It was a beautifully bright and sunny, but cold day as I cycled to Oxford rail station to head to the Oracle offices at Blythe Valley Park for the third RAC & HA SIG of the year.
Future dates for the next couple of RAC & HA SIGs:
Quite a small turnout really, down to 32. Audience this size makes the traditional round of surveying who is running what a bit less interesting.
Some people running 11.1, no one obviously running on 11.2
Practically everyone using physical standby with their RAC systems, no one on active dataguard yet.
Oracle Support Update – Andy Giddons
Andy showed a refreshing level of honesty about the Oracle database product for an Oracle employee.
Very surprised to hear him say that 18.104.22.168 is the terminal release for 11gR1. Quite unusual just having 1 patchset for a release, remember 22.214.171.124 is the base release for 11gR1. 9.2 has 8, 10.1 has 5, as will 10.2, 10.2.0.5 being the terminal release for 10.2.
I pointed out we had hit a number of bugs in testing 11gR1 that were marked only as fixed in 11.2. It really struck me that 11.2 is the release to go for.
Oracle Linux Test Kit – Martin Bach
“The more I looked the more impressed I was”
Martin had a project to replace ageing hardware running on rhel3 moving to rhel5 u3 also a 32bit -> 64bit migration.
He wanted to know who much better performance would be, how much improvement the new hardware would bring, so he could show to the business the value of purchasing this new hardware.
There are several load generators, including swingbench, hammerora, ORION. Also option to use Real Application Testing though this is not free, and they would have had to upgrade their oracle install to 11g.
Using swingbench, found a five-fold increase in transactions per second on the new hardware. found hammerora too complex, particularly the tcl component.
The Oracle Linux Test Kit is an open source project available at the oracle open source software site.
Designed to verify linux kernel functionality and stability
it is a testing tool and not meant for performance comparisons. Used in the oracle validated configuration programme andhas a handy silent install of the oracle software.
Requires a lot of disk space. Comes with both OLTP and DSS benchmarks. You’ve really got to have complete use of the box when running this, as it will bring all instances running on a box down when it is finished.
Supports a variety of filesystems, including NFS, OCFS2 and ASM.
Sounds like it could be a useful tool look at if you have a new hardware install project. Don’t think it supports 11gR2 yet though.
Oracle High Availability for an Age of Austerity – Steve Shaw
The whole driver of this talk was to show how to build a Highly-Available database system for as cheaply as possible. Essentially using Oracle VM to provide cold-failover of a single instance database.
Oracle VM has HA built in. This uses OCFS2 which is built into the linux kernel from 2.6.16
Both Oracle VM and Oracle Enterprise Linux are free to download and use including in production.
The OCFS2 part is used to manage the guest OS’s
Oracle VM can allow your single instance db to start upon another host should the host fail. Apparently live migration will be supported in the future.
talking about using commodity hardware and open-e software to fail between storage nodes.
Mirroring Arrays with Open-E
Steve showed an example of hot adding extra cpu capacity to a VM and then a graph of how the database immediately picked up the increased cpu count.
Claiming nehalem around x15 performance compared to 2004 cpu’s and double compared to previous generation. Also claims Oracle VM gives 84% performance of native performance.
Oracle on VMWARE – Carl Bradshaw
Interesting discussion on Co-scheduling and CPU skewing. Can use esxtop to determine whether va’s are suffering skew.
good idea to reserve the size of the SGA but don’t over reserve as upon failure you may not be able to failover that particular VM.
comparing RAC against VMware HA and Fault tolerance also comparing Dataguard to SRM.
not to keen on Fault tolerance, particularly the fact it’s the 1.0 release of it, and that it is limited to 1vCPU.
Showing how flexible dataguard is, though has mentioned that SRM is for the whole datacenter not just the database.
Virtulisation Panel Discussion
An unusual one this for a UKOUG meeting, and I think there was a reasonable discussion, even if the questions took a while to get started. Virtualisation is clearly a hot topic and it has filtered down to even the database level. I think the panel were still stating not to run mission critical high volume of transactions databases onto a virtualised platform, I’m sure VMware may have a different opinion on that one.
I really was questioning exactly how much cheaper virtualisation truly is, as I’m finding migrating to beefier boxes to run VMware on, once you add in all the additional cards (hba, and extra nics) as well as increased software licenses, and of course each VMware needs far more memory than your standard single application box, a VMware host is a *lot* more expensive than our previous hosts.
Sure I understand that there are savings in power and cooling, and of course your datacenter is fantastically more agile, and you have wonderful HA features, but these are bit harder to put down on a budget spreadsheet, bottom line is the servers you are buying are going to cost more money.
I think this can be a good format and I hope the RAC & HA organising committee pursue this format in future events.
Practical Considerations of a Stretched RAC Cluster Implementation – Deepak Singh
This was the presenters first presentation and he did seem a bit nervous at first. Basically they implemented a Stretched RAC cluster for a baggage handler, sighting the nodes in different terminals.
They used 11gR1 and ASM to mirror across the two storage arrays in the different terminals. Utilised the ASM_PREFERRED_READ_FAILURE_GROUPS to ensure reads only occurred to the local storage array. They had dark fibre connection between the sites, which were only 1KM apart. Oh, used Standard Edition to cut the licensing costs.
Raised an interesting question at the end about what happens if the connection between the arrays goes, but the nodes on either side of the cluster can still communicate with each other, and the voting disk in the 3rd location. There was an interesting discussion saying that this situation would crash the cluster, but that you could then decide to start one side of the cluster back up again.
I had to pull the emergency release cord and leave before the final presentation on Oracle Fusion Middleware.