Below are answers to questions asked during the recent webinar. If you have additional questions, please contact Dean Richards at deanrichards@confio.com.
Q: What would you recommend regarding the storage layout for the VM, i.e. should we configure 1 DB server per LUN?
A: Storage layouts will differ for each VM running a database, just like they do in the physical world. We configure one database per VM to provide more flexibility, but then configure the storage based on that database. Based on I/O and requirements you may share a LUN with many other VMs, e.g. a dev/test and other less critical VMs. Sometimes you may have many LUNs for one VM and no other VM will share those LUNs. So, the answer is that it depends on I/O workloads.
Q: You mentioned that if there are 1000+ concurrent users that you would not virtualize that environment. What is the reason for that?
A: The decision to virtualize an environment should not be based on the number of concurrent users alone. It should be based on the resource consumption of an environment, e.g. storage IOPS, CPU utilization, etc. However, in practicality, an environment with a large number of concurrent users would most likely also use a lot of resources. When beginning a virtualization project, start with the database environments with less resource consumption and build up to an environment like this. You will not only learn many things along the way to help make the project more successful, but you also lessen the risks dramatically.
Q: Is the MAC address the same on Fault Tolerance VMs?
A: Yes. When FT is configured for a VM, a shadow of that VM is created on another physical server. These instances are kept in “lockstep” with each other and access common storage with a single IP and MAC address. Here is a link to a document on the VMware website with information about how FT works:http://www.vmware.com/files/pdf/VMware-Fault-Tolerance-FT-DS-EN.pdf
Q: Can I use OEM (Oracle Enterprise Manager) to monitor my VMware environment?
A: OEM does provide a plug-in for monitoring VMware. However, I personally do not see enough advantages of using this plug-in over the free tools provided via the vSphere GUI on top of a vCenter server. When monitoring databases (Oracle, SQL Server, Sybase, DB2/LUW) running in a VMware environment, Confio’s IgniteVM provides a “single pane of glass” view from the database to the VM down through the physical ESX host and storage layers. Everything you need can be found on a single page which is very unique in the industry.
Q: Is swapping a feature that must be enabled in order to use?
A: No, hypervisor swapping is an inherent feature of VMware used as a last resort when a physical host is under severe memory pressure. Page sharing and ballooning are techniques used to reclaim memory before swapping because of the severe performance degradation that can occur. Here is a link to a document with more information about memory management: http://www.vmware.com/files/pdf/techpaper/vsp_41_perf_memory_mgmt.pdf
Q: Can you define ballooning and what it means?
A: The VM is running a guest O/S like Windows or Linux, and that guest O/S does not know it is running in a virtual environment. Because of that, if the physical host that contains the VMs is running low on memory, the guest O/S will not know this and not free memory up to help. Ballooning helps the guest O/S understand the hosts memory shortage. A balloon driver is installed as part of VMware Tools into the guest O/S and polls the ESX server to get a target balloon size. If the ESX host needs memory, it will post a target balloon size for the balloon driver making it inflate. The hypervisor will see the memory pages pinned by the balloon driver and reclaim that memory. A more thorough description can be found in the vsp_41_perf_memory_mgmt.pdf document reference above.
Q: If we use OS monitoring (CPU , I/O, Memory) on a physical host that has 5 VMs, what will be the accuracy of those OS reports?
A: I assume you have ESX/VMware 4.1 or lower installed and you have ESX running on top of Linux. In that case, the Linux O/S counters should be a good indication of memory and other resource pressure on the physical host. However, note that with ESX 5.0 and higher, the only option is to use ESXi installed directly on the physical host, so the Linux O/S is no longer present. Your O/S counters will no longer work.
Q: In our testing we have seen dropped pings during vMotion. Could you comment on that?
A: I do not have any specific experience with this, but searched for others having issues. It seems that this was more of an issue in earlier versions of VMware. However, when the vMotion is performed, it may take just a little time for everything to get into place, so I could see a failed ping or so occurring. I would not expect to see many occur though, so if that is happening, a call to VMware support may be in order.
Q: Can vSphere send scheduled reports and custom alerts?
A: To my knowledge, the vSphere GUI does not accommodate sending reports on a defined basis. However, many people on the internet have example of scripts that can probably accomplish what you need. There is the concept of a scheduled task in vSphere, but sending reports is not one of the things you can do.
Q: Can you recommend the paravirtualized disk controllers? You did not mention them.
A: Yes, if you have VMs that drive require a lot of I/O throughput, which databases are certainly candidates to do, you should use the paravirtual SCSI (PVSCSI) adapters. These are best suited for environment, especially SAN environments, where applications drive a very high amount of I/O. This is something I will include in my presentation in the future, so thank you for reminding me about it.
Q: We use the VM memory active perfmon counter and monitor for it in SCOM. Are the VM counters in perfmon good to use?
A: Yes, these are supplied by the VMware environment are ok to use. Standard perfmon counters can be incorrect and misleading, but these should be ok. However, if you only use the VM perfmon counters, you will be missing he physical host, storage and other metrics.
Q: As per VMware best practices for SQL Server DB’s they suggest use RDM with specific RAID?
A: I am not aware of a suggestion to use RDM (Raw Device Mapping) with a specific RAID type. My understanding is that any type of RAID can be used underneath a raw device, but I am not a storage expert by any means. The same proven practices you are currently using for SQL Server should also translate to the virtual environment.
Q: Does vMotion/DRS have any overhead on the VMs performance?
A: Not that I am aware of, and it probably only helps. If a physical host is running low on resources while another physical has plenty to spare, vMotion/DRS will help balance the workload for you.