After weeks of qualification and performance testing, Host.net has introduced the new VNX storage system from EMC onto its enterprise production storage platform. Host.net leverages DataCore’s SANsymphony product line as its core storage intelligence. DataCore provides the ability to use just about any storage array from just about any manufacturer. I say “just about any” not to imply that I have found a storage array that is incompatible with DataCore, but to simply say that I have not tried every storage array on the planet. I am batting a thousand so far though.
INTRODUCING THE EMC VNX 5300 For Block
It is worthy to note that performance was not the reason why we purchased the VNX, although performance certainly showed up to the party. Capacity expansion was the primary factor we considered in our due diligence process. The VNX 5300 allows for expansion up to 240TB per system. We considered the 5500 and 5700 series systems, but our decision to go with the 5300 series specifically was based on risk analysis and favorable platform alignment with SANsymphony. The 5300 provided us with a high-density-feature-rich platform with the least amount of risk when compared to other models. In my experience I have found it much better to introduce a greater number of smaller systems rather than a fewer number of larger systems. One could argue that more systems introduce more chance of failure. While I would normally agree with that statement, the more important factor to consider is ‘risk per unit of storage’. Leveraging a greater number of smaller systems mitigates the risk to a smaller data set, which means less data could potentially be negatively affected during a storage system event. The risk of course is further mitigated by the fact that we do everything in pairs with DataCore. (The risk analysis of storage systems is a very interesting but lengthy discussion for another time).
CACHE IS KING
Combining the VNX 5300 pair (each of which includes 400 GB raw/366 GB of usable high-speed enterprise SSD FAST cache and 16GB of controller cache) with our DataCore platform (which provides synchronous mirroring, 128GB of ultra-high speed DDR3 caching, and intelligent coalescing/read-ahead caching algorithms) provides us with four tiers of cache totaling nearly 1TB (Tier 1 – DataCore Cache, Tier 2 – VNX Controller Cache, Tier 3 – VNX FAST cache, Tier 4 – Individual drive cache). By introducing such a large amount of system-wide cache, the slower spinning disk is buffered from most of the heavy burst I/O that is experienced in shared virtual architectures.
The maximum FAST cache configuration supported on the 5300 (using 200GB drives) is 2x 200GB RAID-1 sets (400GB raw/366GB usable) with a hot-spare. The FAST cache provides a system-wide buffer rather than a separate tier of storage. Similar to how FAST (Fully Automated Storage Tiering) works, FAST cache promotes more frequently accessed 64KB blocks into high-speed FAST cache, thus accelerating all spinning disks downstream from the controllers.
At the time of acquisition, the VNX came with Samsung based flash drives. The drives are SATA based (despite what the documentation and sales order says). I pushed EMC to explain the discrepancy and was told that the ESSDs used in the VNX would not suffer any performance degradation due to it being SATA based. The interface does use the SATA 3.0 6Gb/sec specification and the drives use SLC-based flash technology.
BACK-END DISKS AND RAID
A short note about the back-end disks. The EMC VNX supports SAS, SAS-nearline and Flash disk. I personally would never use SATA in an enterprise production environment (as far as spinning disk goes). SATA is half-duplex, uses a less efficient communications protocol, has a less robust error detection and correction mechanism, and a considerably slower spindle speed. SAS is geared for the enterprise, where SATA is adequate for non-enterprise workloads.
EMC recommends (and I completely concur) that when creating storage pools using disks that are 2TBs and larger, that RAID-6 be used. The rebuild times on large capacity drives in conjunction with concurrent production I/O load, could place data at unnecessary risk. And lets not forget that it is possible to have an unrecoverable read error during a RAID-5 rebuild, which would render the entire RAID set inoperable. In the world of drastically increasing storage capacity needs, RAID-6 is definitely the new standard.
The end-to-end experience with the VNX has been excellent. Everything from racking to the installation of the PowerPath software went exactly according to plan with no unexpected gotchas. The new Unisphere interface is exceptionally easier to use than the older Navisphere interface. And although it has no bearing here, I must say the new chassis’ are very impressive looking. We feel that it was the right purchase of the right technology at the right time.
Further information on Fast Cache: EMC VNX FAST Cache Overview