October 20, 2000 Y.Yasu 1
Quality of Service on Gigabit Ethernet for Event Builder Y.Yasu (1) - - PowerPoint PPT Presentation
Quality of Service on Gigabit Ethernet for Event Builder Y.Yasu (1) - - PowerPoint PPT Presentation
DAQ2000, Lyon ATLAS-Japan DAQ Group Quality of Service on Gigabit Ethernet for Event Builder Y.Yasu (1) , Y.Nagasaka (2) , H.Hasegawa (3) , A.Manabe (1) , M.Nomachi (4) , H.Fujii (1) , Y.Watase (1) (1)High Energy Accelerator Research
October 20, 2000 Y.Yasu 2
ATLAS-Japan DAQ Group DAQ2000, Lyon
Contents
Gigabit Ethernet and Quality of Service(QoS) for Event Builder Basic performance of Gigabit Ethernet QoS in Linux QoS performance on Gigabit Ethernet
October 20, 2000 Y.Yasu 3
ATLAS-Japan DAQ Group DAQ2000, Lyon
Congestion Avoidance by Global Traffic Control(Fermilab, KEK) Special Hardware with Barrel shifter etc. by traffic shaping for each node(RD31/ CERN) ATM including QoS(Constant Bit Rate)
Gigabit Ethernet and Quality of Service for Event Builder
How can we use Gigabit Ethernet with TCP/IP ?
October 20, 2000 Y.Yasu 4
ATLAS-Japan DAQ Group DAQ2000, Lyon
Basic performance of Gigabit Ethernet
Hardware configuration and setup Performance measurement of memory copy Tuning network driver Performance measurement of Gigabit Ethernet
October 20, 2000 Y.Yasu 5
ATLAS-Japan DAQ Group DAQ2000, Lyon
Hardware configuration
v0.47 v0.47 v0.47 Network Driver version AceNIC(1MB) AceNIC(1MB) AceNIC(1MB) NIC egcs-2.91.66 egcs-2.91.66 egcs-2.91.66 gcc version 2.2.14-5 2.2.14-12 2.2.14-12 OS version 64-bit/32-bit 32-bit 32-bit PCI(bus width) 128 256 256 Memory size SDRAM SDRAM SDRAM Memory type 133MHz 100MHz 100MHz Bus speed 440BX SuperWorks ServerSet III LE 440GX Chipset 256KB 256KB 512KB Cache 800MHz 733MHz 500MHz CPU Freq. PentiumIII PentiumIII PentiumIII CPU type PC800 PC733 PC500 System
October 20, 2000 Y.Yasu 6 50 100 150 200 250 300 350 400 450 500 0.1 1 10 100 PentiumIII- 733MHz/SuperWorks PentiumIII-500MHz/440GX PentiumIII-800MHz/440BX
ATLAS-Japan DAQ Group DAQ2000, Lyon
Performance measurement of memory copy
Memory size in MB Memory-copy speed in MB/s by memcpy system call
October 20, 2000 Y.Yasu 7
ATLAS-Japan DAQ Group DAQ2000, Lyon
Hardware setup
PC500 PC500 PC500 PC500 Alteon Switch 180 PC733 PC800 PC733 PC800
For measurement of QoS performance and measurement of basic performance For measurement of basic performance
October 20, 2000 Y.Yasu 8
100 200 300 400 500 600 700 800 900 1000 0.01 1 100 10000 tx=20:rx=30 tx=200:rx=300 tx=400:rx=600 tx=2:rx=3 100 200 300 400 500 600 700 800 900 1000 0.01 1 100 10000 tx=400:rx=120 tx=996:rx=1000 tx=2000:rx=2000 tx=40:rx=12
ATLAS-Japan DAQ Group DAQ2000, Lyon
Tuning network driver
Message size in KB Transfer speed in Mbit/s Message size in KB Transfer speed in Mbit/s Tuned parameters of Network driver : /proc/sys/net/core/rmem_max 4000000 /proc/sys/net/core/wmem_max 4000000 TCP buffer size : 524288 bytes Nagle algorithm is enable. Normal frame : tx_coal=996 : rx_coal=1000 Jumbo frame : tx_coal=200 : rx_coal=300 Other parameters : default Used systems : PC500s Measuring tool : Netperf Normal frame Jumbo frame
October 20, 2000 Y.Yasu 9
ATLAS-Japan DAQ Group DAQ2000, Lyon
Transfer speed in normal frame Transfer speed in jumbo frame Bus width dependency(64-bit PCI vs. 32-bit PCI)
Performance measurement of Gigabit Ethernet
October 20, 2000 Y.Yasu 10
100 200 300 400 500 600 700 800 900 1000 0.01 0.1 1 10 100 1000 Pentium500MHz Pentium733MHz Pentium800MHz
ATLAS-Japan DAQ Group DAQ2000, Lyon
Transfer speed in normal frame
Transfer speed does not depend on CPU speed in normal frame
Message size in KB Transfer speed in Mbit/s
October 20, 2000 Y.Yasu 11
100 200 300 400 500 600 700 800 900 1000 0.01 1 100 10000 Pentium500MHz Pentium733MHz Pentium800MHz
ATLAS-Japan DAQ Group DAQ2000, Lyon
Transfer speed in jumbo frame
Transfer speed depends on Chipset performance in jumbo frame.
Message size in KB Transfer speed in Mbit/s
October 20, 2000 Y.Yasu 12
100 200 300 400 500 600 700 800 900 1000 0.01 1 100 10000 32-bit PCI(mtu1500) 64-bit PCI(mtu1500) 32-bit PCI(mtu9000) 64-bit PCI(mtu9000)
ATLAS-Japan DAQ Group DAQ2000, Lyon
Bus width dependency(64-bit PCI vs. 32-bit PCI)
Message size in KB Transfer speed in Mbit/s
Bus width does not affect transfer speed in normal frame. 64-bit PCI improves transfer speed in jumbo frame, up to 990Mbit/s.
On PC733s
October 20, 2000 Y.Yasu 13
ATLAS-Japan DAQ Group DAQ2000, Lyon
QoS performance on Gigabit Ethernet 1x1 system 3x1 system 1x3 system QoS in Linux Class-Based Queueing Token Bucket Filter Traffic control in TCP/IP Queueing discipline for Linux
October 20, 2000 Y.Yasu 14
ATLAS-Japan DAQ Group DAQ2000, Lyon
CBQ can classify incoming packets into multiple classes. CBQ can share and limit the transfer rate.
Class-Based Queueing (CBQ)
estimator Set overlimit classifier Class 1 Class 2 Class 3 packet Packet scheduler
October 20, 2000 Y.Yasu 15
ATLAS-Japan DAQ Group DAQ2000, Lyon
TBF can limit the transfer rate according to the token rate.
Token Bucket Filter (TBF)
Data packet Data Buffer Token buffer rate Bucket size
October 20, 2000 Y.Yasu 16
ATLAS-Japan DAQ Group DAQ2000, Lyon
QoS in Linux
QoS in Linux kernel 2.2.x and Iproute2 for Linux supplies the following functions to control IP packets; For IP packets of inbound and outbound Classifying Prioritizing Sharing Limiting QoS in Linux kernel 2.2.x : in kernel standard distribution Iproute2 : ftp://ftp.inr.ac.ru/ip-routing/iproute2-current.tar.gz Iproute2 supports a typical command called tc. Can we use the QoS such as Class-Based Queueing and Token Bucket Filter for shaping traffic on Event Builder?
October 20, 2000 Y.Yasu 17
ATLAS-Japan DAQ Group DAQ2000, Lyon
Traffic control in TCP/IP
Input de-multiplexing Forwarding Output queueing Upper layers(TCP,UDP,...) Traffic Control
Traffic control is done only at output queueing.
October 20, 2000 Y.Yasu 18
ATLAS-Japan DAQ Group DAQ2000, Lyon
Queueing discipline for Linux
Ῡ Qdisc ̰ CBQ
1: bandwidth:1000Mbit Ῡ Class ̰ CBQ 1:1 rate:1000Mbit
Qdisc (TBF) Rate:100Mbit
Classifier u32: Specified IP Address
U32 is a classifier. TBF limits the transfer rate to 100Mbit/s at this example. The configuration was used at 1x1 system for the evaluation.
October 20, 2000 Y.Yasu 19
100 200 300 400 500 600 100 200 300 400 500 600 Ideal rate Measured rate
ATLAS-Japan DAQ Group DAQ2000, Lyon
QoS performance on 1x1 system
Assigned rate in Mbit/s Measured rate in Mbit/s No QoS
Bandwidth assignment worked well. The overhead of QoS was very small and nearly 1 %.
On PC500s
October 20, 2000 Y.Yasu 20
50 100 150 200 250 50 100 150 200 250 Ideal rate Measured rate on node1 Measured rate on node2 Measured rate on node3
ATLAS-Japan DAQ Group DAQ2000, Lyon
QoS performance on 3x1 system
Assigned rate in Mbit/s Measured rate in Mbit/s
Bandwidth assignment worked well
No QoS
On PC500s
October 20, 2000 Y.Yasu 21
50 100 150 200 250 50 100 150 200 250 Ideal rate Measured rate on node1 Measured rate on node2 Measured rate on node3
ATLAS-Japan DAQ Group DAQ2000, Lyon
QoS performance on 1x3 system
Assigned rate in Mbit/s Measured rate in Mbit/s
Bandwidth assignment worked well at small rate, but did not work at full rate. Packet sharing performance without QoS was better than that with QoS.
No QoS On PC500s
October 20, 2000 Y.Yasu 22