Cameron HaightI just came across an interesting paper on a VMware project called PARDA. The acronym stands for Proportional Allocation of Resources for Distributed Storage Access and represents an attempt to potentially deal with increasing virtualization I/O concerns using a "proportional-share resource scheduler that can provide service differentiation for I/O like VMware already provides for CPU and Memory." Interesting charts that show that the target is to address latency and not necessarily bandwidth. While many clients that I speak to suggest that they focus most on memory in terms of costs, they also cite that from a performance standpoint they often concentrate primarily on storage I/O so this could very interesting. It does though seem to potentially create an interesting scenario for virtualization consumers, i.e.,., do I use a common VM-oriented I/O scheduler that is seemingly independent of the back-end storage, or do I use array-based service controlling mechanisms that may be independent of the underlying VM technology platform?
Category: Uncategorized Tags:
1 response so far ↓
1 Irfan March 8, 2009 at 1:05 am
Hi Cameron:
Thanks for providing coverage for the recently published research paper on PARDA (and for linking to my blog). I’m also writing a series of posts (http://virtualscoop.org/) explaining PARDA in easy to understand terms.
Like you said, one interesting question is whether to use “a common VM-oriented I/O scheduler that is seemingly independent of the back-end storage, or do I use array-based service controlling mechanisms that may be independent of the underlying VM technology platform?” One advantage the PARDA approach might have is the ability to provide per-VM level proportional sharing.
Irfan, VMware, Inc.
http://virtualscoop.org/