Michael Webster is among a small number of Nutanix Platform Experts (NPX-007) and VMware Certified Design Experts (VCDX-066), currently the only NPX and VCDX in New Zealand, and a vExpert, with deep experience delivering project management, ITIL based VMware operational readiness and technical architecture consulting services to enterprise and service provider clients around the world. He works as Principal Enterprise Architect as part of the Engineering / R&D Team for Nutanix, but this is a personal blog and does not represent the views or opinions of anyone else. He has been using VMware products since 1998 and has been designing and deploying VMware solutions since 2002. He specializes in the design and implementation of virtualization solutions for Unix to Linux migrations, business critical applications, disaster avoidance, mergers and acquisitions, public, private and hybrid cloud and software-defined datacenters. Michael has been in the IT industry since 1995 and consulting from 2001 to 2013 before joining Nutanix. He has established a niche as one of the global subject matter experts on virtualizing business critical applications i.e. SAP, Oracle, Java Systems, MS SQL Server, MS Sharepoint, MS Exchange and Monster VM’s. Michael is co-author of the book Virtualizing SQL Server with VMware: Doing It Right by VMware Press with Michael Corey and Jeff Szastak. In addition to this book he was technical reviewer of VCDX Boot Camp and Virtualizing and Tuning Large-Scale Java Platforms, both published by VMware Press. Michael is regularly called on to speak on all aspects of virtualizing business critical applications at events and for organizations all across the globe. Longwhiteclouds.com was recently voted as one of the top 15 virtualization blogs (#13) in the world as listed on vSphere-Land.com.
Visit the VCDX Directory on VMware.com
If you’d like to send me some feedback or contact me regarding a consulting engagement I can be contacted on twitter.com/vcdxnz001 and linkedin.com or please use the form below:
[…] Author […]
[…] Author […]
[…] Author […]
[…] Transparent Page Sharing (TPS). There exists an EMC paper that hints at disabling TPS. The astute Michael Webster thought this contradicted best practices I provided when leading VMware’s performance […]
[…] Michael Webster beschrijft waarom – en hoe – je een Oracle database omgeving op traditionele Unix platformen (zoals HP-UX) zou willen en kunnen migreren naar Linux onder VMware. […]
I read your post according RAC on VMWare.
Would it be possible to get details/scripts for the synthetic tests and testplan (think you use swingbench ?) and how to gather and compare testresults (service time, response time) – that would be very cool.
Whats about the jumbo frames for interconnect (real performance boost ?) – what is important and what have to be changed or what must be supportet on switch layer ?
How long does this take if there is no experience in this testing – days, weeks or more ?
Do you have experience how long it takes to evaluate Real Application Testing (my problem is not much time and budget :-))
Thank you
ng, jochen
Hi Jochen,
For the tests I've done I've used swing bench. VMware has also done a number of benchmarks using swing bench. It is freely available on the internet. Each version of swing bench is very different so the one you choose is the one you'll need to use for all the tests else the results will vary greatly if you cross versions. A documented test plan would be something that's only available as part of a services engagement.
Jumbo Frames on the interconnect is definitely recommended. It will reduce CPU load and improve packet response times in general if using modern NICs. But like anything it should be validated. I've done a lot of testing with Jumbo Frames on 10Gb/s links and it makes a big difference. With 11g R2 you can specify multiple interconnects and the benefit across the links will add up. In order to use Jumbo Frames you need to make sure all the devices in the network path between source and destination support the larger size packets (9000 normally). If you're using this only on the interconnect this should be very easy to achieve as it should be a private network. In that case only the switch ports used by the interconnect need to be modified. Be aware though that some switches need a global setting changed and that might require a reload / restart of the switch. You should seek advice from your network team.
I would say if you have no testing experience it'd take a few weeks to get up to speed. You also need to make sure you have a solid plan on how you will measure the test results. You can use Oracle Enterprise Manager in addition to other tools such as Vmware's vFabric Hyperic, VMware vCenter AppSpeed or AppInsight. AppSpeed is the tool I used to measure the application transaction latency.
Thanks
I have testet also with Swingbench but realized when running the same tests on an RAC with physical hardware – I could not compare it because the I/O and other things rise up to much and I have to stop the test after 2-3 min with only 5 sessions (I have testet to exclude warehouse queries and other but always the same results) 10GB source data.
In comparision to the RAC on VMWare I could run up to 800 sessions and near 50000 tps without a Problem. Setup on phy. RAC and RAC on VMWare is similar.
Also we have troubles with real workload (etl jobs).
The problem is that create index works in parallel as expected but an direct path load with parallel manually or auto DOP works not so good (1-2 subprocesses do the whole work but the other do nothing and it seems therefore we have only ~15% I/O than on physical hardware.
Now we try to tune the VMWare Layout more vmdk files and tune memory and some other things and also VMWare 5.0.
It seems the experience is physical hardware have more ressource and capacity to handle such workload and it works better out of the box than on VMWare you can compensate it a litte to use more clusternodes with less VCPUs – but you buy that with the risk of concurrency and other cluster waits.
Maybe someone have the same or similar problem with parallel (DOP) and can exchange the technical details (RAC on VMWare).
Thx
Jochen
Hi Jochen, In my experience all things being configured equal and optimally the physical RAC should perform no better than 10% at the most. However there is one difference you need to consider for Oracle and that's the effect of hyper-threading. You also need to ensure you're using the correct version of course. So by default Oracle 11g R2 (and possibly earlier versions) assumes you've got multiple threads per CPU (aka Hyper-threading) and therefore allows 2 parallel threads per CPU to be executed. This can degrade performance in cases where you need to get more IO out to disk and you want to dedicate the CPU to a particular process or thread. I've seen some examples where the degradation of performance was quite severe when the Parallel Threads Per CPU in Oracle was set to more then 1 when the Oracle system was virtualized. A VM does not have multiple threads per CPU core. Other than this the configuration of Oracle and the OS should be the same and you should be able to achieve close to native performance.
If you're only getting 15% of the IO throughput on your VM compared to physical there is very likely to be a configuration difference somewhere and this is impacting the comparison. How many LUN's do you have assigned to the physical? How about the virtual? What is the configured queue depths on the physical vs the virtual? What IO scheduler are you using in your OS on the physical vs the virtual (assumes Linux as the OS). What vSCSI adapters are you using and how many etc? One of these things is likely to the major contributor to the performance problem. If you continue to have problems I would encourage you to reach out to one of the partners that specializes in Virtualizing Oracle, or VMware Professional Services, to arrange a services engagement so that you can get all the benefits of virtualizing without sacrificing performance.
Hi Michael
I very keen to get my VCAP5-DCA & VCAP5-DCD this year would you be able to shed some light as to what I need to read up on to prepare for the exams?
Also I've invited you to join my linkedin network.
My Blog @ http://vmexchange.wordpress.com I'm just starting out on the blog thing.
Thanks
Shahidul Alam.
Hi Shahidul, My recommendations are generally the same for all VMware certifications. Carefully read the exam / certification blueprints. Identify where your skills are weakest and study up on those areas. Apply a good dose of real world experience and hands on work with the products. Then once you are happy that you've got a good grasp of all the blueprint areas you can book in for the exams. Be aware that these exams do have certain restrictions on time slots at some testing centers. I would also strongly recommend that you go through the vBrownBag preparation and study material for these exams which is available on professionalvmware.com and through iTunes. The US and APAC vBrownbag team have done a great job of compiling study material that has helped a lot of people pass the exams, provided the right amount of study and real world experience and applied hands on skills. Good luck and I hope you are successful.
Kind regards,
Michael
Hi Michael,
Thank you for pointing me in the right direction and also joining me in LinkedIn.
I look forward to getting to know you; and also "WOW" to one of the best home labs I've seen so far.
[…] an acclaimed futurist, MD of Foresight at Discern Analytics, and a Stanford University professor; Michael Webster, a New Zealand-based virtualization and cloud expert; Mark Burgess, researcher, author and […]