Performance Working Group - Meeting Minutes
February 02, 2010, 12:30 PM - 1:50 PM
US / Canada Mountain
Welcome (5 min)
Co-Chairs: Carla Hunt, MCNC and Chris Hawkinson, CENIC
Community Updates (10 min)
1. Internet2 update, given by Jason Zurawski
- Software Releases
Last official release of perfSonar -ps and the Toolkit in 2009
Roadmap for 2010 - conversion to RHEL lineage, integrating with Nagios
Longer term - AA integration into ps tools, Configuration management, Possible rewrite of NDT and integration with BWCTL, evaluation of TCP implementation - right now is RENO
- Product Adoption:
USATLAS continues to utilize pSPT, preparing recommendations for Tier 3's
USCMS has no official stance on performance measurement
OSG and some performance tools in VDT distribution
GLIF - MANLAN and dynamic circuit interplay demonstration in 2009
Google MLAB- Deployment continues, NDT development in the past year has been driven by MLAB
Continue to be heavily involved in OGF participation. Four working groups: NM, NMC, NML, NSI
I2 assisted REDDnet in a BW Challenge entry
ION and I2 connections to Portland 20G total
perfSONAR measurement infrastructure at SC'09 covered in Aaron's talk today!
Matt Z will be leading SCinet Measurement next year
b) APAN - next week
Open positions on the team!
Discussion about backup of measurement data on toolkit hosts. Consider moving data to alternate location for long term archiving. Toolkit not really designed with robust backup capability.
2. eduPERT activity (http://edupert.geant.net/) Chris Welti, SWITCH
Help desk for users of R&E network users
Typically, NOC doesn't see a problem and hands off an issue for further investigation
Used to be a central PERT for the entire community
Recently implemented a new model where each NREN has implemented its own PERT
This is a federation of PERTS with some central functions like knowledge base, directory, accreditation, mailing list
Range from 2 days to 2 hours minimum response time
Knowledge Base - barriers to contribution low, community-based review of documents, sometimes mailing lists discussions become documents
Reporting on statistics of number of cases worked by PERTs
Also do on-boarding/organized workshops for new organizations that decide to form a PERT
We are welcome to participate if we have an organization that is interested
Federation based on people and relationships. No tools requirement per se.
Discussion - Does the PERT see emergent community-related performance issues or one offs. Typically, one-offs related to TCP buffering issues.
Please see slides for each of the community presentations.
Prasad Calyam, Ohio Supercomputing Center (15min)
"How should an active measurement service operate within an
experimenter's slice (encompassing Internet2 backbone resources)
in the GENI facility?"
What does GENI stand for? Global Environment for Network Integration
What are the metrics for measuring part of a slice? One possibility could be was the slice request fulfilled.
Measurement design so far collects measurement within the same domain as the slice.
Some passive measurement within and outside the slice
How is a slice provisioned?
built on several different substrates
one option is vlan
could be wireless channels
With shared infrastructure, one consideration could be how do I know when another slice impacted my slice?
-queuing and resource utilization considerations
Here in SLC implementation, not possible to request slices that exceed hw capacity.
Katsuhiro Sebayashi, NTT Network Innovation Labs (15min)
"PRESTA 10G, a 10Gbps network interface card (NIC) with highly-accurate traffic-measurement capabilities"
Has presta 10G been looked at for bulk data transfer
Approximate cost of a card is 3000.
Dr. Kenji Shimizu, NTT Network Innovation Labs (15min)
"Development of Measurement Archive (MA) capability for PRESTA 10G that is compatible with the perfSonar measurement infrastructure"
A bit of discussion about the data storage format - why RRD? Not entirely clear why there was a need to develop an HRA interface.