The Systems Engineer

I recently acquired vCenter Operations enterprise. I have had a PoC done by VMware to ensure it met my requirements of a consolidated view of what is really going on in my converged infrastructure. Although I have some extensive knowledge into vCOps I am very interested in getting an under the covers look at how some of the data is calculated. I am also interested on how best to monitor my operation and what KPI’s I should really be looking at. Then I want to learn how best to interpret some of them. Let’s see what this session has to offer….

1. Lol first slide says it’s not just black magic! I like this session already.
2. With virtualization capacity is now fluid, I agree.
3. Invisible walls, with vm CPU and memory issues may not be resolved by adding more, contention can play a significant role. Proper troubleshooting is required.
4. With vmview you have to monitor end users not VM’s. I agree here as well as a user may move between virtual desktops. End user experience is important.
5. The key thing for VC OPs to do for me is to take the tons of metrics I have and to present the end calculation to me. Am I green or red?
6. Dynamic threshold analysis uses competing algorithms, meaning the system actually uses multiple methods to calculate the trend, then checks to see who is right more often and then uses that method. Genius. These are calculated every night.
7. A version change can cause the normal operation of a system to change. Thresholds may not catch this but trends have a better chance. I liked this idea.
8. Trending noise to determine abnormalities, this is really going to help my environment since we use a number of tools that are all sending emails for every little thing. We use our brains today to get a feel for the data center health. I declared this a broken model earlier in the year.
9. Alerts should be an indication of a real problem, yes yes yes! Do not alert on every threshold that is reached. Yes please. May I subscribe to your newsletter!
10. Root cause determination in this product is really root metric determination. It isn’t telling you what the problem was just what metric that was being monitored was the starting metric of the issue. I.e. We saw disk latency go to 100ms before the app crashed.
11. Workload is demand divided by entitlement.
12. Right-sizing is a concept I always support, but VM admins and I seem to be on the front line of this alone on this.

Bookmark to

I like vCenter Configuration Manager more and more as I use it. The reporting and control it gives me is much better than my previous product. I am pointing at my Altiris server with the dust on it. Who am i kidding, I virtualized that product when we installed it. The thought of a physical server not running esx on it is laughable to me. I am going to sit here and eat my pretzel while we wait for the session to start. I assure you that your wait will be shorter than mine.

See? Let’s begin.

1. This session was a little weird because the presenters would interrupt each other and start talking. Odd.
2. VCM integration with vCloud Director looks cool. Auto discovery and understanding of machines with the same name and ip but in different zones.
3. The change detection is tricky for me. I want to monitor everything that could possibly change because of paranoia.
4. Although I do not use it, VCM can provision OS, via pxe boot.
5. Someone agrees that the cloud should include moving the workload back and forth between the public and private cloud.
6. One thing I had not really considered before but I am now, is giving read only access and reports to the server app owners in my org. We keep this info to the server guys right now.

Bookmark to

A few notes from the session.

1. The iterative memory copy that vMotion does was interesting to hear more about in detail. I think too many people may not quite understand what a vMotion is really doing. I often just explain it as black magic.
2. VSphere 5 interests me, the idea of multiple nics being used for vMotion. I wonder how best to configure this in UCS. Is vMotion traffic more than client traffic?
3. VMotion ID once found can give details about vMotion in the vmkernel log.
Tests appear to show that vMotion on vSphere 5 is faster, 37% drop in time on vSphere 5.
4. vMotion obviously gains tremendous performance on 10GbE.
5. Multiple vMotion NICs and vMotion performance gain in vSphere 5.
6. Configure vMotion NIC’s on the same vSwitch.
7. For best vMotion performance place vm swap on SAN.
8. If you use CPU reservations, leave some CPU for the host for vMotion ~30%.
9. Metro vMotion supported in vSphere 5. Doesn’t help us VPlex Geo guys.
10. Conclusion? Upgrade to vSphere 5 ASAP if you want faster vmotions.
11. I’m headed to ask about vMotion over asynchronous distances. If not NDA I will share

Good session for vMotion people.

Bookmark to