latoga labs

One item that I have seen come up again and again recently with my customers is a mis-understanding of ESX and VM Server.  Particularly the performance aspects of both.  These are two entirely different products with entirely different goals that need to be kept in mind when comparing the two.

VM Server is the free server virtualization product from VMware.  It’s purpose was to help seed the market and get people more confortable and familiar with virtualization concepts on the server.  What better way to do that than give something away for free for people to play with!  VM Server is similar to VM Workstation/Fusion in that it runs on top of an OS; so from the ground up you have:

server hardware > operating system > VM Server > virtual machine.

This means VM Server inherets the same performance issues that the OS has.  It also means that it has to rely on the OS for a lot for the management of resources.  Net result:  the performance of VM Server will be impacted by the OS you are running it on top of.

ESX is the data center grade hyperadvisor from VMware.  It’s purpose is to provide virtualization for  business critical data centers.  ESX essentially doesn’t run on top of an OS.  You can think of it as a data center OS as it is the base layer for the Virtual Infrastructure set of products. Similarily, from the ground up you have:

server hardware > ESX > virtual machines

You can see immediately that ESX has eliminated the operating system level, this is the source of the performance gain that ESX provides.  It is running on top of the raw hardware.  It is truely a specialized OS with the sole pupose of virtualization.

One thing that VM Server and ESX share, is the engine to run a virtual machine.  This engine has lots of tricks built into it to manage the physical resources of the box and to distribute them efficiently to the virutal machines that need them.  Of couse, you still suffer from that extra piece of the base operating system under VM Server.  This is why you can have VM Server running with no VMs and still see it utilizing resources.  It’s doing the work of interfacing with the base OS for resource management even when no one is using the resources.

So, if you hear somone complain about their VMs are not performing very well ask them which product they are using.  If they are using VM Server, point them here.  If they are running ESX, then it’s time to dive into the typical operating system and application stack tuning/debugging routine.

(Disclosure: I currently work at VMware as a Solutions Consultant)

Since my mind is on efficiency this morning (see previous post about Facebook) I wanted to share an interesting blog posting I’ve had open in a browser for a few weeks now.  Steve Sounders, web performance guru from Google and previously Yahoo, posted some interesting thoughts on how green is your web page?

Steve did a quick mental experiment of calculating the CO2 emissions caused by bad code on a large website, he used wikipedia as his example.  I find this a bit interesting on the cyclical nature of the topic.  I might be showing my age a bit here, but back when I was a lad learning how to code up on the frozen tundra, we actually took into consideration efficiency and the cost of operations (maybe it was our proximity to Cray Research that drove this…).  I find it interesting that the green movement is causing this topic to be thought of again but in a different way.

I have been doing a little fun project like this myself at home.  A friend loaned my a device called Kill A Watz, which you plug into a power outlet and then plug other electrical devices into the Kill A Watz. The Kill A Watz then measurs how much electicity you are using on that one outlet.  It can track over time and give you the KW over a time period as well as real time watt usage.  I am using this on our home entertainment center to measure how much electricity is uses when it’s in standby mode.  Watch for a posting on that next week.  (I will give a teaser and let you know that a flat panel plasma TV uses twice the electricity when displaying a bright scene than when displaying a dark one…)

A few weeks ago my good friend Jon Hodgson was visiting me over the weekend. Jon and I worked together for over five years at Resonate during the startup, through IPO, to acquisition of the company (quite an interesting 5 year ride). Jon has spent the last few years, from Resonate through today, focusing on multi-tiered application performance tuning. This is a complex area that can contain all sorts of technologies within the Enterprise infrastructure (i.e., firewalls, networks, servers, web servers, application servers, database, middleware). He is currently the guy in this area at OPNET with their Panorama product.

If you have a multi-tier application (and who doesn’t today) and are running into performance issues, you need to give Jon and his team a call. He shared with me some of the work he has been doing in this area with clients, and the advances that he has been involved with in this area since we last worked together are amazing.

He showed me some of the capabilities that Panorama and the OPNET technologies have, and my jaw kept hitting the floor. With Panorama, he showed me how the tool can display the communication between each tier in a multi-tiered application, drill into the part of the communication that is causing the most latency, and then pinpoint the source of that latency within the multi-tiered environment. Like that wasn’t enough, he then combined the Panorama capabilities with ACE to then predict what would happen to the application’s performance if various changes were made to the network environment. Like what would happen if you added a firewall and only 1ms of latency between a Java application server and a database?

Jon and I were involved with building some of the early capabilities for a tool in this area at Resonate before the company was acquired and taken private. The recording, playback and predictive simulation capabilities of the tools that Jon has been involved with building at OPNET are mind blowing compared to what we were working on eight years ago! While I haven’t been involved in this space for a few years, I also know that I haven’t seen anything that comes close to the capabilities of the combined OPNET solutions.

While this tool may not work for everyone, OPNET’s Panorama is definitely one of those hidden gems of the technology world that should be considered by anyone responsible for application performance tuning.