VMdamentals.com

Today Eric delived a great presentation on vSphere troubleshooting. All resources were touched: CPU, Memory, storage and networking.

In the CPU field Eric touched on CPU ready time, which remains the number one metric to check in case of suspected CPU troubles. A good reminder on the ms <--> percentage change in the newer ESX version (see one of my old blogposts: Performance tab “Numbers” demystified on how to get back to your trusted %ready from the VIclient performance tabs instead of esxtop.

On the memory side Eric touches the vague effect that seems to plague some environments were memory limits seem to appear automagically; VMs then get a limit equal to their own memory size. This is not particularly a problem, right until you increase a VM’s memory: ballooning is the result, and ballooning will make sure the extra memory you just gave the VM will not be utilized! I have seen this as well; somehow the limits appear. Gotta dive into that one someday 🙂

On the storage side, Eric claims this is THE number one reason for performance issues nowadays. Hey, been there, done that in the VDI / VMware View envrionments! It is THE one thing you need to consider when designing and scaling your solutions.

Also, Eric has one of my fun 3D vscisiStats stuff included in his presentation! See http://vmdamentals.com/?p=722 or vscsiStats 3D surface graph part 3: Build your own! to build your own like these:





On the networking front, Eric touches on receive buffers, dropped packets and the long way any packet will have to travel. Also VLAN trunking can help. You can then deliver the network data “tagged” straight into the VM; You could otentially use multiple VLANs without the requirement to route between VLANs. Great idea!

Also “load based teaming” is a nice feature to spread your network load more effectively. As soon as a link gets saturated, VMs are migrated to other physical uplinks. At last, a teaming policy that looks at actual network load without the trouble of having to inspect each IP packet to determine the uplink (and even that is no guarantee to spread your load effectively!)

at labs.vmware.com you can download ESXPLOT. Cool way to simplify longterm monitoring!

Another cool tool is RVTools, available at www.robware.net and allows you to read through your vmx files and locates old/unused VMs on your storage.

Thrid and last is the tool XANGATI, available through xangati.com which was also used at VMworld to monitor the labspace there!

4 Responses to “Dutch VMUG: Eric Sloof – vSphere performance troubleshooting”