Print

Print


Hi Alastair,
this is great thanks for this.
The last statement about network connectivity is going to have to be investigated with JANET as a matter of urgency.
The issue here is what do ATLAS expect this to grow to? 12, 15 or 20 Gig?
regards,
Mark

On 18 Nov 2013, at 13:46, Alastair Dewhurst <[log in to unmask]> wrote:

Hi All

Roger got an official reply from ATLAS.

Alastair


Begin forwarded message:

From: Roger Jones <[log in to unmask]>
Subject: Re: ATLAS hardware recommendations for Tier 2s
Date: 18 November 2013 13:40:11 GMT
To: Alastair Dewhurst <[log in to unmask]>

Hi Alastair,
   Borut blessed the following as a correct and accurate summary of the current statements from various CEB and ICB discussions, representing the ATLAS position on new hardware.
   Best
   Roger

The VO card represents a bare minimum for ATLAS usage, and is set so that we can make maximum use of existing (and often aging resource). However, many ATLAS tasks require more than the minimum on the VO card. Indeed, reconstruction, which is an increasing feature on Tier 2s as well as Tier 1s and Tier 0, currently requires 3GB/job slot, even using AthenaMP. While we are working on keeping the memory and other resource usage down, the increasing pile-up makes increases likely in all activities.

Accordingly, ATLAS would wish new purchases to have at least 3GB/job slot.

 

While there are various changes underway on the computing model, including data federation, the best estimate for the disk to CPU balance at a Tier 2 remains essentially unchanged.

 

ATLAS Tier 2s should have ~10GB connectivity. This requirement is likely to rise over run 2.


Best

Roger



--------------------------------------------
Mark Mitchell,
ScotGrid Technical Co-ordinator,                  
Rm 427b, 
Kelvin Building,                      
School of Physics and Astronomy,             
University of Glasgow,                  
G12 8QQ, UK      
Telephone: +44-141-330 6439
E Mail: [log in to unmask]