Post by Chris Mellor (thank you) over at ElReg with a good look under the hood of HUS
Not your average dual-controller array: that’s the conclusion after a canter through HDS’s new HDS Unified Storage (HUS) 100 series array controller architecture.
There are three different controllers in the series for block access and two for file access – NAS heads. Both types use non-commodity hardware.
For block access all three models have symmetric, dual active-active controllers with mirrored cache. The HUS110 and 130 have 2U controller enclosures and the 150 has a 3U one.
A block controller contains a dual core Xeon processor with 3.5GB of RAM. This is responsible for all I/O operations for that controller for the group of LUNs on its management list. It is accompanied by a Data Control ASIC handling RAID control and cache memory, which has associated NVRAM and DDR3 cache.
There is a PCIe 2.0 bus within the controller, with 8-lane connections. There can then be two host I/O Module controllers; one for Fibre Channel and the other for iSCSI. HP calls these Tachyon processors.
At the back end there can be two SAS controllers, each with 8 x 6Gbit/s links to local storage and to expansion trays connected via 16 x 6Gbit/s redundant SAS links.
The HUS 110 supports only 8 SAS links on the back end: a slower processor, a smaller cache and fewer host channels than the HUS 130, which has 16 SAS lines but fewer disks than the HUS 150.
The 2U HUS 110 and 130 controller enclosures can contain 12 x 3.5-inch hard disk drives or 24 x 2.5-inch disk or solid state drives.
For capacity expansion there can be 2U enclosures, holding 24 x 2.5-inch SAS disk or solid state drives or 12 x 3.5-inch 7,200rpm SAS disk drives, or 4U trays holding 48 x 3.5-inch 7,200rpm SAS disk drives.
The HUS 110 only supports the 2U expansion enclosures, and 120 disk drives in total,while the HUS 130 and 150 support both types of expansion tray, with the 130 supporting up to 264 disk drives and the 150 up to 960.
A Josh Krischer white paper (PDF) on HUS provides much more speeds and feeds information.
In principle, any host port can be used to access any LUN managed by either controller, without any need to change LUN ownership. This is the basis of a load-balancing capability whereby LUNs are moved from an overburdened controller in an active-active pair to the other, less burdened one.
Krischer’s white paper also states:
A host accessing a LUN through a Controller-0 can actually have most of the I/O request processed by Controller-1 with little intervention by the processor in Controller-01. This function allows the use of host-based operating system path management and load balancing such as Microsoft Windows MPIO, Solaris MPxIO/traffic manager, IBM AIX MPIO, Linux Device Mapper, Veritas DMP or Hitachi Dynamic Link Manager with minimum overhead penalties. This feature is in particular useful in automated failover.
Files and objects
Contrary to the way El Reg storage desk outlined it in the history of HUS story, the file controllers are essentially BlueArc-based NAS heads using that HNAS product’s FPGA hardware architecture and SiliconFS object-based file system. The HUS 110 and 130 use the 3080 HNAS controller as dual modules that are cluster-capable. The HUS 150 uses the HNAS 3090 with dual-node and 4-node clustering.
The HUS intelligent file tiering feature can move selected files or folders to an appropriate tier of storage, both internally to HUS or, if they are inactive, to an external tier, using content-aware rules and a policy manager. This has the happy benefit of decreasing the size of the overall HUS dataset and thus reducing its data protection resource requirements.
Object storage on the HUS arrays is handled by the NAS heads. Objects are stored as files with specific metadata, which cannot be added to by customers. They are not stored with hash addresses, erasure coding or any addressing scheme related to their content. So these are quasi-objects. Nevertheless they exist as a single logical searchable object pool as HDS’s Data Discovery software can search their metadata and content along with object content and metadata on the HCP true object storage system.
HUS objects cannot be exported to HCP; the HUS automatic tiering system can’t do this, says HDS. Its EMEA chief technical officer Bob Plumridge wouldn’t commit to HDS making this possible. Clearly such a development would be in line with HDS’s ambition to provide a seamless single file, block and object storage facility.
Plumridge said he had seen more change at HDS in the past two years than in the previous 10. It’s apparent that the product range has and is broadening. As well as the low-end F-series filer being added to the HDS product set, a high-end scale-out file product based on Parascale technology should arrive before the end of the year, taking HDS into Isilon territory.
