09-18-2014 12:04 PM
What are the advantages/disadvantages in installing Grid on Virtual Servers rather than Physical Servers?
Is there any SAS presentations/documents regarding this?
Please share your expertise ...
09-18-2014 02:52 PM
There are a lot of papers on this topic.
Some for dedicated hardware
You can find als some as a SAAS approach.
Suppose you would possess a Physical server server that is capable to do the workload. You can buy commercial machines with 64 processors and 1Tb internal memory and many terabytes of external storage including SSD's. You can tune this hardware for the more demanding analytics processes. Suppose that machine is far enough for your requirements.
Question: would is make any sense to define 32 virtual machines on this hardware so you can all your load in a grid process. Adding ca 10% overhead for your virtualization than adding the complexity for your grid and needed shared filesystems for that. Ending up with far from optimal resource usage because every users want all of his resources being reserved.for him?
The virtualization dogma is based on the assumption that many users are just running low demanding jobs on weekly monthly way. You could optimize the hardware by sharing the hardware using virtualization of the OS. This assumption is not a fit with high demanding bi & analytics. The user are bypassing IT staff .by accepting appliances (external supplier is managing running your business). The result of that is missing compliancy to eg regulation requirements. Your into a political discussion mainly not a technical one.
09-18-2014 03:54 PM
Virtual servers are easier and quicker to set up, configure and clone so for a SAS Grid solution there would be a lot less work involved. Also mirroring your SAS Grid for Disaster Recovery would magnify those benefits.
As Jaap has said virtual servers are not suited to high-end analytics like SAS Enterprise Miner and SAS VA. However there is no problem having a mix of virtual servers and physical servers with your high-end analytics restricted to physical servers.
09-18-2014 06:04 PM
Have you considered which type of shared file system will be used by your VMs - I would run some basic i/o tests before making any decisions.
Grid is useful if you want to run your load near to maximum and have the kind of mixed workload that needs queues and prioritisation that is more sophisticated than you can get from the O/S.