Dutch VMUG 2011
- Published: Friday, 09 December 2011
On the 9th of December 2011 the yearly Dutch VMUG event took place. The Dutch VMUG leader opened the event and he even got a visit of the MonsterVM. The keynote was held Willem van Enter and Jeremy van Doorn. Willem started the session with some good examples of the usage of current techniques en how it influence current life. The new employees are looking for a Facebook alternative for business aspects and use their own devices. Consumers are already living their personal cloud and this is moving to companies.
Willem mentioned the three layers VMware is working on Cloud Infrastructure, Cloud Applications Platform and End User Computing. Jeremy continued the session that VMware has currently more than 50 products. He tried to briefly mention some of the products based on the three layers Willem already mentioned. In the Cloud Infrastructure: vCenter appliance, site recovery manager (vSphere replication, automated fallback, planned migration), vStorage appliance (combining local vSphere server disks to a shared storage), vCenter Operations Manager (monitor and only report when real significant changes are noticed), vCloud Director (self service VMs), vCloud Connection (connect the cloud to your infrastructure). Jeremy also touched Application Management via the vFabric AppDirector (incl. middleware, best practice blueprint), (Micro) Cloud Foundry and Application Performance Mangement. In the end user computing layer Project Horizon (self service portal), AppBlast (delivering Windows/Linux applications using HTML5) and VMware Octopus (secure datasync and collaboration service, dropbox alternative). After the keynote the parallel sessions started.
The first session was by Raymond Epping and Jan Willem Lammers (replacement for Duncan Epping). They presented the session vSphere Top 10 counting down from 10 to 1. Nr 10 was vMotion Enhancements (multiNIC support, reduced application overhead, support for higher latency networks). Followed by vSphere Storage Appliance on Nr9 (shared storage using local disks, 2 or 3 vSPhere nodes) and VMFS/VAAI (64GB storage volumes, 1 MB block size, life VMFS upgrade) on nr 8. Nr 7 was the vCenter Appliance (5 hosts, 50 VMs max with embedded database, external database only Oracle supports 1000 host and 10000 VMs. The appliance suffers linked modus, no IP v6, update manager and VAS). Network Storage vMotion enhancements (traffic management of VDS, share/limit principe, guarantee service level, only active when resources are low) was nr5, followed by Profile Drive Storage (placement of VMs based on labels of storage including check if the VM is using the right storage) on number 4. The top three started with Auto Deploy (more about his later on, because I followed the workshop later this day), followed by vSphere High Availability (completely rewritten, Fault Domain Manager, one master, other slaves but all can become master via election, disaster heartbeat for via datastore) on Nr2. The top position was for Storage DRS (automatic selection of best storage, advanced load balancing mechanism, triggers both on disk space as latency, affinity rules applies to DRS).
The second session was Mythbusters goes Virtual by Eric Sloof and Mattias Sandling. They focused on five myths. The first myth was Defrag your guest OS for best performance. VMs on a SAN does not understand of underlying disk geometry. Also defrag is heavy on DISK I/O. Other challenges are: change block tracking, thin provisioning, snapshot and linked clones and using defrag tools. VMware and storage vendors do not recommend using defrag tools and Mattias agrees on that, so the advise is not use defrag tools. Eric continued with the second myth: E1000 is faster than VMXNET3. Eric described the characteristics and origin of the E1000e and the VMXNET 3. Tests done by Eric shows that the VMXNET 3 is faster, more stable and has less CPU overhead. The third myth was Change Block Tracking (CBT) causes signification overhead on your VM. Mattias describes in detail what CBT is create for , how it works and the requirements. Also Mattias asked the developers which overhead CBT causes and these figures are represented in the tests. So use CBT if your back-up solution supports it. HA data stores heartbeats prevent host isolation. Eric explains how the datastore hearbeat works (datastores are used as back-up communication channel to detect host isolation on the management network), but does not prevent the network isolation of a host. The last myth was LSI Logic SCSI is always better than PV SCSI. PV SCSI was introduced in vSphere 4, but made a false start. From 4.1 many improvements are made to the drivers. Test shows that the PV SCSI has a higher performance (12%) and usage less resources on a vSphere host (-18%). In their own tests similar throughput figures were seen, with less host CPU utilization.
After lunch I attended the session vSphere Healtcheck by Gabrie van Zanten. He started his session on which tools (rvtools, Community PowerPack, VMware Resource/Performance Guide) he uses during his healtchecks. He advices to use the default settings, document all changes, review your design by updating to a new version and set-up automated checks (keep the results). Next he mentioned several subject you should check like %Ready (0%-4% ok, 5%-9% be carefull and +10% investigate), Reservations (guarantee physical memory/Shares/Limits(max of physical memory VM can use), CPU rule of thumbs (6 vCPU per core for server VM, 10 vCPU per core for VDI VM, XenApp 1 vCPU=1 logical CPU). Ballooning and active swapfile are the first memory checks. According to Gabrie memory overcommit is no problem as long as used memory is not higher as the available memory. According to the storage free space warnings Gabrie is mentioning that this is still based on % which is difficult for larger databastores. Gabrie advices to set a warning when less than 50 GB is available by using create folders for datastores and configure alarms with different percentages per folder. Also checks on orphaned VMDKs and use waste finder (available disk space with VMs). Snapshot can cause on high I/O system performance, so check on available snapshots. Gabrie also mentioned that different block sizes can cause slow copy (vMotion). Keep in mind that an upgrade VMFS5 (from VMFS3) his block size. Some quick wins during a healthchecks are connected cd-roms/floppy drives, outdated VMtools, old HW level, vmupgradhelper.exe, inconsistent foldersnames VM, Time Sync (use one method VMTools or AD) and free ports on vSwitch. After the log and dump files location Gabrie run out of time and lots of slides keep untouched on his slide deck.
The last session I attended was vCentre Operations Manager Suite (OPS) by Andre v/d Werff. He started with explaining why vCentre OPS is needed. The product is different than other monitor products because it's based on self learning analytics and powerful visualization. Also integration with 3rd party monitor products are possible. Their will be four editions of the product. Andre discussed how the product works, how the output is generated during badges based on health, risk and efficiency. The product will be available probably in Q1 2012.
During the last parallel sessions I attended the workshop vSphere AutoDeploy provided by XTG. The workshop was well organized and three instructors were available. It was a pity that about 50% of the persons showed up. During the workshop we build a AutoDeploy solution using the vSphere Storage appliance, host profiles and the answer file for per server customization. The workshop give a good view of the possibilities and the challenges there are in the product. Personally I think it's a cool technique, but has some limitations and challenges that is suits large infrastructures or a infrastructure that will be upgraded to a next version in the near future.
Just as the last years the event was well organized again with great sessions and lots of attendees. It was good talking to some guys again I did not see for a while, however there were less people I know than the previous version. Viktor and his team did again a great job.