2021 first ovirt network question
Hi, We're using oVirt as VDI solution for awhile. Infrastacture is 2 HPE Gen10 server with 80x2 Core cpu, 2TB RAM, 16GBX2 HBA For getting performance i've splited the networks on physical layer 2X10GB LACP for VM network 2X10GB LACP for Display(Spice) network 1GB for management + Default Route After 40 VMs spice connections start to be laggy. We've found our SAN storage I/O is bad and we'll change it SSD But, on the other hand, RDP connections is flawless. Now, we have 70 VMs and i need to solve this problem. Do you thing this is a network problem? We've also checked dropping packages on the switches, but found nothing.
Hello, On Mon, Jan 4, 2021 at 8:41 AM <ozmen62@hotmail.com> wrote:
Hi, We're using oVirt as VDI solution for awhile. Infrastacture is 2 HPE Gen10 server with 80x2 Core cpu, 2TB RAM, 16GBX2 HBA
For getting performance i've splited the networks on physical layer 2X10GB LACP for VM network 2X10GB LACP for Display(Spice) network 1GB for management + Default Route
After 40 VMs spice connections start to be laggy. We've found our SAN storage I/O is bad and we'll change it SSD But, on the other hand, RDP connections is flawless. Now, we have 70 VMs and i need to solve this problem.
Do you thing this is a network problem? We've also checked dropping packages on the switches, but found nothing. _______________________________________________
40 concurrent spice connections or 40 active VMs? What type of VMs? - Gilboa
Concurrent spice connection. Today 75 VM i exist and all use spice connection Thanks for your interest
Sadly enough we never had more than 20+ concurrent spice connections and most of them were text consoles. In this case, our 40/10GbE link(s) were mostly idle. That said, there's a vast difference between spice displaying VGA resolution and spice displaying 2K/4K resolution. Hence my question about the type of VMs and by extension, the resolution used. BTW, stupid question: Have you tried running iperf3 (or a similar tool) between your client machines and the oVirt cluster to verify that indeed you have 10GbE of usable bandwith? - Gilboa On Wed, Jan 6, 2021 at 10:58 AM <ozmen62@hotmail.com> wrote:
Concurrent spice connection. Today 75 VM i exist and all use spice connection
Thanks for your interest _______________________________________________ Users mailing list -- users@ovirt.org To unsubscribe send an email to users-leave@ovirt.org Privacy Statement: https://www.ovirt.org/privacy-policy.html oVirt Code of Conduct: https://www.ovirt.org/community/about/community-guidelines/ List Archives: https://lists.ovirt.org/archives/list/users@ovirt.org/message/MIXL57E6XTRLUF...
Thanks for advice. Here is the iperf3 test results Connecting to host xyz.example.com, port 6000 [ 4] local XXX.XXX.XXX.XXX port 36140 connected to XYZ.XYZ.XYZ.XYZ port 6000 [ ID] Interval Transfer Bandwidth Retr Cwnd [ 4] 0.00-1.00 sec 81.9 MBytes 687 Mbits/sec 0 352 KBytes [ 4] 1.00-2.00 sec 80.5 MBytes 676 Mbits/sec 0 352 KBytes [ 4] 2.00-3.00 sec 80.5 MBytes 676 Mbits/sec 0 362 KBytes [ 4] 3.00-4.00 sec 79.8 MBytes 669 Mbits/sec 0 362 KBytes [ 4] 4.00-5.00 sec 80.6 MBytes 676 Mbits/sec 0 362 KBytes [ 4] 5.00-6.00 sec 82.0 MBytes 688 Mbits/sec 0 362 KBytes [ 4] 6.00-7.00 sec 81.3 MBytes 682 Mbits/sec 0 362 KBytes [ 4] 7.00-8.00 sec 80.5 MBytes 676 Mbits/sec 0 362 KBytes [ 4] 8.00-9.00 sec 80.5 MBytes 676 Mbits/sec 0 362 KBytes [ 4] 9.00-10.00 sec 79.8 MBytes 669 Mbits/sec 0 362 KBytes Actually, what i was trying to ask is is there any reletion between "default route" and "display" network on oVirt network. Because, if there is, it might be the source of the problem. As i wrote above, my "ovirtmanagement" and "default route" network on the 1Gb port
On Thu, Jan 7, 2021, 16:14 <ozmen62@hotmail.com> wrote:
Thanks for advice. Here is the iperf3 test results
Connecting to host xyz.example.com, port 6000 [ 4] local XXX.XXX.XXX.XXX port 36140 connected to XYZ.XYZ.XYZ.XYZ port 6000 [ ID] Interval Transfer Bandwidth Retr Cwnd [ 4] 0.00-1.00 sec 81.9 MBytes 687 Mbits/sec 0 352 KBytes
[ 4] 1.00-2.00 sec 80.5 MBytes 676 Mbits/sec 0 352 KBytes
[ 4] 2.00-3.00 sec 80.5 MBytes 676 Mbits/sec 0 362 KBytes
[ 4] 3.00-4.00 sec 79.8 MBytes 669 Mbits/sec 0 362 KBytes
[ 4] 4.00-5.00 sec 80.6 MBytes 676 Mbits/sec 0 362 KBytes
[ 4] 5.00-6.00 sec 82.0 MBytes 688 Mbits/sec 0 362 KBytes
[ 4] 6.00-7.00 sec 81.3 MBytes 682 Mbits/sec 0 362 KBytes
[ 4] 7.00-8.00 sec 80.5 MBytes 676 Mbits/sec 0 362 KBytes
[ 4] 8.00-9.00 sec 80.5 MBytes 676 Mbits/sec 0 362 KBytes
[ 4] 9.00-10.00 sec 79.8 MBytes 669 Mbits/sec 0 362 KBytes
Assuming xyz.example.com is in the display network you need to find out what is throttling you at these values.
Actually, what i was trying to ask is is there any reletion between "default route" and "display" network on oVirt network.
No, they should not be related Because, if there is, it might be the source of the problem.
As i wrote above, my "ovirtmanagement" and "default route" network on the 1Gb port _______________________________________________ Users mailing list -- users@ovirt.org To unsubscribe send an email to users-leave@ovirt.org Privacy Statement: https://www.ovirt.org/privacy-policy.html oVirt Code of Conduct: https://www.ovirt.org/community/about/community-guidelines/ List Archives: https://lists.ovirt.org/archives/list/users@ovirt.org/message/TPQXDGMQOVCO2F...
OK, meybe it's not the network issue. So, can you offer me a specific guide/way to increase spice server performance?
Hi, Let me share our experience with VDI and spice. First I do not have a solution to make the spice server more reactive. Here is our current infra: 12 hosts with Xeon processor with 10G port and a couple of hundreds of memory per host. This runs around 55 VMs 40 of them are servers vms and 15 desktops VMs . running Ubuntu/Windows The desktop vms are accessed from a Windows machine on a 1G network. Laggy for us is if we move a window and the mouse does not precisely move with the windows then it is laggy. So with our current setup spice server is not usable. The offset of the mouse vs windows is a couple a inches. It is like moving a window attached to an elastic band. RDP to a windows vm is snappy. RDP to a xrdp server on Ubuntu running Gnome is laaagy. The only usable remote desktop protocol we are able to use is Xpra and the NX protocol. Xpra in seamless mode is by far the snapiest and faster protocol. The problem with Xpra is some stability issues. But the support from the dev is amazing. The NX protocol is also a good choice. The x2go variant is a good seamless option. It is as fast as Xpra but the Xpra has a sharper rendering. There is also nomachine software( creator of the NX protocol) that has the best desktop experience. I think you can try it for free for 30 days and the licencing is not that expensive. I always ask myself the question why RDP on windows is way better than the other options. Even the xrdp server is no match even using the same windows client. Is this related to the windows env.(i.e gnome) or the Xserver.. So if you find a way to make the spice server more snappy i would like to know. Maybe a seamless mode for spice server could help. Regards. On Fri, Jan 8, 2021 at 12:05 AM <ozmen62@hotmail.com> wrote:
OK, meybe it's not the network issue. So, can you offer me a specific guide/way to increase spice server performance? _______________________________________________ Users mailing list -- users@ovirt.org To unsubscribe send an email to users-leave@ovirt.org Privacy Statement: https://www.ovirt.org/privacy-policy.html oVirt Code of Conduct: https://www.ovirt.org/community/about/community-guidelines/ List Archives: https://lists.ovirt.org/archives/list/users@ovirt.org/message/LKKDE7S35REW4L...
Hello, Can you run iperf3 simultaneously on multiple clients? The idea is simple, check that indeed you are getting more than 5-10GbE aggregated bandwidth in/from the machine. As for comparing the performance to rdp - I'm not sure you are comparing apples to apples. As far as I remember, by default rdp clients drop the quality of the display when faced with limited bandwidth. No sure virt-viewer can do the same (I'll let the kind people @RedHat confirm deny the above). Either, I'd suggest you start by verifying that you are indeed getting 10GbE from the clients to the display network. - Gilboa On Thu, Jan 7, 2021 at 4:14 PM <ozmen62@hotmail.com> wrote:
Thanks for advice. Here is the iperf3 test results
Connecting to host xyz.example.com, port 6000 [ 4] local XXX.XXX.XXX.XXX port 36140 connected to XYZ.XYZ.XYZ.XYZ port 6000 [ ID] Interval Transfer Bandwidth Retr Cwnd [ 4] 0.00-1.00 sec 81.9 MBytes 687 Mbits/sec 0 352 KBytes
[ 4] 1.00-2.00 sec 80.5 MBytes 676 Mbits/sec 0 352 KBytes
[ 4] 2.00-3.00 sec 80.5 MBytes 676 Mbits/sec 0 362 KBytes
[ 4] 3.00-4.00 sec 79.8 MBytes 669 Mbits/sec 0 362 KBytes
[ 4] 4.00-5.00 sec 80.6 MBytes 676 Mbits/sec 0 362 KBytes
[ 4] 5.00-6.00 sec 82.0 MBytes 688 Mbits/sec 0 362 KBytes
[ 4] 6.00-7.00 sec 81.3 MBytes 682 Mbits/sec 0 362 KBytes
[ 4] 7.00-8.00 sec 80.5 MBytes 676 Mbits/sec 0 362 KBytes
[ 4] 8.00-9.00 sec 80.5 MBytes 676 Mbits/sec 0 362 KBytes
[ 4] 9.00-10.00 sec 79.8 MBytes 669 Mbits/sec 0 362 KBytes
Actually, what i was trying to ask is is there any reletion between "default route" and "display" network on oVirt network. Because, if there is, it might be the source of the problem. As i wrote above, my "ovirtmanagement" and "default route" network on the 1Gb port _______________________________________________ Users mailing list -- users@ovirt.org To unsubscribe send an email to users-leave@ovirt.org Privacy Statement: https://www.ovirt.org/privacy-policy.html oVirt Code of Conduct: https://www.ovirt.org/community/about/community-guidelines/ List Archives: https://lists.ovirt.org/archives/list/users@ovirt.org/message/TPQXDGMQOVCO2F...
Thanks for all, Clients have 1Gb and iperf3 results are 900Mb per client, if multiple client run at same time. I think client and server network is ok. I've figured out something else. On Host setting there is an option for "console and gpu". separated or consolidated. It's made me suspected that the issue might come from this. I've checked my gpu which is onboard lshw -C Display *-display description: VGA compatible controller product: MGA G200eH3 vendor: Matrox Electronics Systems Ltd. physical id: 0.1 bus info: pci@0000:01:00.1 logical name: /dev/fb0 version: 02 width: 32 bits clock: 33MHz capabilities: pm msi pciexpress vga_controller bus_master cap_list rom fb configuration: depth=32 driver=mgag200 latency=0 resolution=1024,768 resources: irq:17 memory:d8000000-d8ffffff memory:d9b98000-d9b9bfff memory:d9000000-d97fffff memory:c0000-dffff Spice uses gpu, RDP uses cpu. So, maybe if i change my primary gpu with some powerful one it can effect spice in a good way What do you think about that?
Hello, As far as I know, spice doesn't support GPU acceleration on the host. Its CPU only. The vGPU option in the host configuration only works for specific enterprise grade GPUs. At least in my experience, spice performance is limited by three different parameters: A. Bandwidth. B. Network latency. C. Host CPU utilization. BTW, I just fired up ~20 virt-viewer windows, ~1/3 of them GUI w/ 2K resolution, over a 10GbE link and performance was OK, and network utilization was well below 1GbE. - Gilboa On Mon, Jan 11, 2021 at 10:00 AM <ozmen62@hotmail.com> wrote:
Thanks for all, Clients have 1Gb and iperf3 results are 900Mb per client, if multiple client run at same time. I think client and server network is ok.
I've figured out something else. On Host setting there is an option for "console and gpu". separated or consolidated. It's made me suspected that the issue might come from this.
I've checked my gpu which is onboard
lshw -C Display *-display description: VGA compatible controller product: MGA G200eH3 vendor: Matrox Electronics Systems Ltd. physical id: 0.1 bus info: pci@0000:01:00.1 logical name: /dev/fb0 version: 02 width: 32 bits clock: 33MHz capabilities: pm msi pciexpress vga_controller bus_master cap_list rom fb configuration: depth=32 driver=mgag200 latency=0 resolution=1024,768 resources: irq:17 memory:d8000000-d8ffffff memory:d9b98000-d9b9bfff memory:d9000000-d97fffff memory:c0000-dffff
Spice uses gpu, RDP uses cpu. So, maybe if i change my primary gpu with some powerful one it can effect spice in a good way
What do you think about that? _______________________________________________ Users mailing list -- users@ovirt.org To unsubscribe send an email to users-leave@ovirt.org Privacy Statement: https://www.ovirt.org/privacy-policy.html oVirt Code of Conduct: https://www.ovirt.org/community/about/community-guidelines/ List Archives: https://lists.ovirt.org/archives/list/users@ovirt.org/message/HYEKRGLHSCXXQS...
ı've checked my network usage and it's ok. No problem. But, there is another problem which we've already knew. Our storage I/O is bad. It might be the some part of the problem. I know this is getting out of the main subject. But, one thing i want to learn if some of you can help help in os.conf files there ar 2 values os.windows_10x64.devices.display.vramMultiplier.value os.windows_10x64.devices.display.vgamemMultiplier.value i've created override con file to change these, restart engine and install new w10x64 But, does not effect vram value. is's still 4GB shared memory How can i increase the vgamemory to test if makes better performance or not? Thanks
participants (4)
-
Alex K -
carl langlois -
Gilboa Davara -
ozmen62＠hotmail.com