Global cache for HA

fcastro · Mon Feb 01, 2010 8:37 pm

I believe that the next logical step for Starwind after dedicated target cache release is to have a redundant global cache for HA. That will dramatically improve the performance at pair of io development gainning that could allow more vms in less bigger lun's even on old equipment. The base tecnology exists with ram virtual disk and it is the case of making it HA and dedicate it as buffer for HA luns that could use it dinamically and so making the best use of it - also as sugestion for high performance luns: use shares basis and not reservation exclusively.

my intent is for all who concur to quote this topic.

Mon Feb 01, 2010 10:45 pm

1) We have already what you call "global cache" rather then per-target cache.

2) We cannot "use shares and not reservation exclusively" as a) it's hypervisor who locks the LUN b) Hyper-V does lock only region of sectors rather then whole LUN. Hope ESX will do the same soon.

fcastro wrote:I believe that the next logical step for Starwind after dedicated target cache release is to have a redundant global cache for HA. That will dramatically improve the performance at pair of io development gainning that could allow more vms in less bigger lun's even on old equipment. The base tecnology exists with ram virtual disk and it is the case of making it HA and dedicate it as buffer for HA luns that could use it dinamically and so making the best use of it - also as sugestion for high performance luns: use shares basis and not reservation exclusively.

my intent is for all who concur to quote this topic.

fcastro · Tue Feb 02, 2010 9:57 am

could you explain better: - i am having the trouble of creating multiple luns due to my poor disk io but i am also limited on ram and thus i have to divide it for each target. Anton, are you saying that having Starwind the tecnology of Ram disk and HA they will not combine it and pursuit it? The ideia is to have a dedicated Global HA Ram cache to dinamically give to targets as needed or requested at given moment, freeing it for other targets when idle. As for shares and reservation - i'm refering to memory cache.

Tue Feb 02, 2010 11:23 am

I see what you mean now! The answer is - it depends... In the future versions will have the ability to change cache size dynamically. But for whole cache/memory management we'll rely on operating system services.

fcastro wrote:could you explain better: - i am having the trouble of creating multiple luns due to my poor disk io but i am also limited on ram and thus i have to divide it for each target. Anton, are you saying that having Starwind the tecnology of Ram disk and HA they will not combine it and pursuit it? The ideia is to have a dedicated Global HA Ram cache to dinamically give to targets as needed or requested at given moment, freeing it for other targets when idle. As for shares and reservation - i'm refering to memory cache.