Impact of undesirable HTTP traffic on electrical power

Impact of undesirable HTTP traffic on electrical power consumption in the ICT rooms Ivanovic - BUCC/AMRES EUNIS 2012 Vila Real June 22, 2012 connect • communicate • collaborate

Power consumption monitoring Under the influence of Green Technology gathering information about ICT rooms power consumption is becoming popular topic. Monitoring of power consumption is often disregarded? ! Popular monitoring applications are adapted for monitoring of commonly used variables (traffic utilization, port status, services state, …) but not power consumption. Most datacenters are often designed with good electric network power backup (power reserve, UPS, fuel generator). Manufacturers of power supply devices are offering their own software for monitoring. Why can’t we integrate it with custom NMS software we already have? ! Control power consumption and reduce it. Can we have information about power consumption of specific service? ! connect • communicate • collaborate 1/10

AMRES topology connect • communicate • collaborate

Testing environment AMRES policy for web traffic Iron. Ports Devices Six Ironport Web security appliance (S 670) Ironport Management appliance (M 160) Cisco Catalyst 2960 & ASR 1002 Fujitsu-Siemens servers (4 x RX 200 S 5 & 1 x RX 300 S 5) 2 x APC Smart 5000 2 x APC Metered PDU AP 7852 UPS and PDU support SNMP Almost all devices have dual supplies connect • communicate • collaborate 2/10

System monitoring SNMP protocol Net. Flow protocol Ironport Log files connect • communicate • collaborate

SNMP variables APC UPS 5000 VA SNMP variables: –. 1. 3. 6. 1. 4. 1. 318. 1. 1. 1. 4. 2. 3 - The current UPS load expressed in percent of rated capacity. –. 1. 3. 6. 1. 4. 1. 318. 1. 1. 1. 4. 2. 4 - The current in amperes drawn by the load on the UPS. APC PDU SNMP variables: –. 1. 3. 6. 1. 4. 1. 318. 1. 1. 12. 2. 3. 1. 1. 2 - Getting this OID will return the phase/bank load measured in tenths of Amps. –. 1. 3. 6. 1. 4. 1. 318. 1. 1. 12. 1. 16 - Getting this OID will return the Power in Watts. connect • communicate • collaborate 3/10

SNMP variables and accuracy Basic “mrtg-rrd” approach (90 sec sampling) Flat graphs (constant values) Not enough information connect • communicate • collaborate 4/10

SNMP variables and accuracy Linux script - 5 second sampling – Output current ∆Iups=1 A, ∆Ipdu=100 m. A connect • communicate • collaborate 5/10

SNMP variables and accuracy Linux script - 5 second sampling – Output load ∆Wups=40 W, ∆ Wpdu=23 W connect • communicate • collaborate 6/10

Network Traffic and power consumption relation connect • communicate • collaborate

Network Traffic and power consumption relation Linux script #!/bin/bash while true; do c=(`snmpget -v 2 c -c community 147. 91. 7. 1. 1. 3. 6. 1. 2. 1. 31. 1. 6. 23. 1. 3. 6. 1. 2. 1. 31. 1. 10. 23 -Ov. Q`); sleep 10 d=(`snmpget -v 2 c -c community 147. 91. 7. 1. 1. 3. 6. 1. 2. 1. 31. 1. 6. 23. 1. 3. 6. 1. 2. 1. 31. 1. 10. 23 -Ov. Q`); a=`snmpget -v 2 c -c community 192. 168. 99. 16. 1. 3. 6. 1. 4. 1. 318. 1. 1. 12. 1. 16. 0 -Ov. Q`; b=`snmpget -v 2 c -c community 192. 168. 99. 17. 1. 3. 6. 1. 4. 1. 318. 1. 1. 12. 1. 16. 0 -Ov. Q`; in=`expr ${d[0]} - ${c[0]}` out=`expr ${d[1]} - ${c[1]}` echo `date +"%m-%d-%y/%T"` $a $b $in $out connect • communicate • collaborate

Network Traffic and power consumption relation High correlation (r~0. 8) Linear relation Power consumption function y=0. 2407∙x [W/Mbps]+ 2884. 7[W] connect • communicate • collaborate

Network Traffic and power consumption relation Linear relation y=0. 2407∙x + 2884. 7 Initial power consumption W 0=2884. 7 W (y-intercept) Traffic dependency K=0. 2047[W/Mbps] (slope) Network traffic consume about ~11% of total power consumption connect • communicate • collaborate

Prediction of power consumption Error for predicted power consumption expressed in KWh is ~ 0. 47% connect • communicate • collaborate

Prediction of power consumption connect • communicate • collaborate

Impact of undesirable traffic on electrical power consumption Netw. Flow information Separate traffic by services, src&dst AS, src&dst IP…. . connect • communicate • collaborate

Impact of undesirable traffic on electrical power consumption u. Torrent plugin (propagation of 127. 0. 0. 1 requests through the network) Source - Ironport proxy (squid) logs connect • communicate • collaborate

SNMP queries time interval How to choose sampling period (KWh calculation)? Increase time between SNMP queries Reduce load on NMS system connect • communicate • collaborate 7/10

Monitor and reduce power consumption! Can we use traffic results in order to calculate our electric bills? It is easier to collect information about traffic! Reduce the sampling frequency in order to lower the burden on your NMS system. Manufacturers should agree on universal MIB database. (. 1. 3. 6. 1. 2. 1. XY) ($ ) Implementation of SNMP variable that has “memory capability” – SNMP Counters. ($ ) Implementation of good security is necessary Stop the undesirable traffic at the source ($ ) By new equipment ($ ) Be smart, be green! connect • communicate • collaborate 10/10

Campus Best Practices (NA 3/T 4) Addresses key challenges for European campus networks Participating NRENs organise working groups within their country Working groups discuss and agree on common campus best-practices Best practice documents are translated to English and published Results are presented at conferences European level workshops on areas of focus are organised Download ~44 documents from Areas of focus Physical Infrastructure Campus Networking Real-time communications Network Monitoring Wireless Security www. geant. net/About_GEANT/Campus_Best_Practice/ Participating NRENs 20 connect • communicate • collaborate

Q&A Thank you! ivanovic@rcub. bg. ac. rs connect • communicate • collaborate