With the keynote at TechED it is clear that one of the major areas of innovation are in memory computing…
With new advances in in memory computing … , HANA , Exadata etc we see that the everyone including Oracle has their eyes set on in memory computing – which translates into big big boxes with very very powerful CPUs and hardware to support intuitive and innovative ways to access data.
This would also mean that slowly enterprises would shift from having multiple application servers to having one central instance which will house the database etc and then linearly scalable applicances which are in memory appliances which serve the analytical needs.
but then wait….
As Vijay Vijaysankar pointed out in his blog…
Post Teched questions on HANA
what about the client hardware – the question being this … should enterprises essentially have a centralized hardware model ignoring the possible computing power that the client desktop / mobile might have ?
I am typing this on a laptop which has an Intel Dual Core processor with about 3 GB of RAM. When I run a BW query on my laptop , it is only natural to expect that it performs as fast as my laptop. Can I contribute to this ..?
One topic of interest is the Desktop Grid..
here I have given a link to one of the Desktop Grid projects .. what this attempts to do is to have the unused desktops / computers contribute to the absolute computing power and this approach could mean that you can have your desktops in the enterprise contribute their unused computing power to an application servere / central instance which can then use the same to perform its computing activities.
All this requires is some software to help the server co ordinate computing resources – and since most of the clients have some form of a GUI the same software could be included into the installer . SAPGUI for instance and this would mean that all systems running the SAPGUI are included.
It seems like an interesting prespective to have your client systems help you realize server capacity .. but then considering that the minimum recommended RAM for the SAPGUI is 1GB and if you are using Microsoft 2007 then in all likelihood you have 2 GB of RAM and if you have 5000 desktops with an average of 1 GB and assuming that each system is able to contribute 20% of its RAM then you are looking at 1000GB .. this is of course assuming that all systems are switched on which would mean a very high power bill… but then system usage in terms of number of users logged in and number of CPUs available are directly correlated.
I am still not proposing that you remove all applicaiton servers and have a grid based system instead .. all it means is that this is a possible source of computing power that would be very interesting to look at instead of building siloes of powerful servers and applicances. Also the incremental cost of this approach would be lower meaning that a you could add desktops to the enterprise and since these are desktops you can also get productivity gains with employees using the same as opposed to having additional servere which people can connect to but cannot access mail from!!!
On the flip side .. not many enterprise software companies would be interested because :
1. It represents a marked departure for enterprise software which has always meant big servers and clients which have always been consumers and not contributors in terms of computing power.
2. Computing involving client desktops cannot be predicted .. but then when you compare the usage of the system to number of users – then the more the number of users would mean more systems that can be used. This process
however does not consider background processes that run on the system.
I am sure in times to come .. desktops would become more powerful and be considered as a proper alternative to make grid computing something that can be looked into with more seriousness leading to an approach which increases the time existing hardware can be used and delay application server / hardware upgrades leading to revised and higher ROI realizations.
Related terms :
Desktop Grid computng , Distributed computing