New on LowEndTalk? Please Register and read our Community Rules.
All new Registrations are manually reviewed and approved, so a short delay after registration may occur before your account becomes active.
All new Registrations are manually reviewed and approved, so a short delay after registration may occur before your account becomes active.
★ VirMach ★ RYZEN ★ NVMe ★★ $8.88/YR- 384MB ★★ $21.85/YR- 2.5GB ★ Instant ★ Japan Pre-order ★ & More
This discussion has been closed.
Comments
Do you want me to post "work order" for you?
Here we go: #195517
Now give us some week or two break, please.
The better way is no fees and all ports are open I think. Definitely for existing customers.
I dont know if you mean the setup on server (Linux) or including switch (hardware). VLAN / PVLAN are configured on switch side, not software. SolusVM did not take over VLAN configure, it just group IPs according to the IP assignment you did on switch.
If QN turned on PVLAN for your Nodes' switch, it will isolate broadcasting between each Node, thus the broadcast storm wont happen. However on your side (Linux/SolusVM) there're no different.
That's one way to split big VLAN, you can also do VLAN split in the old way however you'll lose flexibility on IP assignment. Another way is to do software VLAN split using Open vSwitch, as I know SolusVM did not integrated this feature, you have to do it yourself.
Packet loss are caused by IRQ (Interrupt Request), when htop on Node root you'll see a CPU core always near its 100%, that's what used to handle network requests, it's overload so it begin to drop packets. Big VLAN > Broadcast Storm > Interrupt Request > Packet Loss
If you confirm this is not a VLAN issue, then DDoS / Coin mining may cause the same symptom. Just ignore this post pls.
Well it works before however they produced some bugs:
VM947:1
Uncaught TypeError: Cannot read properties of undefined (reading 'click')
Wait until they fix this or find another provider may be faster.
PS. Each IP change cost $5 before, if you always change IP, you'd better find some monthly purchased VPS provider instead...
Why would there be a single CPU core processing all the network interrupts?
The hypervisor is supposed to create PCI virtual functions on the Ethernet adapter, and then passthrough a virtual function to each virtual machine.
Then, network interrupts only occur inside the virtual machine, on the CPU core assigned to the virtual machine.
Actually I don't know why but it happened when we have a large amount of VPS in one VLAN before. I got your point, however it occurred, network interrupts occupied 100% of one core on the server Node (Not VPS).
I think the it might be:
Vir Nodes uses single-queue NIC hardware, which can only use single core process IRQ.
They uses old Linux kernel (<2.6.21), does not support multi-queue even the NIC has.
SMP IRQ affinity turned off.
Whatever, even above balanced interrupts to multi cores, it may reduce packet loss but do not solve the root issue, including abnormal traffic income (ARP traffics). A good network should have correct VLAN isolation, test using:
The result should below 100 on each VPS, if above there might be some issue on network design.
Also I've tried irqbalance and manually bind cores on problem Nodes before, no use at all, until I found all issues come from big VLAN in final. Just want Vir avoid detours when I saw the same situation occurred on them. If he check the Nodes and find the same situation then it could be a reference.
Single-queue NIC is for weaklings.
Premium hosting providers should use Intel XL710 or Mellanox ConnectX-5.
40 queues per NIC.
Install multiple NIC if you want to have more than 40 virtual machines.
SR-IOV for everyone.
VLAN is for weaklings too.
Use L3 switch.
Make I-BGP between every host node and the L3 switch.
Block all broadcast traffic.
Traffic comes to a NIC only if it's announced, and then PCI virtual function delivers packets straight to the virtual machine.
Hairpin for traffic between virtual machines on the same host node.
Tried this on a couple of VPS's in Tokyo, and the average is like ~1000 (sometimes up to 1400) rxpck/s, while on an older E5 node, the average is <20. And if the avg. keeps at 800~900, the network remains stable, but once it reaches >1000, ping spikes happens.
You're the expert, any idea on how to solve current issues economically?
I wish my tokyo VPS get rid of this soon: https://imgur.com/PVhdxH9
Quite confirm it's not only caused by abuse.
Move to Oracle Cloud instead.
They have less than 1% packet loss.
Another thing bothers me is AES-NI has been disabled by default.
@VirMach Should I open a ticker for changing the CPU to Host-passthrough?
Noooooooooooooo
when is IPV6 available? @VirMach
One push-up per address.
Minimal request is /96 subnet.
2^(128 - 96) = 2^32 = 2G, fair enough
life long commitment I suppose
I can't even see the button to change ip.
Placed an order over a week ago, paid with account balance. The new vps has been "pending" since and support has been completely unresponsive. So unresponsive that I made a new ticket and this time made it "high priority" only to have support merge the tickets and still not even respond!
this price for one vps or all vps in account ?
New IP change option added. "Ryzen Fix IP" for all Ryzen nodes. You can use this once to either [A] fix a problematic IP on Ryzen, or [B] to re-roll on a new subnet that's more closed off and potentially with better networking and less problems. So you can use this button if you have bad routing. May or may not improve networking.
If enough people use this button I think we should get them more like Los Angeles networking and it'll save us time trying to fix this. We do have it all worked out to migrate later for the rest.
As with everything else, use at your own risk and please provide feedback so we can fine-tune it.
Placed an order over a week ago, paid with account balance. The new vps has been "pending" since and support has been completely unresponsive. So unresponsive that I made a new ticket and this time made it "high priority" only to have support merge the tickets and still not even respond!
No button appears for SJCZ006, although I wouldn't need it for now.
With re-roll button, it's feeling like ISP now.
Every week I change the MAC address on the home router, so that FiOS gives me different IP.
When I had ADSL, I redial the modem every week too.
People on TYOC029 should click that button.
Next time I log into panel (too lazy to find 2fa now) will click that button!
I've logged in, I don't see any "Ryzen Fix IP" or "Change IP" button. Not on BF specials, not on SSD1, not on normal service. I guess was I put on naughty list
@VirMach
Could you review my service id: 662525. I just dit ssh to it one time to test account information. I have not used it because see to many issue here. It still is one original ISO comes from you.
But, at the weekend, you suspended (automatically) my service with title: Phishing site/abuse.
Could you tell me the reason my VPS is suspended?
That's why I prefer to power it down in case I don't have time for initial setup and security hardening.
Keeping it in vanilla state is like waiting for something bad to happen.
The IP change I've been waiting for is finally here, but I can't use the change IP button, the button doesn't work, my node tokyo 40
**When can I connect to my server? pending order -> action=productdetails&id=670351 **
Yes, predicted twice, now I predict ETH pump to 2K in July.
No ip change sjcz002.