I have been working on a large scale EMC and VMware vSphere deployment project and was trying to fill in the gaps to bring operation efficiency on the table. Here I am working EMC VNX 5700 series hardware and VMware vSphere 5.0 based deployment. We have debated over again and again on the fact that how VMware vSphere is going to balance the Storage IO in the back end array. My answer is always use SIOC where ever possible (due to license restriction). But now when we are using EMC VNX and using FAST VP (Fully Automated Storage Tiering for Virtual Pools) why don’t we employ both of them and balance it fully in the Hypervisor layer as well as in the Storage backend layer too. So, here is my deal.
“Storage IO Control and EMC FAST VP always complements each other and we should use it where ever possible.“
So, now let us see how they get together and made our life easy.
Fully Automated Storage Tiering for Virtual Pools, or FAST VP, intelligently manages data placement at the sub-LUN level, thus increasing overall performance. When implemented on a storage system, FAST VP measures, analyzes, and implements a storage-tiering policy much faster and more efficiently than any user could.
Depending on the performance characteristics, FAST VP puts drives of varying performance levels and cost into storage pools. LUNs use the capacity from these pools as needed. FAST VP collects statistics based on I/O activity at 1 GiB slices. These statistics are then used to determine which slices will be promoted to a higher tier of storage.
At the users discretion, relocation of the data can be initiated – either manually or by an automated scheduler.
FAST VP is a licensed feature available on the VNX series supporting a unified approach to automatic tiering for both file and block data.
All data goes through a lifecycle and experiences varying levels of activity over that lifetime.
For example, when data is first created it is typically used heavily and as it ages data is accessed less often. FAST dynamically matches these data characteristics to the tier of disk drives best suited to the data access patterns.Disk drives are segregated into tiers based on the frequency of the data access.
Disk drives also offer varying degrees of performance and are placed in tiers accordingly. Flash or SSD drives perform the fastest level followed by SAS and NL-SAS respectively.
Data that requires a high response time and has a high rate of IOPs should be optimized by allowing FAST VP to migrate this data to these drives.
Ok, so now let us take a look how SIOC can be integrated with FAST VP and get benefited out of it.
FAST is supported with Storage I/O control (SIOC). SIOC is a technology on the VMware side that allows users to configure I/O shares at a datastore level, so that a critical VM receives the I/O throughput and response time when needed, without worrying that another VM on that same datastore won’t hinder the performance of a critical VM. One works at the VM level and the other works at the drive level – both complement each other.
For example, lets say you have two VMs, each running SQL, however one VM is critical and the other is not.
FAST will service the I/O from both SQL databases equally. However, you can designation these VMs as HIGH and LOW in SOIC and ensure the right performance is given to your VMs.
I think it’s probably a situation where we either need to look assign a response time value based upon the weakest link (NL-SAS). So if we expect the performance/response time for
EFD – 15ms
SAS – 30ms
NL-SAS – 30 – 50ms
The FAST VP Pool with NL-SAS would need to be set to 50.
Of course the does not help if all of the data resides on EFD and the LUN experiences contention.