[B! nic] yassã®ãƒ–ãƒƒã‚¯ãƒžãƒ¼ã‚¯

yass id:yass

nicã«é–¢ã™ã‚‹yassã®ãƒ–ãƒƒã‚¯ãƒžãƒ¼ã‚¯ (134)

${{author_name}}$

{{{comment_expanded}}}

{{label}}

{{#is_bookmark}}ãƒªã‚¹ãƒˆ{{/is_bookmark}}{{^is_bookmark}}ãƒªãƒ³ã‚¯{{/is_bookmark}}

${{author_name}}$
{{author_name}}{{created}}
{{ #comment }}{{ comment }}{{ /comment }}
- {{ label }}

${{author_name}}$

{{{comment_expanded}}}

{{label}}

{{#is_bookmark}}ãƒªã‚¹ãƒˆ{{/is_bookmark}}{{^is_bookmark}}ãƒªãƒ³ã‚¯{{/is_bookmark}}

ãƒ‘ã‚±ãƒƒãƒˆå‡¦ç†ã®ãƒžãƒ«ãƒã‚³ã‚¢ã‚¹ã‚±ãƒ¼ãƒ«ã«ã¤ã„ã¦ - Qiita
æ¦‚è¦ ã“ã®è³‡æ–™ã‚’è¦‹ã¦ã€ãƒ‘ã‚±ãƒƒãƒˆå‡¦ç†ã®å‰²ã‚Šè¾¼ã¿ã¨ãƒžãƒ«ãƒã‚³ã‚¢ã«ã‚ˆã‚‹ã‚¹ã‚±ãƒ¼ãƒ«ãŒæ°—ã«ãªã£ãŸã®ã§è‡ªåˆ†ã§ã‚‚èª¿ã¹ã¦ã¿ãŸã“ã¨ã‚’ã¾ã¨ã‚ã‚‹. ã¯ã¦ãªã«ãŠã‘ã‚‹Linuxãƒãƒƒãƒˆãƒ¯ãƒ¼ã‚¯ã‚¹ã‚¿ãƒƒã‚¯ãƒ‘ãƒ•ã‚©ãƒ¼ãƒžãƒ³ã‚¹æ”¹å–„ / Linux network performance improvement at hatena ã¾ãšä¸Šè¨˜è³‡æ–™ã‚’ï¼“è¡Œã§ã¾ã¨ã‚ã‚‹ã¨ å—ä¿¡ãƒ‘ã‚±ãƒƒãƒˆã¯ãƒãƒ¼ãƒ‰å‰²ã‚Šè¾¼ã¿ã¨ã‚½ãƒ•ãƒˆå‰²ã‚Šè¾¼ã¿ã‚’ä½¿ã£ã¦CPUã«å‡¦ç†ã•ã‚Œã‚‹ å—ä¿¡ãƒ‘ã‚±ãƒƒãƒˆã®å‰²ã‚Šè¾¼ã¿ãŒç‰¹å®šã®CPUã«ã—ã‹ã„ã‹ãšãƒžãƒ«ãƒã‚³ã‚¢ã‚¹ã‚±ãƒ¼ãƒ«ã—ãªã„å•é¡Œã«éé‡ RPS (Receive Packet Steering)ã‚’ä½¿ç”¨ã—ã¦ãƒžãƒ«ãƒã‚³ã‚¢ã§å—ä¿¡ãƒ‘ã‚±ãƒƒãƒˆã‚’å‡¦ç†ã—ãƒãƒ¼ãƒ‰ãƒãƒ©ãƒ³ã‚µãƒ¼ã®ãƒãƒƒãƒˆãƒ¯ãƒ¼ã‚¯å‡¦ç†ã®æ€§èƒ½ã‚’å‘ä¸Šã•ã›ãŸ è³‡æ–™ã‚’è¦‹ã¦ã„ã¦ä»¥ä¸‹ã®ç‚¹ãŒæ°—ã«ãªã£ãŸã®ã§ã€ã“ã“ã‚‰è¾ºã‚’ã‚¹ãƒƒã‚ãƒªã•ã›ã‚‹ãŸã‚ã«ã„ãã¤ã‹ã®è³‡æ–™ã‚’èªã¿ã€ã¾ã¨ã‚ã¦ãŠã å‰²ã‚Šè¾¼ã¿ãŒç‰¹å®šã®CPUã«ã—ã‹ã„ã‹ãªã‹ã£ãŸç†ç”± RPS(Receiv
yass 2016/07/16
nic

rss

RPS

interrupt
ãƒªãƒ³ã‚¯
Intent-based retargeting | Datacratic
Datacratic's intent-based retargeting techno logyÂ is now part of iperceptions Get expert guidance in measuring and optimizing the effectiveness of your marketing campaigns with iperceptions Â©2017 Datacratic. All Rights Reserved. All other trademarks, logos and brand names on this website are the properties of their respective owners.
yass 2015/08/30
nginx

ec2

nic

igb

interrupt

rps

interruptThrottleRate

pps

10GbE

sr-iov
ãƒªãƒ³ã‚¯
How to achieve low latency with 10Gbps Ethernet
Good morning! In a recent blog post we explained how to tweak a simple UDP application to maximize throughput. This time we are going to optimize our UDP application for latency. Fighting with latency is a great excuse to discuss modern features of multiqueue NICs. Some of the techniques covered here are also discussed in the scaling.txt kernel document. CC BY-SA 2.0 image by Xiaojun Deng Our expe
yass 2015/07/01
10GbE

nic

latency

toread

rps

rfs
ãƒªãƒ³ã‚¯
How to receive a million packets per second
Last week during a casual conversation I overheard a colleague saying: "The Linux network stack is slow! You can't expect it to do more than 50 thousand packets per second per core!" That got me thinking. While I agree that 50kpps per core is probably the limit for any practical application, what is the Linux networking stack capable of? Let's rephrase that to make it more fun: On Linux, how hard
yass 2015/06/20
linux

pps

numa

socket

nic

packet
ãƒªãƒ³ã‚¯
é«˜ã‚¹ãƒ«ãƒ¼ãƒ—ãƒƒãƒˆãªã‚µãƒ¼ãƒã‚¢ãƒ—ãƒªã‚±ãƒ¼ã‚·ãƒ§ãƒ³ã®ç‚ºã®æ–°ã—ã„ãƒ•ãƒ¬ãƒ¼ãƒ ãƒ¯ãƒ¼ã‚¯â€¨ã€ŒSeastarã€
4. C++14? â€¢ ãƒ•ãƒ¬ãƒ¼ãƒ ãƒ¯ãƒ¼ã‚¯å…¨ä½“ã§C++14ã‚’ä½¿ç”¨ ï¼ˆg++ -std=gnu++1yã€clang not testedï¼‰ â€¢ C++(11Â¦14)ã¯Cloudius Systemsã®æ¨™æº–ä½¿ç”¨è¨€ èªžãªã®ã§â€¦ ï¼ˆOSvã¯C++11ï¼‰ â€¢ ä»–è¨€èªžã«ã¯bindingã§å¯¾å¿œã™ã‚‹äº‹ã‚’æ¤œè¨Ž
yass 2015/02/20
network

nic

dpdk
ãƒªãƒ³ã‚¯
RHEL6ç³»ã§é€šä¿¡ã‚ªãƒ•ãƒãƒ¼ãƒ‰è¨å®šã®æœ‰åŠ¹/ç„¡åŠ¹ã«ã‚ˆã‚‹è»¢é€é€Ÿåº¦ã®å·®ç•° - @uzyexe ã®ãƒŽãƒ¼ãƒˆ
ã»ã¼ãƒ‡ãƒ•ã‚©ãƒ«ãƒˆè¨å®šçŠ¶æ…‹ã® RHEL6 ç³» OS ã§é€šä¿¡ã‚ªãƒ•ãƒãƒ¼ãƒ‰è¨å®šã®æœ‰åŠ¹/ç„¡åŠ¹ã«ã‚ˆã£ã¦ã€é€šä¿¡ã®è»¢é€é€Ÿåº¦ãŒã©ã‚Œã»ã©å·®ç•°ãŒã‚ã‚‹ã‹æ¤œè¨¼ã—ã¦ã¿ã¾ã—ãŸã€‚æ¤œè¨¼ç’°å¢ƒã®æ¦‚è¦ã¯ä¸‹è¨˜ã®ã¨ãŠã‚Šã§ã™ã€‚ ã•ãã‚‰ã®ã‚¯ãƒ©ã‚¦ãƒ‰ã§ä»®æƒ³ã‚µãƒ¼ãƒã‚’2å°ç”¨æ„ã—ã¦ã€ã‚µãƒ¼ãƒé–“ã‚’ã‚¹ã‚¤ãƒƒãƒã§æŽ¥ç¶šã—ã¦é€Ÿåº¦æ¸¬å®šã€‚ æ¸¬å®šå¯¾è±¡ã‚µãƒ¼ãƒã«ã¯ Sientific Linux 6.5 ã‚’ã‚¤ãƒ³ã‚¹ãƒˆãƒ¼ãƒ«ã€‚ å¯¾å‘ã‚µãƒ¼ãƒã«ã¯ Ubuntu 14.04.0 ã‚’ã‚¤ãƒ³ã‚¹ãƒˆãƒ¼ãƒ«ã€‚ å„ã‚µãƒ¼ãƒã® NIC ã®ç†è«–å›žç·šé€Ÿåº¦ã¯å…¬ç§° 1000Mbpsã€‚ï¼ˆæœ€å¤§å®ŸåŠ¹é€Ÿåº¦ã¯500Mbps ç¨‹åº¦ï¼‰ ã‚µãƒ¼ãƒé–“ã‚’çµã¶ã‚¹ã‚¤ãƒƒãƒã®å›žç·šé€Ÿåº¦ã¯ç„¡åˆ¶é™ã€‚ï¼ˆæ¦‚ã NIC ã®å®ŸåŠ¹é€Ÿåº¦ã«ä¾å˜ã™ã‚‹ã“ã¨ã«ãªã‚‹ï¼‰ é€Ÿåº¦æ¸¬å®šã«ã¯ iperf ã‚’ä½¿ç”¨ã€‚ ä»Šå›žã€å¤‰æ›´å¯¾è±¡ã¨ãªã‚‹é€šä¿¡ã‚ªãƒ•ãƒãƒ¼ãƒ‰è¨å®šã¯ã€ä»®æƒ³NIC (Network Interface Card) ã®Â TSO (TCP Segmentation Offload
yass 2015/02/17
offload

nic
ãƒªãƒ³ã‚¯
40 Gigabit Ethernetå¯¾å¿œã®Intelè£½LANã‚«ãƒ¼ãƒ‰ãŒè²©å£²ä¸ ç´„9.4ä¸‡å††ã‹ã‚‰
yass 2014/10/22
intel

nic

infiniband
ãƒªãƒ³ã‚¯
ã‚³ãƒ©ãƒ - ã‚¯ãƒ©ã‚¦ãƒ‰æ™‚ä»£ã®ã‚ªãƒ¼ãƒ—ãƒ³ã‚½ãƒ¼ã‚¹å®Ÿè·µæ´»ç”¨ | ç¬¬53å›žã€€NFVã¨Intel DPDKï¼ˆå‰ç·¨ï¼‰ï½œCTCæ•™è‚²ã‚µãƒ¼ãƒ“ã‚¹ ç ”ä¿®/ãƒˆãƒ¬ãƒ¼ãƒ‹ãƒ³ã‚°
[ITç ”ä¿®]æ³¨ç›®ã‚ãƒ¼ãƒ¯ãƒ¼ãƒ‰ Python UiPath(RPA) æœ€æ–°æŠ€è¡“å‹•å‘ Microsoft Azure Docker Kubernetes ç¬¬53å›žã€€NFVã¨Intel DPDKï¼ˆå‰ç·¨ï¼‰ (ä¸äº•æ‚¦å¸) 2014å¹´10æœˆ ã¯ã˜ã‚ã« å…ˆé€±ã€Red Hat Forum 2014/Tokyoã§æ—¥æœ¬ã®ã‚ªãƒ•ã‚£ã‚¹ã«æ¥ã¦ã„ãŸChris Wrightã¨ä¼šè©±ã‚’ã™ã‚‹æ©Ÿä¼šãŒã‚ã‚Šã€Intel DPDKã«ã¤ã„ã¦ã€ã„ã‚ã„ã‚æ•™ãˆã¦ã‚‚ã‚‰ã„ã¾ã—ãŸã€‚Chrisã¯ã€ä»¥å‰ã¯ã€Linuxã‚«ãƒ¼ãƒãƒ«ã‚„ä»®æƒ³åŒ–ï¼ˆXen/KVMï¼‰ã®é–‹ç™ºã«æ·±ãé–¢ã‚ã£ã¦ã„ã¾ã—ãŸãŒã€ç¾åœ¨ã¯ã€ã€ŒTechnical Director of SDN at Red Hatã€ã¨ã„ã†è‚©æ›¸ã§ã€SDNé–¢ä¿‚ã®ä»•äº‹ã‚’ã—ã¦ã„ã¾ã™ã€‚OpenDaylightã®ãƒœãƒ¼ãƒ‰ãƒ¡ãƒ³ãƒãƒ¼ã‚„OpenStackã®NFVã‚µãƒ–ãƒãƒ¼ãƒ ã«ã‚‚å‚åŠ ã—ã¦ã„ã¾ã™ã€‚ Intel DPDKã®è©±é¡Œã«ãªã£ãŸã®ã¯ã€ã¾ã•ã«ã“ã®
yass 2014/10/21
" Intel DPDKã‚’åˆ©ç”¨ã™ã‚‹ã‚¢ãƒ—ãƒªã‚±ãƒ¼ã‚·ãƒ§ãƒ³ã¯ã€Linuxã‚«ãƒ¼ãƒãƒ«ã‚’ä»‹ã•ãšã«ã€ãƒãƒƒãƒˆãƒ¯ãƒ¼ã‚¯ã‚«ãƒ¼ãƒ‰ã«å±Šã„ãŸãƒ‘ã‚±ãƒƒãƒˆã‚’ç›´æŽ¥ã«å—ã‘å–ã£ã¦å‡¦ç† "

intel

dpdk

nic

linux
ãƒªãƒ³ã‚¯
ç¾Žå¥³ç›´æ’é—´å›´è§‚è§†é¢‘-
ç¾Žå¥³ç›´æ’é—´å›´è§‚è§†é¢‘ä»‹ç»ï¼š â€œè¿™å€’æ²¡æœ‰â€¦â€¦â€çŽ‹æ™“å³°æ‘‡å¤´ã€‚ â€œå¾è€å…è®¸æˆ‘åœ¨è¿™é‡Œè¯´ï¼Ÿâ€ â€œäº”æžšâ€¦â€¦ä¸¤äº¿ï¼Ÿâ€ç›”ç”²é’å¹´çœ¼ç›ä¸€ä¸‹çžªåœ†ï¼šâ€œä½ å–çš„æ˜¯å…å“ä¸¹è¯ï¼Ÿæ€Žä¹ˆä¸åŽ»æŠ¢å‘¢ï¼Ÿâ€â€œæˆ‘ä»¬èƒ½å¤Ÿä¿®ç‚¼â€¦â€¦è€Œä¸”ï¼Œè¿›æ¥å¾ˆå¿«â€¦â€¦â€ ç¾Žå¥³ç›´æ’é—´å›´è§‚è§†é¢‘è¯„æµ‹ï¼š â€œæ˜¯ç¦ä¸æ˜¯ç¥¸ï¼Œæ˜¯ç¥¸èº²ä¸è¿‡â€¦â€¦â€ â€œæ²ˆå“²â€¦â€¦å²‚ä¸å¾ˆå±é™©ï¼â€â€œä¸€ã€ä¸€ä¸‡ä¸¤ï¼Ÿâ€å·®ç‚¹æ²¡æ‹¿ä½æ‰‹ä¸çš„çº¸ï¼Œé±¼è€ä¸€å£°å¹æ¯â€œå°‘çˆ·è¿™æ˜¯è¢«äººéª—äº†ï¼â€ â€œæ˜¯å•Šï¼Œå‰§çƒˆçš„çˆ†ç‚¸ï¼Œå¯ä»¥è®©ç©ºä¸å‡å›ºçš„å…ƒç´ ç²’åï¼Œåˆ†æ•£å¼€æ¥ï¼Œå˜å¾—æ´»è·ƒï¼Œä¸è¿‡ï¼Œè¿™ç§æ´»è·ƒåº¦ï¼Œä¸ä¼šæŒç»å¤šä¹…ï¼Œæœ€å¤šä¸€ä¸ªæ—¶è¾°ï¼Œå°±ä¼šé‡æ–°å‡èšï¼â€ â€œè¿™â€¦â€¦â€çœ‰æ¯›ä¸€è·³ï¼Œæ²ˆä»Žå¿ƒå·®ç‚¹æ²¡æ™•è¿‡åŽ»ã€‚ â€œè§ä¹å„¿â€¦â€¦é‚£æ˜¯æˆ‘çš„ä½ç½®ï¼Œæœ‰ç‚¹â€¦â€¦æœ‰ç‚¹å‘³ï¼Œè¦ä¸ï¼Œæˆ‘è·Ÿä½ æ¢è¿‡æ¥â€¦â€¦â€â€œå·²ç»å¼€å§‹äº†â€¦â€¦â€é’ŸçŽ‰æ¥¼æ‘†äº†æ‘†æ‰‹ã€‚â€œæˆ‘å‡†å¤‡å¥½äº†ææ–™ï¼Œåªæ±‚å¾å®—å¸ˆå‡ºæ‰‹ï¼Œç‚¼åˆ¶å‡ºå…µå™¨ï¼Œæ„¿æ„æ”¯ä»˜å·¨é¢é…¬åŠ³ï¼â€ â€œæ²ˆå®¶å®¶ä¸»å¯åœ¨ï¼Ÿè¿™å°±æ˜¯ä½ ä»¬çš„å¾…å®¢ä¹‹é“ï¼Ÿå¦‚æžœçœŸæ˜¯å¦‚æ¤ï¼Œåœ¨ä¸‹å°±å‘Šè¾žäº†â€¦â€¦â€ ç¾Žå¥³ç›´æ’é—´å›´è§‚è§†é¢‘æ€»ç»“ï¼š â€œäº‘éœ„å¯¸æ‹³
yass 2014/09/18
" NAPIã¯ãƒ‰ãƒ©ã‚¤ãƒãŒå®šæœŸçš„ã«ãƒãƒ¼ãƒªãƒ³ã‚°ã—ã¦ãƒãƒƒãƒ•ã‚¡ã®ä¸èº«ã‚’ã¨ã‚Šã«ã„ããŸã‚ã€ãƒãƒ¼ãƒªãƒ³ã‚°é »åº¦ã‚’ä¸Šã’ã‚‹ã‹ã€ä¸€åº¦ã§å‡¦ç†ã™ã‚‹é‡ã‚’å¢—ã‚„ã—ã¦ã‚ã’ã‚Œã°ãƒãƒƒãƒ•ã‚¡ã«ç©ºããŒã§ãã‚‹ã¯ãšã§ã™ã€‚ ãŸã ã—CPUã®è² è·ã¯ä¸ŠãŒã‚Šã¾ã™ "

nic

i350

network

napi
ãƒªãƒ³ã‚¯
Improving Performance
Many Unix-like operating systems and distributions are tuned for desktop or light use out of the box and not for a production database. This guide describes recommended system performance tunings for operators of new and existing Riak clusters. The tunings present in this guide should be considered as a starting point. It is important to make note of what changes are made and when in order to meas
yass 2014/08/25
riak

kernel

filesystem

nic

performance

nr_requests

tcp
ãƒªãƒ³ã‚¯
10GbE Plugfest 2014 å‚è€ƒè³‡æ–™
2. 40/100GbEã®ç‰©ç†é™ç•Œã‚’çŸ¥ã‚ã† 40GbE PCI Express Gen2 NIC* PCIe Gen 2.0 x8 interface, 32Gbps 8b/12b line encoding, 20% overhead, 25.6Gbps 40GbE PCI Express Gen 3 NIC* PCIe Gen 3.0 x8 interface, 1GB per lane or 64Gbps 64/66 line encoding, 3% overhead, 62.0Gbps 100GbE PCI Express Gen 3 NIC?** PCIe Gen 3.0 x16 interface, 128Gbps 64/66 line encoding, 3% overhead, 124.1Gbps *SOURCE: Evaluation of 40 Gigabit Eth
yass 2014/07/28
10GbE

pps

nic

intel
ãƒªãƒ³ã‚¯
MIIç›£è¦–ã‹ã‚‰ARPç›£è¦–ã«ã‚ˆã‚‹ãƒœãƒ³ãƒ‡ã‚£ãƒ³ã‚°åˆ‡ã‚Šæ›¿ãˆã¸ã®å¤‰æ›´æ–¹æ³•
ã‚µãƒ¼ãƒã®ãƒœãƒ³ãƒ‡ã‚£ãƒ³ã‚°(bonding)ã®åˆ‡ã‚Šæ›¿ãˆãŒèµ·ããšã€ ãƒˆãƒ©ãƒ•ã‚£ãƒƒã‚¯ã‚’å‡¦ç†ã§ããªã„äº‹è±¡ãŒç™ºç”Ÿã—ãŸã€‚ å•é¡Œã¸ã®å¯¾å¿œã‚’æ¤œè¨Žã™ã‚‹ã€‚ â—† éšœå®³å†…å®¹ ã‚µãƒ¼ãƒå´ã§ã€ãƒœãƒ³ãƒ‡ã‚£ãƒ³ã‚°ã®Active-Backup ãƒ¢ãƒ¼ãƒ‰(mode=1)ã‚’åˆ©ç”¨ã—ã€ éšœå®³ã®æ¤œçŸ¥æ–¹æ³•ã¨ã—ã¦MIIç›£è¦–ã‚’ä½¿ã£ã¦ã„ãŸã€‚ ã—ã‹ã—ã€L2ã‚¹ã‚¤ãƒƒãƒãŒãƒªãƒ³ã‚¯ãŒã‚¢ãƒƒãƒ—ã—ã¦ã„ã‚‹çŠ¶æ…‹ã§æ•…éšœã—ãŸãŸã‚ã€ ã‚µãƒ¼ãƒå´ã§ã¯ç•°å¸¸ã‚’æ¤œçŸ¥ã§ããªã‹ã£ãŸã€‚ ãã®çµæžœã‚µãƒ¼ãƒå´ã§ã®çµŒè·¯åˆ‡ã‚Šæ›¿ãˆãŒèµ·ããšã€ãƒˆãƒ©ãƒ•ã‚£ãƒƒã‚¯ã‚’å‡¦ç†ã§ããªããªã£ãŸã€‚ L2SW Â Â L2SW |Â Â â‡’ Ã—Â |ã€€ã€€å·¦çµŒè·¯ã‹ã‚‰å³çµŒè·¯ã«bondingåˆ‡ã‚Šæ›¿ãˆãŒèµ·ããªã„ SERVER â—† ãƒœãƒ³ãƒ‡ã‚£ãƒ³ã‚°åˆ‡ã‚Šæ›¿ãˆã®ç›£è¦–æ–¹æ³•ã®å¤‰æ›´çµµæ¤œè¨Ž ãƒœãƒ³ãƒ‡ã‚£ãƒ³ã‚°ã®Active-Backup ãƒ¢ãƒ¼ãƒ‰ (mode=1) ã§ã¯éšœå®³ã®ç›£è¦–æ–¹æ³•ã¨ã—ã¦ MIIç›£è¦–ã¨ARPç›£è¦–ã®ï¼’é€šã‚Šã®è¨å®šãŒã‚ã‚‹ã€‚ ãƒœãƒ³ãƒ‡ã‚£ãƒ³ã‚°ã‚’åˆ‡ã‚Šæ›¿ãˆã‚‹ãŸã‚ã®ãƒˆãƒªã‚¬
yass 2014/07/13
bonding

arp

nic

network

linux
ãƒªãƒ³ã‚¯
Neutron Network Nodeã®å¤–å´NICã§ã¯groã‚’offã«ã™ã‚‹ - TIMEZONE PST Hatena Branch
ä»®æƒ³ãƒžã‚·ãƒ³ã‚¤ãƒ³ã‚¹ã‚¿ãƒ³ã‚¹ã¨å¤–éƒ¨ã®é€šä¿¡(ftpå—ä¿¡)ã§ã‚¹ãƒ«ãƒ¼ãƒ—ãƒƒãƒˆãŒä½Žã„ é€šä¿¡çµŒè·¯ instance-A ---> hypervisor ---> network node ---> router -(internet)-> ftp.iij.ad.jp ã‚¹ãƒ«ãƒ¼ãƒ—ãƒƒãƒˆãŒã‚„ãŸã‚‰ã¨ä½Žã„ ãƒ†ã‚¹ãƒˆã§ä»®æƒ³ãƒžã‚·ãƒ³ã‚¤ãƒ³ã‚¹ã‚¿ãƒ³ã‚¹(instance-A)ã‹ã‚‰ftp.iij.ad.jpã‹ã‚‰CentOSã®LiveCDã‚¤ãƒ¡ãƒ¼ã‚¸ã‚’ãƒ€ã‚¦ãƒ³ãƒãƒ¼ãƒ‰ã—ã¦ã¿ã‚‹ã¨ã€ä»¥ä¸‹ã®é€šã‚Šå¹³å‡400kB/sç¨‹åº¦ã—ã‹ã‚¹ãƒ«ãƒ¼ãƒ—ãƒƒãƒˆãŒå‡ºãªã„ CentOS-6.5-x86_64-LiveCD.iso: ETA: 21:34 116.51/649.00 MB 421.35 kB/s network nodeã‹ã‚‰ç›´æŽ¥åŒã˜æ“ä½œã‚’å®Ÿè¡Œã—ã¦ã¿ã‚‹ã¨...6MB/sä»¥ä¸Šå‡ºã¦ã„ã‚‹ CentOS-6.5-x86_64-LiveCD.iso: ETA: 0:50 327.9
yass 2014/07/04
nic

offload

vitualization
ãƒªãƒ³ã‚¯
Linux packet-forwarding
2. ã¯ã˜ã‚ã¾ã—ã¦ï¼Ÿ â€¢ æ–°æ½ŸçœŒä¸‰æ¡å¸‚ã«ã‚ã‚‹åœ°åŸŸ ISP æœ‰é™ä¼šç¤¾éŠ€åº§å ‚ã§ã‚µãƒ¼ãƒç®¡ç†ã‚„ãƒãƒƒãƒˆ ãƒ¯ãƒ¼ã‚¯ç®¡ç†ã‚’ã—ã¦ã„ã¾ã™ â€¢ å¼Šç¤¾ã§ã¯ä¸Šæµ ISP(OCN ã•ã‚“ã¨æ–°æ½ŸçœŒå†…ã® ISP) ã‚„ Echigo-IX(æ–°æ½ŸçœŒ å†…ã®åœ°åŸŸ IX) ã¨ã®æŽ¥ç¶šã« Linux(Vyatta Core) ã‚’ä½¿ã£ã¦ BGP æŽ¥ç¶šã‚’ ã—ã¦ã„ã¾ã™ â€¢ 2010 å¹´é ƒã‹ã‚‰ Linux ã‚«ãƒ¼ãƒãƒ«ã®ãƒ‘ã‚±ãƒƒãƒˆè»¢é€å‡¦ç†ã«ã¤ã„ã¦èª¿ã¹ InternetWeek ã§ç™ºè¡¨ã—ãŸã‚Šã—ã¦ã„ã¾ã—ãŸ â€¢ è¶£å‘³ã§ã¯ IETF ã§æ¨™æº–åŒ–ãŒé€²ã‚ã‚‰ã‚Œã¦ã„ã‚‹ MAP(Mapping of Address and Port with Encapsulation) ã¨ã„ã†ãƒ—ãƒãƒˆã‚³ãƒ«ã‚’ Linux ã‚«ãƒ¼ãƒãƒ«ã«å®Ÿè£…ã— Vyatta Core ã«çµ„ã¿è¾¼ã¿ ASAMAP ã¨ã„ã†åå‰ã§ ã‚ªãƒ¼ãƒ—ãƒ³ã‚½ãƒ¼ã‚¹ã§å…¬é–‹ã—ãŸã‚Šã—ã¦ã„ã¾ã™ 2 3. ASAMAPï¼Ÿ â€¢ IPv4 over
yass 2014/06/26
linux

Network

nic

napi

interrupt

numa

rss

kernel
ãƒªãƒ³ã‚¯
https://github.com/xtao/PF_RING/blob/master/drivers/PF_RING_aware/intel/igb/igb-5.2.5-zc/README
yass 2014/06/21
intel

nic

igb

linux

network

driver
ãƒªãƒ³ã‚¯
Linux Base Driver for 10 Gigabit Intel(R) Ethernet Network Connection
yass 2014/06/21
intel

nic

10GbE

ixgb

linux

kernel

ethtool

sysctl
ãƒªãƒ³ã‚¯
10GbE Adapter Iperf Performance Test Results
This testing looked at iperf performance and CPU utilization (using vmstat) of the Neterion Xframe1 and the Intel IXGB 10GbE adapters. I tested different sysctl and MMRBC parameters along with varying numbers and directions of flows. Summary: In the best case of single stream, single direction throughput, the Xframe1 sustained ~6.4Gb/s. The IXGB sustained 5.6-5.9Gb/s. CPU utilization was high for
yass 2014/06/21
" I tested different sysctl and MMRBC parameters along with varying numbers and directions of flows. "

10GbE

nic

linux

kernel

intel

ixgb
ãƒªãƒ³ã‚¯
Linux TCP Tuning
The aim of this post is to point out potential kernel tunables that might improve network performance in certain scenarios. As with any other post on the subject, make sure you test before and after you make an adjustment to have a measurable, quantitative result. For the most part, the kernel is smart enough to detect and adjust certain TCP options after boot, or even dynamically, e.g the Sliding
yass 2014/06/20
linux

network

nic

10GbE
ãƒªãƒ³ã‚¯
NUMAã‚¢ãƒ¼ã‚ãƒ†ã‚¯ãƒãƒ£ã¨ãƒãƒƒãƒˆãƒ¯ãƒ¼ã‚¯æ€§èƒ½ï¼ˆï¼’ï¼‰ï¼å®Ÿè·µç·¨ - Blog
å‰å›žã€NUMAã‚¢ãƒ¼ã‚ãƒ†ã‚¯ãƒãƒ£ãŒãƒãƒƒãƒˆãƒ¯ãƒ¼ã‚¯æ€§èƒ½ã«ä¸Žãˆã‚‹å½±éŸ¿ã«ã¤ã„ã¦ã®æŠ€è¡“çš„ãªèƒŒæ™¯ã‚’èª¬æ˜Žã—ãŸã€‚ (å‰å›žè¨˜äº‹) NUMAã‚¢ãƒ¼ã‚ãƒ†ã‚¯ãƒãƒ£ã¨ãƒãƒƒãƒˆãƒ¯ãƒ¼ã‚¯æ€§èƒ½ï¼ˆï¼‘ï¼‰ï¼èª¬æ˜Žç·¨ - Interconnect Your Future - InfiniBand and RoCE æœ¬ç¨¿ã§ã¯ã€å®Ÿéš›ã«Linuxç’°å¢ƒã§ãƒãƒƒãƒˆãƒ¯ãƒ¼ã‚¯æ€§èƒ½ã‚’æ¸¬å®šã—ã€NUMAãŒä¸Žãˆã‚‹å½±éŸ¿ã«ã¤ã„ã¦ã®äº‹ä¾‹ã¨ãƒŽã‚¦ãƒã‚¦ã‚’ç´¹ä»‹ã—ãŸã„ã€‚ æœ¬ç¨¿ã§ã¯ã€æœ€çµ‚çš„ã«ã¯ä¸‹è¨˜æ¸¬å®šä¾‹ã‚’ã¾ã¨ã‚ã¦ã„ã‚‹ã€‚NUMAæ§‹æˆãŒç•°ãªã‚‹å ´åˆã€æ€§èƒ½ãŒå…¨ãç•°ãªã£ã¦ã—ã¾ã†ã€‚æ¥µç«¯ãªè©±ã€40GbE NICãªã®ã«4-5Gbpsã—ã‹å‡ºãªã„ã“ã¨ã«ãªã£ã¦ã—ã¾ã†ã€‚å®Ÿéš›ã«ã¯æ€§èƒ½å‡ºã¾ã™ã€‚æ³¨æ„ã€‚ NUMAæ§‹æˆä¸Šé ã„æ§‹æˆã§ã®iperfæ€§èƒ½ï¼š1ãƒ—ãƒã‚»ã‚¹(1ã‚³ã‚¢)Â 2.00Gbpsã€12ãƒ—ãƒã‚»ã‚¹(6ã‚³ã‚¢)Â 4.84Gbps NUMAæ§‹æˆä¸Šè¿‘ã„æ§‹æˆã§ã®iperfæ€§èƒ½ï¼š1ãƒ—ãƒã‚»ã‚¹(1ã‚³ã‚¢)Â 14.7Gbpsã€12ãƒ—ãƒã‚»
yass 2014/05/26
" NUMAæ§‹æˆã®è¨å®šï¼ˆã‚¢ãƒ•ã‚£ãƒ‹ãƒ†ã‚£(Affinityï¼‰è¨å®šï¼‰ã¯éžå¸¸ã«é‡è¦ã§ã‚ã‚‹ã€‚æ€§èƒ½çš„ã«ä¸åˆ©ãªæ§‹æˆã§æ€§èƒ½æ¸¬å®šã‚’è¡Œã†ã¨ã€æœ¬ä¾‹ã®ã‚ˆã†ãªæ¥µç«¯ãªæ§‹æˆã§ã¯ã€40GbE NICãªã®ã«4-5Gbpsã—ã‹æ€§èƒ½ãŒå‡ºãªã„ã¨ã„ã£ãŸã“ã¨ã‚‚èµ·ã“ã‚Šã†ã‚‹ã€‚"

numa

nic

affinity

cpu

iperf

pcie
ãƒªãƒ³ã‚¯
NUMAã‚¢ãƒ¼ã‚ãƒ†ã‚¯ãƒãƒ£ã¨ãƒãƒƒãƒˆãƒ¯ãƒ¼ã‚¯æ€§èƒ½ï¼ˆï¼‘ï¼‰ï¼èª¬æ˜Žç·¨ - Blog
NUMA(Non-Uniform Memory Access)ã¨ã¯ Non-uniform memory access - Wikipedia, the free encyclopedia NUMA - Wikipedia NUMAï¼ˆNon-Uniform Memory Accessã€ãƒŒãƒžï¼‰ã¨ã¯ã€å…±æœ‰ãƒ¡ãƒ¢ãƒªåž‹ãƒžãƒ«ãƒãƒ—ãƒã‚»ãƒƒã‚µã‚³ãƒ³ãƒ”ãƒ¥ãƒ¼ã‚¿ã‚·ã‚¹ãƒ†ãƒ ã®ã‚¢ãƒ¼ã‚ãƒ†ã‚¯ãƒãƒ£ã®ã²ã¨ã¤ã§ã€è¤‡æ•°ãƒ—ãƒã‚»ãƒƒã‚µãŒå…±æœ‰ã™ã‚‹ãƒ¡ã‚¤ãƒ³ãƒ¡ãƒ¢ãƒªã¸ã®ã‚¢ã‚¯ã‚»ã‚¹ã‚³ã‚¹ãƒˆãŒã€ãƒ¡ãƒ¢ãƒªé ˜åŸŸã¨ãƒ—ãƒã‚»ãƒƒã‚µã«ä¾å˜ã—ã¦å‡ä¸€ã§ãªã„ã‚¢ãƒ¼ã‚ãƒ†ã‚¯ãƒãƒ£ã§ã‚ã‚‹ã€‚ è¤‡æ•°CPUãŒã‚ã‚‹å ´åˆã€è¿‘ã„ãƒ¡ãƒ¢ãƒªã¨é ã„ãƒ¡ãƒ¢ãƒªãŒæ§‹æˆä¸Šå˜åœ¨ã—ã€ãƒ‡ãƒ¼ã‚¿ãŒç½®ã‹ã‚Œã‚‹å ´æ‰€ã«ã‚ˆã£ã¦ã€æ€§èƒ½ã«å·®ãŒå‡ºã¦ã—ã¾ã†ã‚¢ãƒ¼ã‚ãƒ†ã‚¯ãƒãƒ£ã§ã‚ã‚‹ã€‚ ãƒãƒƒãƒˆãƒ¯ãƒ¼ã‚¯æ€§èƒ½ã‚’æ¸¬å®šã™ã‚‹å ´åˆã‚‚ã€NUMAæ§‹æˆä¸Šä¸åˆ©ãªçŠ¶æ…‹ã§æ€§èƒ½æ¸¬å®šã‚’ã—ã¦ã—ã¾ã†ã¨ã€æ€ã†ã‚ˆã†ãªæ€§èƒ½ãŒå‡ºãªã„å ´åˆãŒã‚ã‚‹ãŸã‚ã€æ³¨æ„ãŒå¿…è¦ã§ã‚ã‚‹ã€‚æœ¬ç¨¿ã§ã¯ã€NUMAåŠã³CPU
yass 2014/05/26
" NUMAãƒŽãƒ¼ãƒ‰ã®è¿‘ã„ãƒ¡ãƒ¢ãƒªã«å¯¾ã—ã¦ã€é ã„ãƒ¡ãƒ¢ãƒªã¯åŠåˆ†ã®ãƒãƒ³ãƒ‰å¹…ã—ã‹æ€§èƒ½ãŒå‡ºã¦ã„ãªã„ / ãƒ¬ã‚¤ãƒ†ãƒ³ã‚·ãƒ¼ï¼ˆå‡¦ç†å®Œäº†æ™‚é–“ï¼‰ã«ã¤ã„ã¦ã‚‚ã€é ã„ãƒ¡ãƒ¢ãƒªã¯è¿‘ã„ãƒ¡ãƒ¢ãƒªã‚ˆã‚Šç´„60%ã‚‚æ™‚é–“ãŒæŽ›ã‹ã£ã¦ã—ã¾ã† "

NUMA

nic

network

cpu

latency

PCIe

affinity
ãƒªãƒ³ã‚¯
1 2 3 4 5 6 7 æ¬¡ã®ãƒšãƒ¼ã‚¸