IOPS throttling
Posted: Fri Nov 12, 2010 4:04 pm
Hi,
I've been quite close to HW virtualization, terminal solutions (Terminal services, Citrix) or desktop virtualization (VmView, XenDesktop,...) etc. etc.
The common feature of such solutions is that they run on virtual hardware. (ESX VMWare, Hyper V, XenServer etc.)
These In many cases these solutions share many resources. (Storage, CPU, memory, ....)
Consider architecture like this:
1000 virtual Windows desktops running on top of
30 ESX VmWare servers which are connected to
1 enterprise storage
(This is pretty common.)
The risks:
Shared sources can be easily depleted.
On CPU level this is pretty well handled.
On ram level this has been also covered. (Either you have that ram or not.)
With storage the situation is a bit specific. Normally it is considered that disk CAPACITY is a problem. Which is not true because in such scenarios IOPS are the actual problem. (Which is normally not considered.) Moreover there does not exist a any product which would be capable of throttling IOPS!
I believe that FancyCache is the optimal product for implementation of IOPS throttling feature. (It would of course potentially limit the performance of the single machine, but it would ensure that other machines (sharing the same local storage) are still ok and not affected by high IOPS demand of single hungry machine.
At the same time this will be a key selling point for enterprises. Considering also all FC's caching capabilities, it would be a killer app. (I of course understand that such feature is not interesting for people running FC on their PCs or physical servers with dedicated local storage. (=local disks
) )
I've been quite close to HW virtualization, terminal solutions (Terminal services, Citrix) or desktop virtualization (VmView, XenDesktop,...) etc. etc.
The common feature of such solutions is that they run on virtual hardware. (ESX VMWare, Hyper V, XenServer etc.)
These In many cases these solutions share many resources. (Storage, CPU, memory, ....)
Consider architecture like this:
1000 virtual Windows desktops running on top of
30 ESX VmWare servers which are connected to
1 enterprise storage
(This is pretty common.)
The risks:
Shared sources can be easily depleted.
On CPU level this is pretty well handled.
On ram level this has been also covered. (Either you have that ram or not.)
With storage the situation is a bit specific. Normally it is considered that disk CAPACITY is a problem. Which is not true because in such scenarios IOPS are the actual problem. (Which is normally not considered.) Moreover there does not exist a any product which would be capable of throttling IOPS!
I believe that FancyCache is the optimal product for implementation of IOPS throttling feature. (It would of course potentially limit the performance of the single machine, but it would ensure that other machines (sharing the same local storage) are still ok and not affected by high IOPS demand of single hungry machine.
At the same time this will be a key selling point for enterprises. Considering also all FC's caching capabilities, it would be a killer app. (I of course understand that such feature is not interesting for people running FC on their PCs or physical servers with dedicated local storage. (=local disks
