1 LNL SADIRC2000 SADIRC2000 Resoconto 2000 e Resoconto 2000 e Richieste LNL per il 2001 Richieste LNL per il 2001 Berti Berti 30% 30% Biasotto 100% Biasotto 100% Gulmini Gulmini 50% 50% Maron Maron 50% 50% Toniolo Toniolo 30% 30% Le percentuali sono condivise con INFN-G Le percentuali sono condivise con INFN-G
24
Embed
LNL 1 SADIRC2000 Resoconto 2000 e Richieste LNL per il 2001 L. Berti 30% M. Biasotto 100% M. Gulmini 50% G. Maron 50% N. Toniolo 30% Le percentuali sono.
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
1
LNL
SADIRC2000SADIRC2000
Resoconto 2000 eResoconto 2000 eRichieste LNL per il 2001Richieste LNL per il 2001
LNLSwitch packets loss in special conditionSwitch packets loss in special condition
First packet forwarded correctly
Some packets lost
( 10-30 packets )
All the following packetsforwarded correctly
• The switch lose some packets in particular conditions
• When the transmission restart after a pause >60 Secs
• Probably related to the MAC address table management
• Problem solved by the Error recovery procedure
10
LNLTest conditionsTest conditions
• BDN+BCN in the same network• No Readout commands from EVM to RU• Clear/Allocate optimization: the clear
command is sent together an allocate for a new event
• No command or event aggregation: each packet transport a command or a data frame relative to a single event
11
LNLTest conditionsTest conditions
• Full data transfer: data moved from/to the pc memory
• Tests performed in the 400-4000 Bytes range
12
LNLThroughput vs number of allocated eventsThroughput vs number of allocated events
15 x 15
13
LNLThroughput vs event sizeThroughput vs event size
15 x 1
1 x 15
15 x 15
14
LNLEvent rate vs event sizeEvent rate vs event size
15 x 1
1 x 15
15 x 15
15
LNLBU balancingBU balancing
15 x 15
16
LNLEvent builder scalingEvent builder scaling
15 x 15
17
LNLAggregate throughputAggregate throughput
15 x 15
18
LNLConclusions for the Square Event BuilderConclusions for the Square Event Builder
• Event Builder Demonstrator 15x15 based on copper Gigabit Ethernet interfaces tested
• Good performances and scalability
• Actual limit is in the Readout Unit (?)
• Error recovery work fine, systematic tests need
19
LNLNNext Steps for 2000ext Steps for 2000
• Asymmetric Event Builder (4x40)
• Event Builder Protocol Simulations
20
LNLNext Steps for 2001: InfiniBand (I)Next Steps for 2001: InfiniBand (I)
TODAYTODAY
21
LNLInfiniBand (II)InfiniBand (II)
Legacy host architectureLegacy host architecture
The Infiniband ModelThe Infiniband Model
22
LNLInfiniBand (III)InfiniBand (III)
• What?What?– Initial single link signaling rate of 2.5Gbaud Initial single link signaling rate of 2.5Gbaud
• Means unidirectional transfer rate of 250MB/sec with a Means unidirectional transfer rate of 250MB/sec with a theoretical full duplex rate of 500MB/sectheoretical full duplex rate of 500MB/sec
– Initial support for single, 4, and 12 wide link widthsInitial support for single, 4, and 12 wide link widths
– Point to point switched fabricPoint to point switched fabric
– Message based with multicasting supportMessage based with multicasting support
LNL SADIRC Requests for 2001SADIRC Requests for 2001
• Simple test system (4 servers + Storage Area Network + Simple test system (4 servers + Storage Area Network + Network) forNetwork) for 20012001 is possibleis possible
• Early access to the productsEarly access to the products
• Test BedsTest Beds
• Requests 2001 (valuations) Requests 2001 (valuations) • 4 servers4 servers 50 Ml50 Ml• 1 IBA Switch1 IBA Switch 20 Ml20 Ml• IBA AdaptersIBA Adapters 10 Ml10 Ml
24
LNL SADIRC2000 Global RequestsSADIRC2000 Global Requests