Transcript slides

Panel discussion about
networking architecture
issues
[email protected]
www.renater.fr
GRID@LARGE Workshop, Lisbon, 29-30 August 2005
1
DEDICATED vs SHARED
• Network configuration <--> GRID interaction
• DEISA will be provided with an star OPN (VPN today)
• Future of HPC infrastructure should be a mix of core network
and access infrastructure. With multiple connections to all HPC
sites and not all of them on the core part!
GRID@LARGE Workshop, Lisbon, 29-30 August 2005
2
DEDICATED vs SHARED
• Network configuration <--> GRID interaction
• LHC (T0-T1 and T1-T1) will be provided with an OPN
• LHC (T1-T2) will used a shared resource, with dynamic services
configuration (BoD) ==> EGEE
• Dedicated network (unique OPN) is the cheapest solution if there
was a single pan-European project (LCG). Reality is not that, and
not a technical optimum either (answer to Kors Bos!).
• eVLBI is using a VPN, and will migrate eventually towards an
OPN in the future
GRID@LARGE Workshop, Lisbon, 29-30 August 2005
3
Scalability
• Star topology is the initial sketch for all projects, which
contradicts completely the grid spirit.
• Ideal full mesh topology is not scalable
• Star topology is not scalable either! (If it was, it should
not be resilient)
• Internet model is the most scalable (made of core +
access infrastructure)
• OPN must comply with this
• Central overall management is required. Shared AUP (CA
at large) is required.
GRID@LARGE Workshop, Lisbon, 29-30 August 2005
4
OVERLAP
• Grid community is discovering what telecommunications
know and have been performing for long time already :
•
•
•
•
24/7,
upgrade cost and complexity,
interdomain and interoperability issues,
monitoring, central management or standardization forces with
operational interoperability of management,
• Security.
GRID@LARGE Workshop, Lisbon, 29-30 August 2005
5
?
Questions
GRID@LARGE Workshop, Lisbon, 29-30 August 2005
6