv just happends to be the only protocol built-into
>>    the kernel and having several incompatible versions flying around.
> 
> Actually I wonder if that is more of a myth. Most communities use either the 
> compat version 14 (releases from mid 2011- end 2013) or compat version 15 
> (since 2014), which was designed with future backwards compatibility in mind.
> I also don't think we should use anything < version 15 for either management 
> or test networks, so I don't really see the problem.

i may be preaching to the choir, but...
in DeltaLibre, a month ago we sysupgraded a bunch of ~30 tplink wdr3500
from 2012.4.0 (compat14) to 2014.2.0 (compat15) in one go, with no
issues so far; experience was smooth: same performance, same stability,
same set of bugs :P (just kidding)
only one glitch was suspicious (ipv6 traffic was acting funny on one
node) but it turned out to be a igmp snooping bug^Wfeature introduced in
openwrt-bb-rc3 (so, nothing to do with our loved/hated bat-hero)

one myth-believer less here ;)
	
> 
>>    Also in case kexec seems too far out, this could at least allow
>>    to load different variants of the the batman-adv kernel module.
>>    I'd suggest some low-footprint layer-3 protocol, OLSRv1 would be
>>    a good choice imho, as most people are familiar with it and it
>>    doesn't have too crazy dependencies. As that depends on having
>>    IPv4 addressing configured, I further suggest using AHCP or even
>>    more simple hacks like using the last 24 bits of the MAC address
>>    to enumerate devices in the 10.0.0.0/8 network...
>>    However, I'd be happy with anything which still allows painlessly
>>    hacking on batman-adv without risking to break access to the mgmt
>>    net.
> 
> I wonder if changing from batman-adv (which is already in all the test 
> scripts) to OLSRv1/AHCP will be more reliable ... I didn't hear that test 
> management stability/reliability was a problem last year, but this is more a 
> question for the people who actually performed the test (i.e. not me).

+1,
if the fear is about the mgmt network being irreversibly broken by an
update, wibed (err... battlemesh firm) overlay scheme would prevent
that, by rolling back the faulty overlay (i.e. with deadly
batman-adv.ko) after a set time, right?

(btw: what happened to the "snapshot" tool blogic suggested last year? i
saw some simple userspace cli being included in BB, but it seems the
development path didn't go further?)