On 25/10/07 15:40, Matt Doidge wrote:
> What struck me as being odd with our situation is that dcache used the
> larger pool even though it had less space. Nothing appears to be
> broken, but some weighting somewhere seems screwed. Could someone
> please remind me if the pool choice can be weighted by factors other
> then CPU and Space?
Nope, just CPU and Space at the moment. Actually, looking at the
documentation, the CPU cost seems to now be called the performance cost:
http://www.dcache.org/manuals/Book/cf-pm-cm.shtml
"The load of a pool is determined by comparing the current number of
active and waiting transfers to the maximum number of concurrent
transfers allowed."
Greig
> On 24/10/2007, Ross, D (Derek) <[log in to unmask]> wrote:
>> Hi Matt,
>>
>> Yup, IME dCache will normally pick the pool with most free space.
>>
>> Derek
>>
>>> -----Original Message-----
>>> From: GRIDPP2: Deployment and support of SRM and local storage
>>> management [mailto:[log in to unmask]]On Behalf Of Matt
>>> Doidge
>>> Sent: 24 October 2007 16:52
>>> To: [log in to unmask]
>>> Subject: Re: New dcache Pool being ignored.
>>>
>>>
>>> Sorry to reply to myself.
>>>
>>> After getting some wisdom from Greig during the storage meeting I
>>> tried an experiment where I removed dteam access to all pools except
>>> the new one, and tried copying files in. This worked fine, with all
>>> the files going into the largest of the pool partitions (the new pool
>>> node hass eight 1TB partitions and one 1.5TB partition). However if I
>>> enabled dteam on even just one of the old pools (I choose the one with
>>> the least amount of space) then that pool was used in preference. I've
>>> tried fiddling with the CPU/Space factors either way but nothing
>>> changed this-old pools were always favoured. One important thing to
>>> note is that the new pools are individually smaller then the old (1 TB
>>> compared to 1.5 TB), so there could be infact nothing to worry about.
>>> I'd just like someone elses opinion on that.....
>>>
>>> cheers,
>>> Matt
>>>
>>>
>>>
>>> On 22/10/2007, Matt Doidge <[log in to unmask]> wrote:
>>>> Hello,
>>>> We've installed a new pool node here at Lancaster but it seems to be
>>>> being ignored by the dcache, i.e. files aren't being put
>>> into it. It's
>>>> been left for a few days now of normal use but still no user files
>>>> have shown up on it. Firewalls check out okay (after a bit
>>> of a fiddle
>>>> on friday), all the cell services are talking. The pool
>>> node has doors
>>>> in it which seem to talk with the rest of the dcache
>>> without trouble.
>>>> I've tried some experiments and successfully pulled off
>>> manual pool to
>>>> pool transfers from other pools to the new one. I also tried a test
>>>> where I deleted all replicas of a few files except the copies on the
>>>> new pool and tried to access them through the various normal ways,
>>>> which also worked. The logs aren't giving any signs of what's up.
>>>>
>>>> My best guess is that this is a pool selection
>>> configuration problem,
>>>> but I can't see any differences in the new pool and old
>>> pools with the
>>>> same settings (only atlas and dteam). Any ideas of where to
>>> look next
>>>> would be appreciated.
>>>>
>>>> cheers,
>>>> Matt
>>>>
|