Name
Abstract | ||
|---|---|---|
Multi-tiered storage systems using tiers of SSD and traditional hard disk is one of the fastest growing trends in the storage industry. Although using multiple tiers provides a flexible trade-off in terms of IOPS performance and storage capacity, we believe that providing performance isolation and QoS guarantees among various clients, gets significantly more challenging in such environments. Existing solutions focus mainly on either disk-based or SSD-based storage backends. In particular, the notion of IO cost that is used by existing solutions gets very hard to estimate or use. In this paper, we first argue that providing QoS in multi-tiered systems is quite challenging and existing solutions aren't good enough for such cases. To handle their drawbacks, we use a model of storage QoS called as reward scheduling and a corresponding algorithm, which favors the clients whose IOs are less costly on the back-end storage array for reasons such as better locality, read-mostly sequentiality, smaller working set as compared to SSD allocation etc. This allows for higher efficiency of the underlying system while providing desirable performance isolation. These results are validated using a simulation-based modeling of a multi-tiered storage system. We make a case that QoS in multi-tiered storage is an open problem and hope to encourage future research in this area. |
Year | Venue | Keywords |
|---|---|---|
2012 | HotStorage | storage QoS,QoS guarantee,desirable performance isolation,Efficient QoS,storage industry,Multi-tiered storage system,multi-tiered storage,storage capacity,SSD-based storage backends,multi-tiered storage system,IOPS performance,back-end storage array |
Field | DocType | Citations |
Disk array,Converged storage,Working set,Computer data storage,Scheduling (computing),Temporal isolation among virtual machines,Computer science,IOPS,Quality of service,Distributed computing | Conference | 8 |
PageRank | References | Authors |
0.54 | 18 | 4 |
Name | Order | Citations | PageRank |
|---|---|---|---|
| Ahmed Elnably | 1 | 34 | 3.08 |
| Hui Wang | 2 | 48 | 3.89 |
| Ajay Gulati | 3 | 573 | 27.79 |
| Peter J. Varman | 4 | 700 | 83.23 |