VMworld Archive

At VMworld I attended the excellent “Networking Deep Dive” (TA2525) by Srinivas Neginhal.  Here are just a few things he discussed that jumped out me as either new information (not in the standard documentation) or something that expanded on my current understanding.

ESX hosts keep a local cache of the vDS and DVPort information, to use when vCenter is unavailable (both are non-editable binary files):
Host DVPort state:    /etc/vmware/dvsdata.db
VM DVPort state:      /vmfs/volumes/<storage>/.dvsData

DVPort binding types:

• Static – assigned to the virtual adapter and immediately pushed to the host. Written to the host’s cache and written in the VM’s vmx file.
• Dynamic – assigned when the VM is powered on, and then pushed to the host. There is no guarantee that the VM will get the same DVPort on the next  power on.  However it uses a concept similar to DHCP in that if the same port is available then it will renew that one.
• Ephemeral – a new port on every power-on. The port is destroyed when the VM disconnects from the port. Essentially this is what happened on all vSwitch ports on ESX 3.x and lower.  This is the only option if you are connecting the client directly to the host, vCenter is required to create the other 2 types of binding.

He also explained the Traffic Shaping terminology, which seems obvious in retrospect but was something that I hadn’t “got”.  Simply put, its all relative to the switch.  So Ingress traffic is stuff coming in, whether from a VMNIC (physical adapter) or a vNIC (VM’s adapter).  Egress traffic is leaving the switch.

And one last thing which stood out was a table he used to quickly show some of the advanced vNIC driver capabilities.  I’d not seen this laid out in this way, and is much more specific than the standard documentation (in fact if this is correct, which I’m sure it is, then the current documentation is downright misleading – N.B I need to update my vSphere reference card with this).  It also makes it much easier to remember.

TSO – TCP Segmentation Offloading
RSS – Receive Side Scaling http://www.microsoft.com/whdc/device/network/NDIS_RSS.mspx
MSI/MSI-X – Message Signaled Interrupts http://blogs.sun.com/gnunu/entry/how_does_msi_x_work

Update: VMware has just released this performance whitepaper discussing the merits of the new vmxnet3 driver.

I also hit “vSphere4 Storage Troubleshooting and Log Analysis” (TA1394) which was presented by Mostafa Khalil at VMworld this year.  Wow, he was a wealth of information.  Unfortunately he was pushed for time, and I had to leave this one early to get to my next session.  If a video of this ever popped up on youtube I think I’d watch it at least 5 times through.

Anyway, one thing he introduced which I thought was rather nifty was a new element to the Event Handling.  Now in vCenter, when certain events occur, the event’s description givess a hyperlink which takes you directly to an associated KB article explained what happened, why and what you might need to do to resolve things.  Currently this is limited to the 16 most common events, but they intend to add more links along with vSphere updates.

Since coming back from VMworld this year, I have been hunting around for the dates for next year.  I had a lot of fun, met some really interesting people and of course improved my knowledge of the products.  However simple google searches weren’t turning up the actual dates, just that it would be sometime in September.  All the top hits were about the news that VMworld Europe was changing its dates next year.

Well I was reading through some of my notes last, and just as I closed the notebook there was the answer.  The dates are on the back of this year’s red VMworld booklet   So before I forget what the dates are again, they are 30th Aug – 2nd Sep 2010.

On Wednesday I attended the VMFS3 session entitled “VMware vStorage VMFS-3: Architectural Advances since ESX 3.0″.  It was a very interesting little session discussing some of the under-the-covers work through the evolution of VMFS3.

However, why am I blogging about it now?  Well one thing jumped out at me during the presentation.  The speaker (I think it was Satyam Vaghani), introduced a slide which showed the testing that they were doing on the next implementation of VMFS3 (3.45).  To be honest I can’t remember the exact detail of the slide (and unfortunately the uploaded version currently on the VMworld site doesn’t have the same slide), but what I do recall is the upper limits of the testing.  The slide was showing the effects of new locking mechanisms.  It showed the performance of 8 hosts connecting to 512 VM on 1 VMFS LUN! Holy smoke. Ka-Pow.  Imagine being able to do that.

What sort of effect would that have on your datastore provisioning?  Time to re-think your VDI solution?  I can’t wait