Difference between revisions of "Storage Networking World Convention"
| Line 46: | Line 46: | ||
Look up things on ISER technology - There appears to be nothing covering it at this conference. | Look up things on ISER technology - There appears to be nothing covering it at this conference. | ||
| + | ===36 Gigaflops to 18 Teraflops in 30 days=== | ||
| + | Full title: 36 Gigaflops to 18 Teraflops in 30 days: Implementation of the Fastest Academic Supercomputer in the United Kingdom | ||
| + | |||
| + | [[Paul Calleja]], University of Cambridge | ||
| + | |||
| + | Cambridge used MOAB for their scheduling software. | ||
| + | |||
| + | Exludus - prefetches data to local storage for use | ||
| + | |||
| + | Had many problems with 10gbit ethernet cards - eventually settled on Chelsio cards after much testing. | ||
Revision as of 19:49, 17 April 2007
This is a convention I went to from April 16th-19th, 2007.
Contents
Talks
IP Storage Protocols: iSCSI
John Hufferd, Brocade Inc.
- PDU - Protocol Data Unit
- TOE - TCP/IP Over Ethernet
- iSNS - Internet Storage Network Server
- iSCSI Boot
- Static
- Bios must support it
- May be an adapter
- DHCP
- More flexible
- PXE Boot
- Everything can be done in software
- Static
- Spoke with Todd Burkey of Xiotech, who works with storage communications over InfiniBand.
Storage Grid: Relevance and Overview
Abbot Schindler, Hewlett Packard
- FAN - File Area Network
PCI Express, Infiniband, and Storage
Ron Emerick, Sun Microsystems Paul Millard, Xyratex
- Fibrechannel over Ethernet
- PCIe - External connections
- PCIe IOV - IO Virtualization
The Infiniband Revolution
Full Title: The InfiniBand Revolution: Parallel Computing is Out-Screaming High-Performance Clusters are in
Lloyd Dickman, Qlogic
Open Fabrics Alliance - working to bring about single fabric
Check out compute cluster IB Storage - Direct connect IB to storage, thus bypassing IO servers
IB Optical transcievers - 100m Length
QSFP pluggable modules - next generation
FC latency - 20µsec
Look up things on ISER technology - There appears to be nothing covering it at this conference.
36 Gigaflops to 18 Teraflops in 30 days
Full title: 36 Gigaflops to 18 Teraflops in 30 days: Implementation of the Fastest Academic Supercomputer in the United Kingdom
Paul Calleja, University of Cambridge
Cambridge used MOAB for their scheduling software.
Exludus - prefetches data to local storage for use
Had many problems with 10gbit ethernet cards - eventually settled on Chelsio cards after much testing.
