Hi,
Enrico summarized it quite well.
The decision to use the CE as the NFS server or not should be based on
the size of your site.
If you have a large site (more than 50 WNs) I would use a separate node
a the NFS server.
This is just to avoid any load on the CE that is not related to the
CE's function.
markus
On Tuesday, May 18, 2004, at 14:20 Europe/Zurich, Enrico Ferro wrote:
> On Tue, 18 May 2004 12:47:43 +0100 Barry MacEvoy
> <[log in to unmask]> wrote:
>
>> Hmm. Having had a chance to think through this
>> overnight, I'm still not clear as to what
>> is going on - and why. Let me explain.
>>
>> At present, the LCG2 default is to create
>> a directory structure on the SE:
>>
>> SE:/stage/lcg2-data/*
>> SE:/stage/exp_software/*
>>
>> where * = alice, atlas, cms etc.
>>
>> The software area is mounted by autofs from
>> the WN to avoid problems with mounting
>> in the case of the WN being re-booted when the
>> SE is down.
>>
>> I understand that the SE is an attractive
>> place to store the experiment software if
>> only because some of it is very large, and the
>> SE, by definition, is likely to have high
>> storage capacity.
>>
>> On the other hand, the CMS model, as described
>> to me yesterday, is also attractive, because
>> one would expect close coupling between the
>> CE and WN. But then we require, certainly at IC,
>> a bigger disk on the CE than we presently have !
>> Is it really the role of the gatekeeper
>> to act as a repository ? Personally, I think not.
>>
>> What's the official policy, then ?
>> The default is for installation on the SE,
>> but our software manager - Marco - intends to
>> install on the CE...
>>
>> Steve - could you please advise as to what
>> configurational changes are required in LCFG
>> for all of the software to be shifted over to the CE:
>>
>> CE:/stage/exp_software/*
>>
>> What happens to the existing software on the SE ?
>> Will people looking for $VO_*_SW_DIR end up
>> in the right place ?
>>
>> In the meantime, we have people external to the site
>> installing software which we know nothing about.
>> This leaves me deeply uneasy...
>
> For INFN Production Grid we added the missing LCFG configuration on LCG
> profiles so every CE is exporting the directory to the WN for the VO
> software.
> To summarize it is important that:
>
> 1) uid & gid are correctly configured for the vo-sw directories
> 2) the "super-grid-user jobs" are correctly mapped into the right
> localuser
> 3) wn are mounting that area in read-write.
> 4) the environment variables are ok
>
> So if these requirements are ok, IMHO eveything should work perfectly
> also if
> you are exporting this area from the SE or whatever machine, I do not
> think
> there could be particular troubles.
>
> Best regards,
> --
> Enrico Ferro Ist. Naz. di Fisica Nucleare - Padova
> Address: via F. Marzolo, 8 - 35131 Padova - ITALY
> Email: [log in to unmask] Phone: +39-0498277154
>
>
************************************************************************
*******
Markus Schulz
CERN IT
|