[HCoop-Discuss] Frank's confusion

[Derrick Brashear]

On Wed, Aug 19, 2009 at 8:41 PM, Adam Chlipala<adamc at hcoop.net> wrote:
> Clinton Ebadi wrote:
>> Adam Chlipala <adamc at hcoop.net> writes:
>>
>>> This seems like a deliberately disingenuous argument.  If we weren't
>>> using a network filesystem, then a failure of network hardware wouldn't
>>> break the filesystem.  It's that simple.
>>>
>>
>> Yes, it would break everything instead.
>>
>> Instead we are merely lacking the read only copy of our afs volumes and
>> a secondary KDC which is causing the underlying issues with deleuze to
>> be more noticeable (transient failure to check credentials -> failure
>> instead of checking krunk and succeeding).
>>
>
> We must be talking past each other here.  krunk is filling a role that
> doesn't exist in a system without distribution.  Naturally, if the
> network card fails on the machine where the files live, then services
> based on those files aren't going to get out into the world.  The
> problem in our case is that, compared to a monolithic server with a
> local filesystem, we have strictly increased the number of network cards
> whose failures can break things.  Under any reasonable assumptions about
> independence of card failures, this increases that chance that some
> network card will fail in a way that interferes with file access.

i have to say, i've had one network card fail out of hundreds,
including servers i managed, in like 15 years. my assumption is if a
network card fails, i replace it and move on with my life, and figure
out what the next failure, probably unrelated, will be. like
single-bit memory failures, or head crash, or cpu fan death, or....

-- 
Derrick