Network? performance issue with HPVM

[Bill Costigan]

I think I have an issue with VMs communicating between themselves with TCP/IP.

We have some HPVMs on an HPUX hosts. They are all connected to the same vSwitch. This is an SAP implementation with the Oracle DB running on one VM and the SAP batch application on the other.

After running various performance tests we discovered that the batch process is handled primarly by two processes. One oracle process and one SAP process. The SAP process requests data from the oracle process and then processes that data and asks for more.

Running both processes in the same VM shows that either one process or the other is running. If you add up the CPU utilization of both processes it equals 100% or a single CPU.
This seems to indicate that both are single threaded and they do not run simultaneously.
Now if we move the SAP process to a second VM on the same host and have them communicate through the vSwitch using TCP/IP the overall performance drops quite a bit. Adding the CPU utilizations together we only see 75% of a single CPU. The duration also increases a corresponding amount of time.

So the question is where does the other 25% of the time go? Before moving to VMs, the two processes were running on separate HPUX servers connected through a physical Ethernet switch. We did not experience this lost time with that set-up.

[Olivier Masse]

Are you using the AVIO lan driver on both VMs? If not, try that as a first step.

[Bill Costigan]

Yes, I'm using avio_lan for the network interface.

One more detail, the waits that I'm assuming are tied to the network are IPC waits. When both processes where on the same server the waits were SOCKT waits.

[Tonny Sejr Kromann]

Hi, I'm experiencing about the same, but we have both SAP and Oracle within the same virtual machine. Everything within the machine is fine. Heavy disk i/o is ok, but when Oracle creates many logs, we trigger dataprotector to back them up. This will happen over network and causes the virtual machine to be more or less unresponsive till the network traffic has finished. It does not help to run the backup over lan in only one thread.
This is HPVM4.0 on 11.31/0809, and I'm planning to install PHNE_36857/PHNE_38689 to see if this helps.

--
Regards
Tonny

[Origin Energy]

Hi

We are also facing the same problem,when ever the dataprotect is trying to backup then the servers are hanging and we are forcebly restarting the servers through HP CM console.

we have installed SAP CRM DB/CI on one virtual machine and SAP PI DB/CI on another virtual machine but both are running on single Base OS(HP BL870c - 8CPU/32GB RAM. We have logged a call with HP Global support and they have recommended the following patches and we are going to deploy these patches; let's wait ...

Install the Patches.
PHSS_39591
PHKL_40441

Thanks
Venugopal
+61 402161730

[Bill Costigan]

Thank you.

We are on 4.1 so the first patch doesn't fit. However the vm patch might be a good idea. HP suggested a vm patch but it was the one pervious to the one you listed.