From GEANT2-JRA1 Wiki

Contents 1 EGEE 1.1 Expectation from the network 1.2 Use cases 2 EXPReS 2.1 Expectation from the network 2.2 Use Cases

EGEE

https://edms.cern.ch/file/475908/2/EGEE-JRA4-TEC-475908-NetworkPerformanceMetrics-2.0.pdf

Expectation from the network

• IP link capacity and utilisation
• Interface output drops and errors
• Delay measurements
• MTU
• Topology information
• Optional run iperf and delay test to/from servers located on the networks.
• For AA, they will be using VOMS.
• Want to access the data through the NM-WG interface as a mean of accessing underlying frameworks.

Use cases

1. Middleware: https://edms.cern.ch/file/672906/1/EGEE-JRA4-TEC-672906-NPM_UseCasesAndRequirements_RBMW-v1-1.pdf
   • Note that it is not foreseen in the short term to have the middleware making use of network data to take decisions.
   • Metrics:
     • TCP achievable bandwidth between end-sites: However, they realise that it may not be realistic to run such tests between any two locations. So they would rather like to have a model using different metrics and which could provide them an estimate of teh TCP achievable bandwidth.
     • Round-trip packet loss between end-sites
     • Round-trip time between end-sites
     • Available bandwidth between sites (NS note: can be provided by concatenating backbone measurement along the path).
     • IPDV between sites.
   • Usage
     • Two groups would make use of those data for GridFTP transfers:
       • Data Management to find out the best combination of storage element to transfer data to/from.
       • Workload Management to find out the best combination of storage element and Compute Element to transfer data to/from.
       • The following applications are targeted: file transfer, real rime application and audio-streams.
   • User
     • What is the middleware? Can someone behind the middleware access the network data? How can we make sure that it isn't the case?
     • How does the authentication works?
2. Operations (NOC and GOC):
   • Metrics:
     • Hop list (between end-sites)
     • Path MTU (the information can be provided by perfSONAR in addition to the link capacity).
     • Delay (OW or RT) (between the sites, but also to dedicated backbone measurement points).
     • Loss (OW or RT) (Personal note: to provide a better understanding of what's happening, output drops and input errors would also be needed along a path)
     • Packet re-ordering
     • Availability
   • User
     • Two groups of operational users: ROC (Regional Operations Centre) and ENOC (EGEE Network Operations Centre).
       • ROC are the distributed operational teams (almost one in each federations, http://egee-sa1.web.cern.ch/egee%2Dsa1/ROC-support.htm). They may have to access to data, but through NPM only in my opinion (Network Performance Diagnosis Tool).
       • ENOC is the centralized team who deals with all network related support in EGEE. It's more a 2nd level support. They will use the same kind of tools as ROCs but they may need also to access to lower level data (like MTU), maybe not through NPM.
     • How does the authentication work?

EXPReS

Expectation from the network

• IP link capacity and utilisation
• Interface output drops and errors
• Delay measurements
• MTU
• Topology information
• Access to iperf probes

Use Cases

• They wish to have the possibility of doing detailed bwctl tests (from begin to endpoint and to and from intermediate points) about one week and one day before doing an actual evlbi experiment.
  • Right now we have a few intermediate points available, not enough to be really useful.
  • If Geant would take care of regular (short) iperf tests between different nodes we could use this info to get an idea of the generally available bandwidth.
  • For the time being, EVN poeple at the stations (astronomers in charge of vlbi experiments) would make use of access to your iperf boxes, and we would certainly do it on a very limited scale.
• Would rather see the information on graphs on the web (Fast and convenient to interpret).
• Data Access: EVN people at the stations (astronomers in charge of the vlbi experiments)
• Run tests: EVN people at the stations (astronomers in charge of the vlbi experiments)
• Authentication: currently, they don't seem to have an authentication infrastructure and are using the GEANT username and password to access web data and have exchange AES keys to configure the BWCTL tools.