What is the biggest attack in GBPS you stopped
-
With esxtop running
http://youtu.be/_dimZ1_DO_o
-
With esxtop running
http://youtu.be/_dimZ1_DO_o
Your wait time is insane on the hypervisor.
The way ESXi works is that it needs all 8 cores free before it will allow the CPU in your VM to process data. So if you've over-subcribed CPUs, your wait time goes through the roof while the VM waits for all 8 cores to become free to the VM; hence it's waiting for free cores to process data.
You can impair a VM considerably because of wait time and CPU availability. It's one of the reasons why Oracle wants you to test their products on iron and will not support you if you're not using their hypervisor. If you aren't aware of how wait times work at the hypervisor level, it'll bite you in the ass.
In this case perhaps the hypervisor kernel is using multiple cores on the CPU to accept data from the NIC. As you hammer that NIC, CPU wait times go up because the hypervisor kernel has a higher priority than your VM, so your VM is left waiting.
When we tested FreeBSD and CentOS, my hypervisor was crushed. It wasn't readily apparent if the issue was with the hypervisor or the VM, but I believe it was a combination of the two.
If you can, remove the hypervisor layer so all you have to deal with is the hardware and BIOS.
-
Its accumulated idle time. Not the actual VM waiting…
Run, %RUN:
-
This value represents the percentage of absolute time the virtual machine was running on the system.
-
If the virtual machine is unresponsive, %RUN may indicate that the guest operating system is busy conducting an operation.
-
When %RUN is near zero and the virtual machine is unresponsive, it means that the virtual machine is idle, blocked on an operation, or is not scheduled due to resource contention. Look at other values (%WAIT, %RDY, and %CSTP) to identify resource contention.
-
When %RUN is near the value of the number of vCPUS x 100%, it means that all vCPUs in the virtual machine are busy. This is an indicator that the guest operating system may be stuck in a operational loop. To investigate this issue further, you may need to engage the appropriate operating system vendor for assistance in identifying why the guest operating system is using all of the CPU resources.
-
If you have engaged the guest operating system vendor, and they have determined that the issue is caused by the VMware Tools or the virtual machine hardware, it may be pertinent to suspend the virtual machine to collect additional diagnostic information.
Wait, %WAIT:
-
This value represents the percentage of time the virtual machine was waiting for some VMkernel activity to complete (such as I/O) before it can continue.
-
If the virtual machine is unresponsive and the %WAIT value is proportionally higher than %RUN, %RDY, and %CSTP, then it can indicate that the world is waiting for a VMkernel operation to complete.
-
You may observe that the %SYS is proportionally higher than %RUN. %SYS represents the percentage of time spent by system services on behalf of the virtual machine.
-
A high %WAIT value can be a result of a poorly performing storage device where the virtual machine is residing. If you are experiencing storage latency and timeouts, it may trigger these types of symptoms across multiple virtual machines residing in the same LUN, volume, or array depending on the scale of the storage performance issue.
-
A high %WAIT value can also be triggered by latency to any device in the virtual machine configuration. This can include but is not limited to serial pass-through devices, parallel pass-through parallel , and USB devices. If the device suddenly stops functioning or responding, it can result in these symptoms. A common cause for a high %WAIT value is ISO files that are left mounted in the virtual machine accidentally are either deleted or moved to an alternate location. For more information, see Deleting a datastore from the Datastore inventory results in the error: device or resource busy (1015791).
-
If there does not appear to be any backing storage or networking infrastructure issue, it may be pertinent to crash the virtual machine to collect additional diagnostic information.
Ready, %RDY:
-
This value represents the percentage of time that the virtual machine is ready to execute commands, but has not yet been scheduled for CPU time due to contention with other virtual machines.
-
Compare against the Max-Limited, %MLMTD value. This represents the amount of time that the virtual machine was ready to execute, but has not been scheduled for CPU time because the VMkernel deliberately constrained it. For more information, see the Managing Resource Pools section of the vSphere Monitoring and Performance Guide or Resource Management Guide.
-
If the virtual machine is unresponsive or very slow and %MLMTD is low, it may indicate that the ESX host has limited CPU time to schedule for this virtual machine.
Co-stop, %CSTP:
-
This value represents the percentage of time that the virtual machine is ready to execute commands but that it is waiting for the availability of multiple CPUs as the virtual machine is configured to use multiple vCPUs.
-
If the virtual machine is unresponsive and %CSTP is proportionally high compared to %RUN, it may indicate that the ESX host has limited CPU resources, simultaneously co-schedule all vCPUs in this virtual machine.
-
Review the usage of virtual machines running with multiple vCPUs on this host. For example, a virtual machine with four vCPUs may need to schedule 4 pCPUs to do an operation. If there are multiple virtual machines configured in this way, it may lead to CPU contention and resource starvation.
-
-
Its accumulated idle time. Not the actual VM waiting…
I disagree. Try dropping the cores down to two and see if that wait time drops.
I actually had a conversation with VMware about this yesterday. Our SQL farm is getting crushed by wait time due to CPU over-subscription. We're about to start a VM right-sizing exercise to address the issue.
-
%RUN = 290.51 (could be 400%) so it means not all CPU's are busy
%WAIT=408.70 (Could be my NAS at home where the VMresides, Synology 1813+)
%RDY=0.18 (% time a vCPU was ready to be scheduled on a physical processor but couldn’t due to processor contention. Threshold: 10% per vCPU)
%MLMTD=0.00 (% time the VM was ready to run but wasn’t scheduled because it would violate the CPU Limit set. Threshold: 0%)
%CTSP=0.00% (% time a vCPU in an SMP virtual machine is “stopped” from executing, so that another vCPU in the same virtual machine could be run to “catch-up” and make sure the skew between the two virtual processors doesn’t grow too large. Threshold: 3%)It doesnt indicate that the hypervisor subsystem suffers in any way, but storage wise(maybe).
-
Yes, not all CPU will be busy, but all CPUs allocated to a VM must be present for the VM to compute.
For example, if you have 16 cores and 12 are busy, and a VM with 8 cores comes along, it will wait until 8 cores are available to process data. It cannot use the 4 cores available in the example above, it must find 8 available cores and then compute whether or not it needs all 8 cores. It's an allocation model. So CPU contention can create excessive wait times while CPU utilization is very low.
-
Yes but it didnt :)
I can reissue 8 cores and test again. :)
-
Even with 8 cores it doesnt get past 0.33 %CSTP :)
-
I rebooted the server behind that receives the traffic while attacking.
Packet loss still occured and then I shut it down completely.
In the end you see the traffic settle in on around 7.5mbit/s and become a straight line.
Then packets begin to flow and CPU4 settles below 100%.
What you reckon Tim?
-
This is the IOPS I see on the storage attached as NFS during an attack.
-
What you reckon Tim?
I reckon it's going to be very complicated troubleshooting an apparent kernel issue in FreeBSD while it's running on a hypervisor.
-
Yes, not all CPU will be busy, but all CPUs allocated to a VM must be present for the VM to compute.
For example, if you have 16 cores and 12 are busy, and a VM with 8 cores comes along, it will wait until 8 cores are available to process data. It cannot use the 4 cores available in the example above, it must find 8 available cores and then compute whether or not it needs all 8 cores. It's an allocation model. So CPU contention can create excessive wait times while CPU utilization is very low.
Correct. https://communities.vmware.com/message/2275523
IF anything you'd be better off reducing the number of cores a VM needs as minimum rather than give it the most it can use, as the overhead switching or timeslicing at the host (ESXi) level can be further reduced just like sometimes its faster to run software single threaded on a single core than a multithread app across all cores, as the latter introduces more locks & overheads at the sw level as well as the cpu level not to mention having to share the bus to devices like the hw, nics or ram.
Besides running alot of OS tended to be faster as a VM instead of bare metal, eg I can install W7x32 as a VM in just 6mins 23 seconds from hitting play on the VM for the first time, various reboots, assigning username & password, updates settings all the way to being on the desktop for the very first time. I might be able to shave a few more seconds off by not running other VM's at the same time, but I cant do that time installing W7x32 on the same bare metal machine which is a testament to VMware.
-
Wait, %WAIT:
-
This value represents the percentage of time the virtual machine was waiting for some VMkernel activity to complete (such as I/O) before it can continue.
-
If the virtual machine is unresponsive and the %WAIT value is proportionally higher than %RUN, %RDY, and %CSTP, then it can indicate that the world is waiting for a VMkernel operation to complete.
-
You may observe that the %SYS is proportionally higher than %RUN. %SYS represents the percentage of time spent by system services on behalf of the virtual machine.
-
A high %WAIT value can be a result of a poorly performing storage device where the virtual machine is residing. If you are experiencing storage latency and timeouts, it may trigger these types of symptoms across multiple virtual machines residing in the same LUN, volume, or array depending on the scale of the storage performance issue.
-
A high %WAIT value can also be triggered by latency to any device in the virtual machine configuration. This can include but is not limited to serial pass-through devices, parallel pass-through parallel , and USB devices. If the device suddenly stops functioning or responding, it can result in these symptoms. A common cause for a high %WAIT value is ISO files that are left mounted in the virtual machine accidentally are either deleted or moved to an alternate location. For more information, see Deleting a datastore from the Datastore inventory results in the error: device or resource busy (1015791).
-
If there does not appear to be any backing storage or networking infrastructure issue, it may be pertinent to crash the virtual machine to collect additional diagnostic information.
You had an 800%-900% wait time in the video you posted. Holy crap, that's insane! See the part above that talks about latency. This is a smoking gun!
I've been very consistent when I've said that testing pfSense on a VM isn't going to give you tangible data because you'll also need to troubleshoot the hypervisor layer at the same time.
Please, please, please stop wasting your time testing this issue on a hypervisor. Put pfSense on bare metal and test it there.
-
-
I couldn't agree more.
Please, please, please stop wasting your time testing this issue on a hypervisor. Put pfSense on bare metal and test it there.
-
I couldn't agree more.
Please, please, please stop wasting your time testing this issue on a hypervisor. Put pfSense on bare metal and test it there.
Ditto
Too much shot-gunning going on in this effort instead of a methodical systematic approach.
Begin with minimalist install / config (bare metal, no packages, no services, etc.) and work up to point of failure.
If the minimalist install / config fails then go back even further to either earlier pfSense versions, or better yet to FreeBSD itself until the issue does not exist. Then systematically move forward adding to that config until the issue appears.
-
Jun 4 09:28:36 check_reload_status: Reloading filter
Jun 4 09:28:36 check_reload_status: Restarting OpenVPN tunnels/interfaces
Jun 4 09:28:36 check_reload_status: Restarting ipsec tunnels
Jun 4 09:28:36 check_reload_status: updating dyndns Yousee
Jun 4 09:06:36 check_reload_status: Reloading filter
Jun 4 09:06:33 check_reload_status: Syncing firewall
Jun 4 09:05:08 check_reload_status: Reloading filter
Jun 4 09:05:06 check_reload_status: Syncing firewall
Jun 4 09:01:12 kernel: em0: promiscuous mode enabledWhen this happens the firewall encounters packetloss.
Disabling promiscious mode on em0 and em1 solves it and makes it endure.
Running CRON job every 60 seconds will make you survive a SYN flood.
-
So whats the number of CPU's and cores available?
What VM's are you running and how many CPU's/Cores do each need as a minimum?
Possibly the best way to visualise this is a bit like a game of tetris but various sized horizontal sized blocks which are your minimum core requirements for the VM OS.
If you have a 4 core baremetal then your Tetris game is just four blocks wide, if its a 2 cpu, 8 core, the your Tetris game is 16 blocks wide.
When the blocks reach the base, thats your hypervisor's timeslice to work on the physical cpu's/cores.
Your task is to try and fit as many horizontal blocks representing the core requirement of your VM OS's to use up all the space. This is why its best to run VM's as a minimum sized cores/vCPU's in most cases.
eg say you have 4 OS's and 1 cpu/4 cores.
OS1 needs 2 cores min but can use 10 cores max.
OS2 needs 4 cores min but can use 8 cores max
OS3 needs 1 core min but can use 4 cores max
OS4 needs 1 core min but can use 2 cores max.Whats the most efficient way to set these OS's up as VM's in order to maximise the time slices for each?
If you went:
OS1 4cores
OS2 4cores
OS3 4cores
OS4 2coresWhen ever OS4 had its time slice you waste 2 physical cores.
This approach would also make for a clunky setup because if any OS needs to talk to another, then you have no two OS's running in the same time slice to communicate with each other, they will all have to wait 4 time slices to before they can get back and process their stuff.
If you went
OS1 2cores
OS2 4cores
OS3 1cores
OS4 1coresThen OS3 & OS4 can run when ever OS1 runs so you now have a "block" where you in effect have OS1, OS3 and OS4 that can operate in one timeslice and OS2 can operate in another timeslice. ESXi only has to swap between two different blocks of OS's (OS1,OS3 & OS4) and OS2, which makes for a more responsive setup as the OS's in the first block can communicate between themselves if need be in teh same timeslice and so the only extra wait time for any communication is swapping from block 1(OS1, OS3 & OS4) to block 2(OS2) which is running OS2.
Does that make sense and easier to understand?
Its more complicated than that because next you have RAM requirements to consider as well, but a similar principle exists, ie if you err towards less ram, ESXi doesnt have to spend time loading and unloading ram for each VM running in the timeslice. Its best to have the ram requirements for each VM' fit within the physical ram.
So if you take the 2nd example above.
OS1 2cores
OS2 4cores
OS3 1cores
OS4 1coresYou have lets say 32Gb of physical ram.
OS1 can use 8GB to 32GB
OS2 can use 4GB to 16GB
OS3 can use 2GB to 4GB
OS3 can use 4GB to 16GBOS1 2cores/32GB
OS2 4cores/16GB
OS3 1cores/4GB
OS4 1cores/16GBThen even though the first block(OS1,OS3&OS4) can share the physical CPU's, they cant share the physical as you would need 42GB of physical ram.
But if you went
OS1 2cores/16GB
OS2 4cores/16GB
OS3 1cores/4GB
OS4 1cores/8GBThen the first block (OS1,OS3&OS4) can share the physical ram as the total amount is 32GB, and the 2nd block(OS2) will use 16Gb with 16Gb going spare doing nothing.
If you notice I gave OS1 16GB, this is because even though OS4 can also use 16GB its only got 1 core.
However you then also need to look at what the tasks are that are going to be running on each VM.
Databases love ram, the more ram you have the more you can load the DB into ram already sorted into the most popular views that users use the most.
MS Exchange is similar ie its just a big DB but it offloads alot of its work to the workstation so Outlook will often have a copy of what is stored in Exchange so outlook only has to access the local disk, but Exchange will have lots of connections like keep-alive open for smart phones so that it can "push" emails & other things to the phones.
Webservers depends on what they are doing, some maybe front for DB's running on the same VM or maybe not, but hopefully that will give you a better overview of whats going on at a lower level. :)
-
I have divided the 2 running VM's here at home on 2 different sockets with 4 cores each.
Didnt matter at all until I ran the cron job disabling promiscous mode.
Its in the reject state on the hypervisor Vswitch allready….
-
I couldn't agree more.
Please, please, please stop wasting your time testing this issue on a hypervisor. Put pfSense on bare metal and test it there.
IT ISN'T BETTER ON BAREMETAL. Problem still exist. I tried several times on my bare metal supermicro.
Read the thread and the other threads again. You will see the history.- i am though not using pfsense anymore. So i can no more test
-
I have divided the 2 running VM's here at home on 2 different sockets with 4 cores each.
Didnt matter at all until I ran the cron job disabling promiscous mode.
Its in the reject state on the hypervisor Vswitch allready….
You know about promiscuous mode can mess up some nics? You'll see this note in the packet capture amongst other places.
What are your System:Advanced:Networking, Networking tab, Network checkboxes set to?
How did you install the VM?
Did you install from iso on the ESXi server or setup pfsense on a different bare metal/VMware host like VMware workstation, cloned it then moved it to the ESXi server in question?pfsense is one vm, whats the other VM?
