Hi,
Burke, S (Stephen) пишет:
>
> That's basically what LHCb are doing on LCG. It seems to have been
> pretty effective in maximising the use of resources.
I don't think maximisation of resource usage is the driving force.
It's just a more convenient way of managing production-scale tasks.
ATLAS production system works in a similar manner, half-way (before
it hits various grids)
>>Other users have huge datasets where you don't know beforehand what
>>data the job will need, and where you really want to access your data
>>through a database search, not by file transfer, because of the way
>>the data is structured.
>
>
> Also one of the HEP use cases which hasn't been tackled so far is to do
> partial reads of files - if you only want one event from a file it would
> be a lot more efficient to just read that event from a remote SE, than
> to replicate the whole file locally, copy it again to a local disk and
> then read through it to find the event ...
I don't think this is what Leif had in mind - I guess it's, again,
more of a case of jobs like ATLAS simulation, where, say, detector
geometry or calibration data are read by each job from an external
database. If this kind of data is changing often, it's plain
impossible to transfer them as a file. It's even not very
straightforward - if possible at all - to make properly synchronised
local replicas of such databases.
> How many of the operational problems we have in LCG are related to
> firewalls? 50%? How much manpower goes in solving them? Of course, one
> reason it's so bad is that the people who designed the internet
> protocols didn't take security into account. Nice to know that EGEE
> middleware has built in security right from the start, like it said in
> the proposal documents ... :)
... which reminds me of one serious advice on how not to overload a
database server: do not query it. How to keep your cluster secure?
Don't turn it on. Everything a man had made, another man can (and
will) destroy. If internet had a security built in, it would never
spread around like it did, and, well, it would have been cracked
sooner or later anyway. Did anybody think why Web grew in such a
fantastic manner? No, not because it was born at CERN ;-) Because it
was - and still is - anonymous. And because port 80 is not
firewalled ;-)
Cheers,
Oxana
|