To boost HCP disk i/o performance we thought of placing LUNs for data nodes in a HDT pool - is it supported ?
Thanks Michael. I think the conclusion will be that it's of dubious value to use HDT. We discussed alternatives with HDS and fortunately there's a number of techniques to improve search . But if HDT could have done it, it would bee a convenient solution.
Hi Kenneth Pedersen
acctually vsp has no problem with HDT pool because HCP doesn't see what type lun youre assign to HCP.
it's my suggestion
Hi Audhie Ardhana
Thanks a lot for your answer. I'll create a new HDT pool and try it out.
Technically HDT is supported.
But thinking in other terms.
- HCP contains data that is WORM type. Means the data is usually not active and hence it is being archived.
- So the usual selection would be to place it in NLSAS disks.
Thanks for your comment. I think you got a point here. We discussed this issue in terms of relocation of object pages and metadata pages in particular, from tier2 (3 TB SAS) to tier1(SSD) , to better service search operations. But I think it's doubtful if we can achieve accesspatterns that ensures relocation of relevant pages to tier1. Also it will probably be problematic to maintain these pages in tier1. We expect a lot af search activity though, but the frequency of search against metadata belonging to a specific object will vary a lot.
Hi Kenneth Pedersen,
Please also take into account that HCP runs services that constantly iterate over objects in a repository.(and the default schedule is pretty packed). It is going to be really really hard to make any sense or prediction of when each object is accessed. It is certainly not a static system. Yes, HCP will work with HDT - because it really doesn't care where its LUNs come from (unless you're using spin-down). But you most likely won't get any performance or other benefit from using HDT.
Retrieving data ...