Intel E810: erinevus redaktsioonide vahel
Allikas: Imre kasutab arvutit
Mine navigeerimisribaleMine otsikasti
| 339. rida: | 339. rida: | ||
Tulemusena saavutab iga ühendus nii 7-11 Gbit/s. Samal ajal virtuaalsed iperf3 serverid ei ole os mõttes koormatud, interrupt load on seotud ühe cpu'ga (ju see on iperf3 töötamist või kasutamise eripära). |
Tulemusena saavutab iga ühendus nii 7-11 Gbit/s. Samal ajal virtuaalsed iperf3 serverid ei ole os mõttes koormatud, interrupt load on seotud ühe cpu'ga (ju see on iperf3 töötamist või kasutamise eripära). |
||
| + | |||
| + | ===Koormustestimine - tava lahendus=== |
||
| + | |||
| + | Võrgunduse setup on sama, testid on samad. Muudatus seisneb selle, et ovs töötab non-dpdk ja non-vhost-user-protokoll režiimis. |
||
===Probleemid=== |
===Probleemid=== |
||
Redaktsioon: 10. juuli 2024, kell 12:30
Sissejuhatus
Üldistel teemadel sobib tutvuda tekstiga https://www.auul.pri.ee/wiki/Mellanox_ConnectX-6_Lx
Kasutada on võrgukaardi riistvara
# lspci | grep 810 81:00.0 Ethernet controller: Intel Corporation Ethernet Controller E810-XXV for SFP (rev 02) 81:00.1 Ethernet controller: Intel Corporation Ethernet Controller E810-XXV for SFP (rev 02)
dpdk abil liikluse kohale toomine ovs switchi juurde
Väited
- eesmärgiks on füüsiliselt võrgust kohale tuua ovs switchi peale füüsilise võrgukaardi juures dpdk lahendust kasutades võimalikult palju liiklust
- switchi peale on tekitatud üks ovs internal port (internal port on näha ka operatsioonisüteemi jaoks, st tema poole saab eemalt pöörduda, nt pve webgui kasutamiseks ja ssh abil sisse logimiseks)
- ei tegelda liikluse edasi jõudmisega ovs switchi külge kinnitatud virtuaalse arvuti juurde
- tegevused toimuvad PVE v. 8.2 keskkonnas
- midagi ei kompileerita st kõik paigaldatakse Debian ja PVE tava apt repost
Tulemuseks ovs switch kasutamise osa seadistus paistab selline
# ovs-vsctl show
26f5a7ed-dfe6-49bb-978f-062dd420f331
Bridge vmbr0
datapath_type: netdev
Port vmbr0
Interface vmbr0
type: internal
Port dpdk-p0
Interface dpdk-p0
type: dpdk
options: {dpdk-devargs="0000:81:00.0"}
Port inter
Interface inter
type: internal
ovs_version: "3.1.0"
ning ovs üldosa seadistus
# ovs-vsctl list open_vSwitch
_uuid : 26f5a7ed-dfe6-49bb-978f-062dd420f331
bridges : [4b2f6152-da5c-495b-bd7f-c3020d8e4012]
cur_cfg : 17
datapath_types : [netdev, system]
datapaths : {}
db_version : "8.3.1"
dpdk_initialized : true
dpdk_version : "DPDK 22.11.5"
external_ids : {hostname=valgustaja1, rundir="/var/run/openvswitch", system-id="257085fe-47b8-4b5d-b894-3063fbb645e2"}
iface_types : [afxdp, afxdp-nonpmd, bareudp, dpdk, dpdkvhostuser, dpdkvhostuserclient, erspan, geneve, gre, gtpu, internal, ip6erspan, ip6gre, lisp, patch, stt, system, tap, vxlan]
manager_options : []
next_cfg : 17
other_config : {dpdk-init="true", dpdk-lcore-mask="0x1", dpdk-socket-mem="4096", pmd-cpu-mask="0x0ff0", userspace-tso-enable="false", vhost-iommu-support="true"}
ovs_version : "3.1.0"
ssl : []
statistics : {}
system_type : debian
system_version : "12"
ning
root@valgustaja1# ifconfig
inter: flags=4419<UP,BROADCAST,RUNNING,PROMISC,MULTICAST> mtu 1500
inet 10.40.134.18 netmask 255.255.255.248 broadcast 10.40.134.23
inet6 fe80::b470:b7ff:fe0c:8ea9 prefixlen 64 scopeid 0x20<link>
ether b6:70:b7:0c:8e:a9 txqueuelen 1000 (Ethernet)
RX packets 9783611 bytes 14414601922 (13.4 GiB)
RX errors 0 dropped 0 overruns 0 frame 0
TX packets 4824979 bytes 409852147 (390.8 MiB)
TX errors 0 dropped 0 overruns 0 carrier 0 collisions 0
lo: flags=73<UP,LOOPBACK,RUNNING> mtu 65536
inet 127.0.0.1 netmask 255.0.0.0
inet6 ::1 prefixlen 128 scopeid 0x10<host>
loop txqueuelen 1000 (Local Loopback)
RX packets 307863 bytes 56590252 (53.9 MiB)
RX errors 0 dropped 0 overruns 0 frame 0
TX packets 307863 bytes 56590252 (53.9 MiB)
TX errors 0 dropped 0 overruns 0 carrier 0 collisions 0
Sellise olukorra saavutamiseks Ubuntu v. 24.04 operatsioonisüsteemil paigaldatakse tarkvara OVS-DPDK, sobib taustaks vaadata juhendeid
- 'How to use DPDK with Open vSwitch' - https://ubuntu.com/server/docs/how-to-use-dpdk-with-open-vswitch
- 'Using Open vSwitch with DPDK' - https://docs.openvswitch.org/en/latest/howto/dpdk/
Tarkvara paigaldamine, ice ehk intel e810 mudeli dpdk teek ei paigaldata muu hulgas sõltuvusena koos openvswitch-switch-dpdk paketiga, kuigi hulka librte-net-xxx pakette seejuures paigaldatakse
valgustaja1# apt-get install openvswitch-switch-dpdk valgustaja1# apt-get install librte-net-ice23
Peale tarkvara paigaldamist on ovs käivitatud olekus ning sobib seda edasi seadistada
valgustaja1# echo 6144 > /sys/kernel/mm/hugepages/hugepages-2048kB/nr_hugepages valgustaja1# ovs-vsctl set Open_vSwitch . "other_config:dpdk-socket-mem=4096" valgustaja1# ovs-vsctl set Open_vSwitch . "other_config:pmd-cpu-mask=0x100000100000" valgustaja1# ovs-vsctl set Open_vSwitch . "other_config:vhost-iommu-support=true" valgustaja1# ovs-vsctl set Open_vSwitch . "other_config:userspace-tso-enable=false" valgustaja1# ovs-vsctl set Open_vSwitch . "other_config:dpdk-lcore-mask=0x1" valgustaja1# ovs-vsctl set Open_vSwitch . "other_config:dpdk-init=true"
librte dpdk teegi kasutamiseks sobib öelda
valgustaja1# modprobe vfio-pci valgustaja1# modprobe vfio valgustaja1# dpdk-devbind.py --bind=vfio-pci 0000:81:00.0 valgustaja1# dpdk-devbind.py --bind=vfio-pci 0000:81:00.1
Tulemusena paistab
# dpdk-devbind.py --status Network devices using DPDK-compatible driver ============================================ 0000:81:00.0 'Ethernet Controller E810-XXV for SFP 159b' drv=vfio-pci unused=ice 0000:81:00.1 'Ethernet Controller E810-XXV for SFP 159b' drv=vfio-pci unused=ice ..
OVS switchi tekitamine ja füüsilise kaardi pordi ühendamine
valgustaja1# ovs-vsctl add-br vmbr0 -- set bridge vmbr0 datapath_type=netdev valgustaja1# ovs-vsctl add-port vmbr0 dpdk-p0 -- set Interface dpdk-p0 type=dpdk options:dpdk-devargs=0000:81:00.0 valgustaja1# ovs-vsctl add-port vmbr0 inter -- set interface inter type=internal valgustaja1# ifconfig inter 10.40.134.18/29 valgustaja1# route add default gw 10.40.134.17
Ootus on et tulemusena saab pve host arvutist pingida gw ip aadressi ja arvuti on võrku ühendatud.
Probleemid, kui librte-net-ice23 pakett on puudu, siis saab selliseid vigu
TODO
dpdk abil liikluse kohale toomine ovs switchi kaudu qemu virtuaalse arvuti juurde - vhost user protokoll
Tulemuseks ovs switch kasutamise osa seadistus paistab selline
# ovs-vsctl show
26f5a7ed-dfe6-49bb-978f-062dd420f331
Bridge vmbr0
datapath_type: netdev
Port vmbr0
Interface vmbr0
type: internal
Port vhost-user-1
tag: 3564
Interface vhost-user-1
type: dpdkvhostuserclient
options: {vhost-server-path="/var/run/vhostuserclient/vhost-user-client-1"}
Port vhost-user-2
tag: 3564
Interface vhost-user-2
type: dpdkvhostuserclient
options: {vhost-server-path="/var/run/vhostuserclient/vhost-user-client-2"}
Port dpdk-p0
Interface dpdk-p0
type: dpdk
options: {dpdk-devargs="0000:81:00.0", n_rxq="8"}
Port inter
Interface inter
type: internal
ovs_version: "3.1.0"
kus
- vhost-user-1 port vastab ühe virtuaalse arvuti ühele võrguliidesele
- vhost-user-2 port vastab teise virtuaalse arvuti ühele võrguliidesele
- /var/run/vhostuserclient/vhost-user-client-1 on unix socket, mille tekitab qemu/kvm protsess virtuaalse arvuti käivitamisel (selle tekitamist õpetab /etc/pve/qemu-server/1102.conf seadistus)
OVS osa tekitatakse eelmises punktis, lisaks on oluline tekitada OVS juurde port, nt
valgustaja1# mkdir /var/run/vhostuserclient valgustaja1# ovs-vsctl add-port vmbr0 vhost-user-1 -- set Interface vhost-user-1 type=dpdkvhostuserclient "options:vhost-server-path=/var/run/vhostuserclient/vhost-user-client-1" valgustaja1# ovs-vsctl set port vhost-user-1 tag=3564 valgustaja1# ovs-vsctl add-port vmbr0 vhost-user-2 -- set Interface vhost-user-2 type=dpdkvhostuserclient "options:vhost-server-path=/var/run/vhostuserclient/vhost-user-client-2" valgustaja1# ovs-vsctl set port vhost-user-2 tag=3564
kasutada sobiva seadistusega PVE virtuaalset arvutit, nt ühe arvuti puhul
valgustaja1# cat /etc/pve/qemu-server/1102.conf agent: 1 args: -machine q35+pve0,kernel_irqchip=split \ -device intel-iommu,intremap=on,caching-mode=on \ -chardev socket,id=char1,path=/var/run/vhostuserclient/vhost-user-client-1,server=on \ -netdev type=vhost-user,id=mynet1,chardev=char1,vhostforce=on,queues=4 \ -device virtio-net-pci,mac=12:4A:8D:1E:33:3D,netdev=mynet1,mq=on,vectors=10,rx_queue_size=1024,tx_queue_size=256 bios: ovmf boot: order=virtio0;ide2;net0 cores: 8 cpu: host efidisk0: local-to-valgustaja-1-lvm-over-mdadm:vm-1102-disk-0,efitype=4m,pre-enrolled-keys=1,size=4M hugepages: 1024 ide2: none,media=cdrom machine: q35 memory: 8192 meta: creation-qemu=9.0.0,ctime=1720472955 name: imre-ubuntu-2404-01 numa: 1 ostype: l26 rng0: source=/dev/urandom scsihw: virtio-scsi-single serial0: socket smbios1: uuid=eff2a3c6-e3e6-4ccf-889b-33babc49c3fc sockets: 1 tpmstate0: local-to-valgustaja-1-lvm-over-mdadm:vm-1102-disk-1,size=4M,version=v2.0 vcpus: 8 vga: virtio virtio0: local-to-valgustaja-1-lvm-over-mdadm:vm-1102-disk-2,backup=0,iothread=1,size=20G vmgenid: ec8bc8c2-1279-43d2-bfca-9d7b9081c97f
kus
- TODO
ja teise arvuti puhul
TODO
OVS jaoks täiendavate arvutusressursside andmiseks sobib öelda
valgustaja1# ovs-vsctl set Interface dpdk-p0 "options:n_rxq=8" valgustaja1# ovs-vsctl set Open_vSwitch . other_config:pmd-cpu-mask=0x0ff0
dpdk abil liikluse kohale toomine ovs switchi kaudu qemu virtuaalse arvuti juurde - vf representor
TODO
Koormustestimine - vhost user lahendus
Koormustestimise võrgujoonis
- valgustaja1 - 10.40.134.18 - füüsiline pve node/host arvuti (25 Gbit/s)
- ubuntu-2404-01 - 10.40.135.66 - virtuaalne arvuti
- ubuntu-2404-02 - 10.40.135.67 - virtuaalne arvuti
- ubuntu-2404-03 - 10.40.135.68 - virtuaalne arvuti
- gen-1 - 10.40.13.242 - füüsiline arvuti (25 Gbit/s)
- gen-2 - 10.40.13.246 - füüsiline arvuti (25 Gbit/s)
syn flood
Kolmes virtuaalses arvutis ei käivitata midagi spetsiaalselt, lihtsalt paketifilter ei takista vms. Syn paketile vastatakse rst paketiga.
gen-1 arvutis käivitatakse kaheksa protsessi (24 on random tühi port)
# cat run-hping3-to-66.sh timeout 60 hping3 -S -c 400000000 --flood -p 24 10.40.135.66 -m 200 & timeout 60 hping3 -S -c 400000000 --flood -p 24 10.40.135.66 -m 200 & timeout 60 hping3 -S -c 400000000 --flood -p 24 10.40.135.66 -m 200 & timeout 60 hping3 -S -c 400000000 --flood -p 24 10.40.135.66 -m 200 & timeout 60 hping3 -S -c 400000000 --flood -p 24 10.40.135.66 -m 200 & timeout 60 hping3 -S -c 400000000 --flood -p 24 10.40.135.66 -m 200 & timeout 60 hping3 -S -c 400000000 --flood -p 24 10.40.135.66 -m 200 & timeout 60 hping3 -S -c 400000000 --flood -p 24 10.40.135.66 -m 200 &
ning ka analoogne 'run-hping3-to-67.sh'.
gen-2 arvutis käivitakse analoone 'run-hping3-to-68.sh'.
vnstat abil vaadeldes paistab
- üks hping3 protsess tekitab 200 kpps sekundis väljuvat liiklust (ootus oleks et tekib rohkem)
- ühe hping3 protsessi tagasihoidlikku väljndit kompenseeritakse mitme hping3 protsessi samaaegse käivitamisega (üks 8 protsessi komplekt tekitab nii 1200 kpps)
- tulemusena tekitatakse kokku gen-1 ja gen-2 peal väljuvat liiklust ca 3 x 1200 = 3.6 mpps
- see liiklus jõuab virtuaalsetele arvutite koplekti juurde kohale 'vnstat -l -i enp0s2' abil hinnates, seejuures on virtuaalsed arvutid cpu mõttes väga koormatud
Virtuaalsete arvutite jõudlust määrab protsessorite arv (8) ning nendega peab olema kooskõlas qemu virtuaalse arvuti seadistus
cores: 8 memory: 8192 numa: 1 sockets: 1 vcpus: 8
ja ovs port seadistus eriti rxq osas, ühe arvuti puhul peab see olema 8, kolme puhul ehk 3 x 8, peab uurima
# ovs-vsctl show
...
Port dpdk-p0
Interface dpdk-p0
type: dpdk
options: {dpdk-devargs="0000:81:00.0", n_rxq="16"}
Tulemusena peavad saama kõik protsessorid virtuaalses arvutis nn interrupt load'i (vaadata top + 1 abil, ja top + H abil). Ja PVE host peal võiks ka kaheksa CPU PMD 100% koormus nihkuda 'us' pealt 'sy' peale.
iperf3
Kolmes virtuaalses arvutis on kävitatud 'iperf3 -s' serverid.
gen-1 arvutis käivitatakse
gen-1# iperf3 -t 180 -c 10.40.135.66 gen-1# iperf3 -t 180 -c 10.40.135.67
gen-2 arvutis käivitatakse
gen-2# iperf3 -t 180 -c 10.40.135.68
Taustal jälgitakse võrguliidesel vnstat abil väljuva liikluse statistikat
gen-1# vnstat -l -i enp129s0f0 gen-2# vnstat -l -i enp129s0f0
Tulemusena saavutab iga ühendus nii 7-11 Gbit/s. Samal ajal virtuaalsed iperf3 serverid ei ole os mõttes koormatud, interrupt load on seotud ühe cpu'ga (ju see on iperf3 töötamist või kasutamise eripära).
Koormustestimine - tava lahendus
Võrgunduse setup on sama, testid on samad. Muudatus seisneb selle, et ovs töötab non-dpdk ja non-vhost-user-protokoll režiimis.
Probleemid
Probleem tekib siis kui füüsilisele võrgukaardile vastav librte teek on paigaldamata, logis on siis
valgustaja1# less /var/log/openvswitch/ovs-vswitchd.log ... 2024-07-09T00:26:12.303Z|00052|dpdk|ERR|EAL: Driver cannot attach the device (0000:81:00.0) 2024-07-09T00:26:12.303Z|00053|dpdk|ERR|EAL: Failed to attach device on primary process 2024-07-09T00:26:12.303Z|00054|netdev_dpdk|WARN|Error attaching device '0000:81:00.0' to DPDK 2024-07-09T00:26:12.303Z|00055|netdev|WARN|dpdk-p0: could not set configuration (Invalid argument) 2024-07-09T00:26:12.303Z|00056|dpdk|ERR|Invalid port_id=32
ning
root@valgustaja1# cat ovs-vsctl-show-katki
e3c33e5f-ba41-45e8-b0af-f7310a63f5ce
Bridge vmbr0
Port dpdk-p0
Interface dpdk-p0
type: dpdk
options: {dpdk-devargs="0000:81:00.0"}
error: "Error attaching device '0000:81:00.0' to DPDK"
...
Mõisted
- epct - ethernet port configuration tool
Riistvara seadistamine
arvuti BIOS
TODO
HII
TODO
devlink utiliit
TODO
epct utiliit - os käsurida
epct võiks olla väga asjakohane tööriist, aga 2024 aasta kevadel ta praktiliselt eriti ei tööta, ei linux kernel v 6.x ega v. 5.x puhul, https://www.intel.com/content/www/us/en/download/19437/ethernet-port-configuration-tool-linux.html
kui ice draiver on laaditud (kasutatakse kernel native ice draiverit, analoogne tulemus on epct v1.41.03.01)
# ./epct64e -devices Ethernet Port Configuration Tool EPCT version: v1.40.05.05 Copyright 2019 - 2023 Intel Corporation. Cannot initialize port: [00:129:00:00] Intel(R) Ethernet Controller E810-XXV for SFP Cannot initialize port: [00:129:00:01] Intel(R) Ethernet Controller E810-XXV for SFP Error: Cannot initialize adapter.
kui ice draiver ei ole laaditud
# ./epct64e -devices Ethernet Port Configuration Tool EPCT version: v1.40.05.05 Copyright 2019 - 2023 Intel Corporation. Base driver not supported or not present: [00:129:00:00] Intel(R) Ethernet Controller E810-XXV for SFP NIC Seg:Bus:Fun Ven-Dev Connector Ports Speed Quads Lanes per PF === ============= ========= ========= ===== ======== ====== ============ 1) 000:129:00-01 8086-159B SFP 2 - Gbps N/A N/A Error: Base driver is not available for one or more adapters. Please ensure the driver is correctly attached to the device.
epct utiliit - efi rakendus
epct efi rakendus natuke töötab, aga ei võimalda siis 'svio enable' teha, https://www.intel.com/content/www/us/en/download/19440/ethernet-port-configuration-tool-efi.html
nvmeupdate efi rakendus iseenesest ei anna viga, aga ei saa ka aru, et süsteemis oleks e810 kaart, https://www.intel.com/content/www/us/en/download/19629/non-volatile-memory-nvm-update-utility-for-intel-ethernet-network-adapters-e810-series-efi.html
Misc
Kasutada on üks füüsiline kahe pordiga võrgukaart
# lspci | grep -i net 81:00.0 Ethernet controller: Intel Corporation Ethernet Controller E810-XXV for SFP (rev 02) 81:00.1 Ethernet controller: Intel Corporation Ethernet Controller E810-XXV for SFP (rev 02)
Operatsioonisüsteemi tavalised võrguseadmed on sellised
root@pve-02:~# ethtool enp129s0f1np1
Settings for enp129s0f1np1:
Supported ports: [ FIBRE ]
Supported link modes: 1000baseT/Full
25000baseCR/Full
25000baseSR/Full
1000baseX/Full
10000baseCR/Full
10000baseSR/Full
10000baseLR/Full
Supported pause frame use: Symmetric
Supports auto-negotiation: Yes
Supported FEC modes: None RS BASER
Advertised link modes: 25000baseCR/Full
10000baseCR/Full
Advertised pause frame use: No
Advertised auto-negotiation: Yes
Advertised FEC modes: None RS BASER
Link partner advertised link modes: Not reported
Link partner advertised pause frame use: No
Link partner advertised auto-negotiation: Yes
Link partner advertised FEC modes: Not reported
Speed: 25000Mb/s
Duplex: Full
Auto-negotiation: on
Port: Direct Attach Copper
PHYAD: 0
Transceiver: internal
Supports Wake-on: g
Wake-on: d
Current message level: 0x00000007 (7)
drv probe link
Link detected: yes
root@pve-02:~# ethtool -i enp129s0f1np1
driver: ice
version: 6.5.13-5-pve
firmware-version: 4.40 0x8001c7d4 1.3534.0
expansion-rom-version:
bus-info: 0000:81:00.1
supports-statistics: yes
supports-test: yes
supports-eeprom-access: yes
supports-register-dump: yes
supports-priv-flags: yes
Paistab devlink vaatest sedasi
# devlink dev info
pci/0000:81:00.0:
driver ice
serial_number 00-01-00-ff-ff-00-00-00
versions:
fixed:
board.id K58132-000
running:
fw.mgmt 7.4.13
fw.mgmt.api 1.7.11
fw.mgmt.build 0xded4446f
fw.undi 1.3534.0
fw.psid.api 4.40
fw.bundle_id 0x8001c7d4
fw.app.name ICE OS Default Package
fw.app 1.3.36.0
fw.app.bundle_id 0xc0000001
fw.netlist 4.4.5000-2.15.0
fw.netlist.build 0x0ba411b9
stored:
fw.undi 1.3534.0
fw.psid.api 4.40
fw.bundle_id 0x8001c7d4
fw.netlist 4.4.5000-2.15.0
fw.netlist.build 0x0ba411b9
...
RoCEv2 kasutamine
Oluline on, et see ip konf millega opereeritakse järgnevalt oleks kinnitatud füüsilise seadme külge (st mitte nii, et arvutis on kasutusel nt OVS virtual switch ning ip on vlan47 küljes, sama ovs bridge küljes on füüsiline enp129s0f1np1 jne, st siis ei toimi)
root@pve-02:~# ip addr show dev enp129s0f1np1:
3: enp129s0f1np1: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc mq state UP group default qlen 1000
link/ether 3c:ec:ef:e6:69:b9 brd ff:ff:ff:ff:ff:ff
inet 10.47.218.226/24 scope global enp129s0f1np1
valid_lft forever preferred_lft forever
inet6 fe80::3eec:efff:fee6:69b9/64 scope link
valid_lft forever preferred_lft forever
Muu hulgas toetab see seade iwarp ja rocev2 protokolle, devlink vaatest paistab see nii
# devlink dev param
pci/0000:81:00.0:
name enable_roce type generic
values:
cmode runtime value true
name enable_iwarp type generic
values:
cmode runtime value false
pci/0000:81:00.1:
name enable_roce type generic
values:
cmode runtime value false
name enable_iwarp type generic
values:
cmode runtime value true
Korraga saab olla aktiivne üks või teine, nende vahel valmine toimub nt selliselt devlink abil
# devlink dev param set pci/0000:81:00.1 name enable_iwarp value false cmode runtime
rocev2 kasutamiseks on asjakohen süsteemidesse paigaldada omajagu rdma ja infiniband traditsiooniga tarkvara, nt
# apt-get install rdma-core # apt-get install perftest # apt-get install ibverbs-utils # apt-get install infiniband-diags # apt-get install ibverbs-providers
Asjasse puutuvad sellised driverid
- ice - põhi draiver
- irdma - intel võrgukaartide (2024 aastal e810 ja midagi vanemat ka) - 'intel rdma'
Seadmed kui rdma seadmed paistavad
root@pve-01:~# rdma link link rocep129s0f0/1 state ACTIVE physical_state LINK_UP netdev enp129s0f0np0 link rocep129s0f1/1 state ACTIVE physical_state LINK_UP netdev enp129s0f1np1
Kahe otse kaabliga süsteemi uurimiseks sobib öelda ühes ja teises arvutis nt nii, pve-02 on nö rping server ja pve-01 on rping klient; seejuures on iseloomulik, et tavalise võrguliidese peal tavalisel viisil võrku pealt kuulates ei ole midagi näha
root@pve-02:~# rping -s -a 10.47.218.226 -v server ping data: rdma-ping-0: ABCDEFGHIJKLMNOPQRSTUVWXYZ[\]^_`abcdefghijklmnopqr server ping data: rdma-ping-1: BCDEFGHIJKLMNOPQRSTUVWXYZ[\]^_`abcdefghijklmnopqrs server ping data: rdma-ping-2: CDEFGHIJKLMNOPQRSTUVWXYZ[\]^_`abcdefghijklmnopqrst server ping data: rdma-ping-3: DEFGHIJKLMNOPQRSTUVWXYZ[\]^_`abcdefghijklmnopqrstu server ping data: rdma-ping-4: EFGHIJKLMNOPQRSTUVWXYZ[\]^_`abcdefghijklmnopqrstuv server DISCONNECT EVENT... wait for RDMA_READ_ADV state 10 root@pve-01:~# rping -c -a 10.47.218.226 -v -C 5 ping data: rdma-ping-0: ABCDEFGHIJKLMNOPQRSTUVWXYZ[\]^_`abcdefghijklmnopqr ping data: rdma-ping-1: BCDEFGHIJKLMNOPQRSTUVWXYZ[\]^_`abcdefghijklmnopqrs ping data: rdma-ping-2: CDEFGHIJKLMNOPQRSTUVWXYZ[\]^_`abcdefghijklmnopqrst ping data: rdma-ping-3: DEFGHIJKLMNOPQRSTUVWXYZ[\]^_`abcdefghijklmnopqrstu ping data: rdma-ping-4: EFGHIJKLMNOPQRSTUVWXYZ[\]^_`abcdefghijklmnopqrstuv client DISCONNECT EVENT...
Veel huvitavaid utiliite, seejuures on esimene seade 'devlink dev param ...' abil seadistatud rocev2 režiimi ja teine iwarp režiimi
# ibv_devices
device node GUID
------ ----------------
rocep129s0f0 3eeceffffee667b2
iwp129s0f1 3eeceffffee667b3
# ibv_devinfo
hca_id: rocep129s0f0
transport: InfiniBand (0)
fw_ver: 1.71
node_guid: 3eec:efff:fee6:67b2
sys_image_guid: 3eec:efff:fee6:67b2
vendor_id: 0x8086
vendor_part_id: 5531
hw_ver: 0x2
phys_port_cnt: 1
port: 1
state: PORT_ACTIVE (4)
max_mtu: 4096 (5)
active_mtu: 1024 (3)
sm_lid: 0
port_lid: 1
port_lmc: 0x00
link_layer: Ethernet
hca_id: iwp129s0f1
transport: iWARP (1)
fw_ver: 1.71
node_guid: 3eec:efff:fee6:67b3
sys_image_guid: 3eec:efff:fee6:67b3
vendor_id: 0x8086
vendor_part_id: 5531
hw_ver: 0x2
phys_port_cnt: 1
port: 1
state: PORT_ACTIVE (4)
max_mtu: 4096 (5)
active_mtu: 1024 (3)
sm_lid: 0
port_lid: 1
port_lmc: 0x00
link_layer: Ethernet
Töötaval juhul roce seadmega
root@pve-02:~# ib_send_bw -i 1 -d rocep129s0f1
************************************
* Waiting for client to connect... *
************************************
---------------------------------------------------------------------------------------
Send BW Test
Dual-port : OFF Device : rocep129s0f1
Number of qps : 1 Transport type : IB
Connection type : RC Using SRQ : OFF
PCIe relax order: ON
ibv_wr* API : OFF
RX depth : 512
CQ Moderation : 1
Mtu : 1024[B]
Link type : Ethernet
GID index : 1
Max inline data : 0[B]
rdma_cm QPs : OFF
Data ex. method : Ethernet
---------------------------------------------------------------------------------------
local address: LID 0x01 QPN 0x0004 PSN 0x807cd5
GID: 00:00:00:00:00:00:00:00:00:00:255:255:10:47:218:226
remote address: LID 0x01 QPN 0x0004 PSN 0xba4fee
GID: 00:00:00:00:00:00:00:00:00:00:255:255:10:47:218:225
---------------------------------------------------------------------------------------
#bytes #iterations BW peak[MB/sec] BW average[MB/sec] MsgRate[Mpps]
Conflicting CPU frequency values detected: 3696.643000 != 3000.000000. CPU Frequency is not max.
65536 1000 0.00 2762.05 0.044193
---------------------------------------------------------------------------------------
ja esimene arvuti kliendina
root@pve-01:~# ib_send_bw -i 1 10.47.218.226 -d rocep129s0f1
---------------------------------------------------------------------------------------
Send BW Test
Dual-port : OFF Device : rocep129s0f1
Number of qps : 1 Transport type : IB
Connection type : RC Using SRQ : OFF
PCIe relax order: ON
ibv_wr* API : OFF
TX depth : 128
CQ Moderation : 1
Mtu : 1024[B]
Link type : Ethernet
GID index : 1
Max inline data : 0[B]
rdma_cm QPs : OFF
Data ex. method : Ethernet
---------------------------------------------------------------------------------------
local address: LID 0x01 QPN 0x0004 PSN 0xba4fee
GID: 00:00:00:00:00:00:00:00:00:00:255:255:10:47:218:225
remote address: LID 0x01 QPN 0x0004 PSN 0x807cd5
GID: 00:00:00:00:00:00:00:00:00:00:255:255:10:47:218:226
---------------------------------------------------------------------------------------
#bytes #iterations BW peak[MB/sec] BW average[MB/sec] MsgRate[Mpps]
Conflicting CPU frequency values detected: 3698.021000 != 3000.000000. CPU Frequency is not max.
65536 1000 2757.88 2757.84 0.044126
---------------------------------------------------------------------------------------
kus
- tundub, et ib_send_bw on infiniband põlvnevusega utiliit
- Transport type on IB ehk infiniband, st nii nagu roce protokolli tööpõhimõte on
- Link type on ethernet
Mittetöötav juhum - kuna kasutatakse iwarp seadmel
root@pve-02:~# ib_send_bw -i 1 -d iwp129s0f1
************************************
* Waiting for client to connect... *
************************************
---------------------------------------------------------------------------------------
Send BW Test
Dual-port : OFF Device : iwp129s0f1
Number of qps : 1 Transport type : IW
Connection type : RC Using SRQ : OFF
PCIe relax order: ON
ibv_wr* API : OFF
RX depth : 512
CQ Moderation : 1
Mtu : 1024[B]
Link type : Ethernet
GID index : 0
Max inline data : 0[B]
rdma_cm QPs : OFF
Data ex. method : Ethernet
---------------------------------------------------------------------------------------
ethernet_read_keys: Couldn't read remote address
Unable to read to socket/rdma_cm
Failed to exchange data between server and clients
Failed to deallocate PD - Device or resource busy
Failed to destroy resources
Mittetöötav juhtum - kuna seade on küll arvutis olemas, aga seal ei ole kinnitatud kõnealust ip aadressi
root@pve-02:~# ib_send_bw -i 1 -d rocep129s0f0
************************************
* Waiting for client to connect... *
************************************
---------------------------------------------------------------------------------------
Send BW Test
Dual-port : OFF Device : rocep129s0f0
Number of qps : 1 Transport type : IB
Connection type : RC Using SRQ : OFF
PCIe relax order: ON
ibv_wr* API : OFF
RX depth : 512
CQ Moderation : 1
Mtu : 1024[B]
Link type : Ethernet
GID index : 0
Max inline data : 0[B]
rdma_cm QPs : OFF
Data ex. method : Ethernet
---------------------------------------------------------------------------------------
Failed to modify QP 5 to RTR
Unable to Connect the HCA's through the link
Kõige praktilisem kasutusjuht on iscsi kasutamine iser + rocev2 viisil. Tulemusena on andmevahetuse kiirus 2x suurem võrreldes nö tavalisega. Target näeb välja selline
/> ls / o- / ......................................................................................................................... [...] o- backstores .............................................................................................................. [...] | o- block .................................................................................................. [Storage Objects: 1] | | o- iscsi_block_md127 ............................................................. [/dev/md127 (27.9TiB) write-thru activated] | | o- alua ................................................................................................... [ALUA Groups: 1] | | o- default_tg_pt_gp ....................................................................... [ALUA state: Active/optimized] | o- fileio ................................................................................................. [Storage Objects: 0] | o- pscsi .................................................................................................. [Storage Objects: 0] | o- ramdisk ................................................................................................ [Storage Objects: 0] o- iscsi ............................................................................................................ [Targets: 1] | o- iqn.2003-01.org.setup.lun.test .................................................................................... [TPGs: 1] | o- tpg1 .......................................................................................... [no-gen-acls, auth per-acl] | o- acls .......................................................................................................... [ACLs: 1] | | o- iqn.1993-08.org.debian:01:b65e1ba35869 ................................................... [1-way auth, Mapped LUNs: 1] | | o- mapped_lun0 ..................................................................... [lun0 block/iscsi_block_md127 (rw)] | o- luns .......................................................................................................... [LUNs: 1] | | o- lun0 ........................................................ [block/iscsi_block_md127 (/dev/md127) (default_tg_pt_gp)] | o- portals .................................................................................................... [Portals: 1] | o- 10.47.218.226:3261 ............................................................................................. [iser] o- loopback ......................................................................................................... [Targets: 0] o- srpt ............................................................................................................. [Targets: 0] o- vhost ............................................................................................................ [Targets: 0] o- xen-pvscsi ....................................................................................................... [Targets: 0]
ning initiator
root@pve-01:~# iscsiadm -m discovery -t st -p 10.47.218.226:3261 root@pve-01:~# iscsiadm -m node -T iqn.2003-01.org.setup.lun.test -p 10.47.218.226:3261 -o update -n iface.transport_name -v iser root@pve-01:~# iscsiadm -m node -T iqn.2003-01.org.setup.lun.test -p 10.47.218.226:3261 -l root@pve-01:~# lsscsi -s ... root@pve-01:~# iscsiadm -m node -T iqn.2003-01.org.setup.lun.test -p 10.47.218.226:3261 -u root@pve-01:~# iscsiadm -m discoverydb -t sendtargets -p 10.47.218.226:3261 -o delete root@pve-01:~# iscsiadm -m discoverydb
Scalable Functions
Kasulikud lisamaterjalid
- https://metonymical.hatenablog.com/entry/2021/07/28/201833
- https://www.youtube.com/watch?v=G6D-jaCs6sc
- https://xryan.net/p/309
Kasulikud lisamaterjalid
- 'Intel E800 devices now support iWARP and RoCE protocols' - https://access.redhat.com/documentation/en-us/red_hat_enterprise_linux/8/html/8.7_release_notes/new-features
- https://forum.proxmox.com/threads/mellanox-connectx-4-lx-and-brigde-vlan-aware-on-proxmox-8-0-1.130902/
- https://docs.nvidia.com/networking/display/mftv4270/updating+firmware+using+ethtool/devlink+and+-mfa2+file
- https://edc.intel.com/content/www/ca/fr/design/products/ethernet/appnote-e810-eswitch-switchdev-mode-config-guide/eswitch-mode-switchdev-and-legacy/
- nvidia-mlnx-ofed-documentation-v23-07.pdf
- https://enterprise-support.nvidia.com/s/article/How-To-Enable-Verify-and-Troubleshoot-RDMA
- https://enterprise-support.nvidia.com/s/article/howto-configure-lio-enabled-with-iser-for-rhel7-inbox-driver
- https://dri.freedesktop.org/docs/drm/networking/devlink/ice.html
