georges' blog

August 30, 2012

VMworld 2012 Takeaways, Day 4

Filed under: Technology,VMworld 2012 — kendall @ 1:36 pm
The Thursday keynote at VMworld generally highlights some inspirational technologies and their creators, generally not at all related to virtualization. In the 2012 Thursday keynote, Genius Machines, the speakers talked about autonomous algorithms that run the world, humanoid robots, and self-driving cars… In short, we are closer to the robot wars than ever.
Kevin Slavin gave a talk about how algorithms run our world. It was very similar to the Ted Talk he gave July of 2011. Check it out, it’ll terrify you. Dr. Dennis Hong, Director, RoMeLa (Robotics and Mechanisms Lab), Virginia Tech University, talked about the why of creating humanoid robots… and why make them play soccer. Chris Ormson of the Google self-driving car project also talked about making driving safer, more efficient, and more productive… and also making it accessible for people with disabilities like the blind. Check out this video to see what I mean:
The one session I attended apart from the keynote was Virtualizing SQL 2012: Doing It Right, important for our organization. Here are the take aways:
  1. You can do it, because you can create monster VMs that give you all the resources you will need.
  2. Yes, do it in tier 2 first.
  3. Collaborate, don’t keep it a secret.
  4. Do basic throughput testing, SQLIO/IOMETER, prior to deployment
  5. Unattended installs, SQL sysprep
  6. The OS and databases don’t know they are virtualized
Architecting for Performance
  1. Design for workload types or mix workloads
  2. Pay attention to storage types
  3. Understand your physical infrastructure’s limitations
  4. Use VMFS and NOT RDM. The performance difference is statistically insignificant but you get more features with VMFS.
  5. Use Thick Provisioned, Eager Zeroed disks for best performance.
  6. >80% of problems with virtualization happen at the storage layer
  7. Avoid over-committing CPU resources. Use one vCPU per core. Do not count a hyperthread as a full CPU core.
  8. Some important memory stuff. Check the slides. Sounded like there were a bunch of important memory considerations and settings on both the server and host.
  9. Mike says, “Avoid shares and reservations.”
  10. Jeff says, “Don’t use limits.”
  11. Don’t turn off ballooning.
  12. Enable jumbo frames.
  13. SQL best practice analyzer
Here’s the slide deck. (But you’ll need to sign in.) A lot of slides. A lot of info. Good stuff. I took the VCP exam at VMworld.  I recommend doing this because it is half price.  I don’t recommend doing this because when will you study?  It worked out for me.  I passed.

August 29, 2012

VMworld 2012 Takeaways, Day 3

Filed under: Technology,VMworld 2012 — kendall @ 11:47 am
Day three started off with no keynote. That is probably because there are a lot of parties Tuesday night, so a lot of reasons to sleep in. Personally, my first session of the day started at 9:30, What’s New with vSphere Automation. They were talking about new commands in esxcli and PowerCLI. The PowerCLI guy sounds a bit like Ricky Gvais, but not so funny. Takeaways:

  1. 82 new commands in esxcli.
  2. A bunch of snap-ins, over 270 commandlets in PowerCLI

I literally get to take away a PowerCLI and esxcli quick reference poster. Kinda cool since there isn’t much 5.1 documentation at yet.

Session 2, VM Scare? Heterogeneous Virtualization’s Impact. This guy is not into heterogeneous hypervisors. He’s from Gartner, so it isn’t because he’s trying to sell us something. In fact, Gartner makes their living off complexity–if the world is simple, what do we need consultants for? He’s basically saying there aren’t good reasons for using multiple hypervisors.

  1. He says if you aren’t moving toward standardization, then you are not doing things enterprise; you aren’t doing things like a service provider; you are doing things like a small IT shop. Those are fighting words. But I think he has a point.
  2. If you are trying to avoid vendor lock-in for cost reasons, you need to understand there is a cost associated with switching for one hypervisors to another. Also, there is inherent waste in having multiple hypervisor silos.
  3. You will have to have multiple management tools, because you will need vendor specific tools to do some advanced operations.
  4. Public cloud doesn’t save you money. The vm is cheap, the management and DR is what costs you.
  5. Do not use different hypervisors for test/dev and production. It defeats the purpose and causes problems when moving from test/dev to production.
  6. Check out Gartner’s reference architecture for cloud management SaaS.

I got to ask about how far this recommendation extends to VDI, and he gave a bit of an “it depends” answer. He actually said that VDI is perhaps the only place where having a siloed approach makes sense. He had a couple of observations:

  1. CAPEX is a major consideration with VDI. You get CAPEX savings by bundling your VDI with your hypervisor.
  2. You may decouple VDI from hypervisor because in many/most shops the desktop guys are not the infrastructure guys. So, the infrastructure commitment to the VDI management is generally limited. Let the infrastructure guys choose the hypervisor; let the desktop guys choose the VDI.

I like that last observation. It may be just because choosing a VDI solution is a complex process and I’d be happy to pass the job off to our user team. My first order of business when I get back to work next week.

I went to the the session, Securing the Virtual Environment: How to Defend the Enterprise. What I learned? Our security guys need to understand virtualization. Today over 60% of servers are virtualized. There are special considerations that apply to virtualized infrastructure. The old rules and methods may not apply. It will become unacceptable to halt forward progress because the security guys don’t know current technology. Our security guys should get this book, Securing the Virtual Environment: How to Defend the Enterprise Against Attack.

August 28, 2012

VMworld 2012 Takeaways, Day 2

Filed under: Technology,VMworld 2012 — kendall @ 2:23 pm
The day started with a less dramatic keynote than Monday. Steve Herrod, VMware CTO, gave an update on end-user computing. I like the prospects of Operation Horizon–the promise of delivering the user a single place to go for all their apps, desktops, and data. Last year Steve demoed mobile OS virtualization on an Android device. That was pretty slick, but it was for Android only. This year he demoed some interesting developments in the iOS space. While the Android solution is still more impressive and elegant, at least VMware was able to demonstrate a commitment to developing for iOS, since that is what most customers want. The biggest eyebrow-raiser for me is the integration of XenApp into Horizon. VMware is going head on at Citrix in the desktop virtualization space, and other fronts: cloud file storage, user portal, self-provisioning… It is interesting that they seem to be throwing in the towel on application virtualization. Not sure if they are running into too much resistance on the electronic medical records front, where hospitals seem to be largely committed to XenApp or what. Or, if they are having trouble innovating with respect to ThinApp. In any case, it is an interesting development. Personally, I’m not sure I want to support infrastructure for both View and Xen if I don’t have to.

Takeaways from session vCAT 3.0: Architecture to Implementation in 5 Easy Steps:
First of all, total false advertising. I was intrigued by the notion that there were five easy steps to architecting anything, much less VMware. As I suspected it wasn’t easy. It turns out there are five steps, and these are essentially the same as any project management plan–five very involved steps that are probably easy enough for someone used to a complex project management process. But anyhow, there are some important take aways.

  1. vCat 3.0 is the latest release of a toolkit for VMware architects to do their own designs… methodologies and considerations for building and deploying VMware technologies.
  2. Use the VMware implementation examples docs; these provide real examples of VMware designs.
  3. Get vCat.

Now for the arguably geekiest session yet, Virtualizing Oracle for Disaster Recovery with Data Guard and VMware vCenter Site Recovery Manager. Takeaways:

  1. Use SRM to failover web and app vms, but use DataGuard to failover the databases to the secondary servers. Okay, but are you going to explain why? I think what they are saying is that it is preferred to use DataGuard for database failover rather than SRM as DataGuard is designed specifically to avoid database corruption whereas SRM is not. Good point. Touché. King’s to you.
  2. Don’t use VMware vShpere Replication as it does not currently deliver synchronous replication. In fact, it does not guarantee less than a 15 minute RPO. So, some data loss is virtually guaranteed.

How about benefits of virtualizing Oracle?

  1. Performance is not an issue.
  2. Server consolidation
  3. Simplify deployment
  4. More…

Daily Fails. So today had some fails that warrant mentioning:

  1. Yesterday when Steve Herrod mentioned that they would be live polling the audience during Tuesday’s keynote I was doubtful that it would work. If you’ve ever been to a large concert or sporting event you know how difficult it can be to make a call, send a text, or check-in on Facebook. Now, VMware is not a wireless networking provider, but they should know if you get 20,000 geeks together in one room you are going to have 40,000 wireless devices. Consequently, just getting on wifi was tricky. Navigating to the polling website was darned near impossible. Deep in the bowels of the Moscone, AT&T’s 3G suffered from the same problem. I was able only to respond to two of the polls and that with much effort. All of that work that your device does to try to connect, try to send, resending… It kills your battery while you get next to nothing done. During the keynote, I depleted half my battery. I don’t know who is providing this conference’s wifi, but i suspect they aren’t using the latest advances in this space. There are a number of companies innovating in this wireless space–large scale extremely high user density. What I’m talking about is using many radios, an array of multiple uni-directional antennas, and pairing devices to the right radio, frequency, and data rate, based on the profile of the device. I’m thinking of vendors like Xirrus (disclosure: I’ve gotten nothing from Xirrus for this mention, though if they wanted to give me something, I’d take it). Ultimately, this lousy ability to respond to the polls probably doesn’t matter. If you have a basic knowledge of statistics, you understand that you only need a relatively small sample size to have a reasonably accurate poll–not good enough for voting, when every vote should count–but good enough for a poll. So, my complaint isn’t about the validity of the poll, but about the user experience. The irony was not lost on me, that the audience participation aspect of this keynote about end-user computing/customer experience pretty well stunk.
  2. Yesterday Steve promised that all five of VMworld’s platinum sponsors would be delivering four minute live demos. And that we would be able to vote for the best in an American-Idol-America-chooses fashion, hence the need for live polling. NetApp ultimately won the competition, but I dispute that outcome. As I’ve already discussed I don’t dispute the validity of the poll. My beef with respect to the outcome is not in regard to voter fraud, or some such thing. NetApp didn’t satisfy the criteria of the competition. Not only were the polls supposed to be live, but also the demos. Of the five competitors only EMC provided a live demo. The others used video demonstrations. HP was the sneakiest. Their presenter was kind of miming the mouse movements during the video. Not sure if it was intentional. Seemed that way at first, but later it was obvious he wasn’t doing it live. In the case of Dell, the mouse movements and screen refreshes were faster than humanly possible. Cisco didn’t even do a demo, but a song (lip sync’ed) and essentially a PowerPoint. When you advertise a live demo, deliver a live demo, not a sales presentation. So, my kudos go out to EMC who actually did a demo–of a file recovery–in nearly the allotted time. (disclosure: the University of Oklahoma is an EMC customer, but we get storage from a lot of different vendors; EMC gave me nothing for this mention).

August 27, 2012

VMworld 2012 Takeaways, Day 1

Filed under: Technology,VMworld 2012 — kendall @ 12:31 pm
VMworld got off to a dramatic start this morning with a drums and dancing spectacle and a passing of the baton from outgoing CEO Paul Maritz to former EMC COO Pat Gelsinger. Can’t say I have no concerns about continuity of vision and momentum. I’m also wondering what is behind the leadership change up after only four years at the helm. Word is Paul will still be involved as a board member and still walking the halls of VMware.

There were a few other notable announcements like dropping the vRAM-based pricing model and announcing a per CPU pricing model with no core count, memory, or VM count restrictions. Other interesting announcements were a focus on end-user computing and management tools. I wonder what implications this has for “coopetition” between VMware and Citrix and between VMware and any number of management tools providers currently on display in the Solutions Exchange.

Anyhow, how about some takeaways from my first session, Architecting Auto Deploy for Availability and Scalability:

  1. Deploy your auto deploy servers as VMs. Build a small management ESX cluster that does not use auto deploy for your auto deploy servers and other management servers, like vCenter.
  2. Deploy one auto deploy server per 40 hosts.
  3. Don’t install auto deploy on your vCenter server, it uses a lot of CPU. It is not multi-threaded, so it doesn’t benefit from many cores, but it will benefit from more cycles. Best practice is a 2 vCPU VM.

A tip from my second session, esxtop for Advanced Users: Set the power regulation setting in the BIOS to OS control to get more useful data out of esxtop. Otherwise the OS, ESXi in this case, will not be able to differentiate CPU/core activity and will only display total utilization. At least that’s what think he was saying. By-the-way, there sure seem to be a lot of people in this session… And all glassy eyed. Not a good choice for after lunch. They are probably all trying to make people think they are total wizards by sitting in on such a geeky session. That’s why I’m here. I’m probably going to sound like a geek here, but this session and a later session, Become a Rock Star with PowerCLI and vCenter Orchestrator nearly made me want to take some time away from my conference agenda to get on console and mess around with esxtop and PowerCLI. I have three back to back parties to go to and I’m wondering when I can remote into a server. What is wrong with me? It will have to wait. I’m sure I’ll be in no condition for command lining by the time I get back to my hotel room.

Daily Fails: While VMware deserves a lot of commendation for pulling off such a massive conference, with a lot of complexity, and a lot of very impressive technology–and this year with a record 20,000+ attendees–I gotta report on a couple of fails from the conference so far:

  1. The Hands-on Labs crapped out on Sunday. I was excited by the prospect of getting some labs knocked out before the conference started. The labs were scheduled to open at 11 AM on Sunday. At 10:30 AM the queue was snaking all over the lobby of Moscone West. Some nerds got in line first chance they could like it was some kind of Star Wars premier. That didn’t work out too well, as by mid-afternoon technical difficulties were creating an estimated three-hour wait. I went back that evening after the welcome reception. While the beers in my system weren’t cooperating too well with my learning objectives, there was no wait.
  2. During my first session, I meant to log into the VMworld mobile app to check my schedule and was greeted by a message saying the server was not available due to maintenance. It would be bad for any company for their website to go down precisely when the users need it, but this is particularly bad for a company whose whole business is about delivering daylight hours maintenance without interrupting the service and dynamically scaling out capacity on demand, and which advocates going all-in on a cloud paradigm. Well, the cloud failed me. Fortunately, the service was back up in time to get me to my next session on time, but my trust in the cloud was shaken if ever-so slightly.
  3. This has to be said… The alumni lounge this year is fantastic… By comparison with a partitioned off little area in the basement like it was last year at VMworld 2011 it is a serious upgrade. It’s at Jillian’s a billiards hall and sports bar on the corner of the Metreon, so it is super-conveniently located–right in the middle of all the Moscone venues. Soft drinks are free as is the adequate selection of snackables. Seating is adequate and the booths and sofas particularly comfortable. They could use use some help with their wifi config. They’ve got three different public wifi networks available–all with no authentication which is preferred. Can’t they find someone in San Francisco who could help them clean that up though? Here’s the real problem… Where is the friggin’ coffee?!

Powered by WordPress