Most datacenters jumped on the virtualization bandwagon years ago, and while the premise of server consolidation allows for efficiency, it also creates new storage growth patterns. Virtual and physical machines need to be backed up in accordance with service level agreements. Anybody that has worked with recovery point objectives knows that it is certainly more comforting to rely on a disk array than tape repositories. Tapes are expensive, fail frequently, and are cumbersome to store.
There are tons of enterprise grade tier 3 storage arrays on the market, but the objective of this series is to present a proposal for building an in-house solution that meets and exceeds the "out-of-the-box" solutions. In the series we will discuss the storage proposal, configuration, performance benchmarking, and maintenance.
Architect, develop and deploy a scalable and cost-effective backup storage solution.
Generally, a tier 3 storage system is comprised of multiple inexpensive disk arrays. The arrays should be
highly scalable, but not necessarily highly available. Redundancy within the array should allow for data
protection in the event of multiple drive failures.
Currently, we use an Exagrid system as a backup repository for both headquarters and disaster
recovery sites. Backups are first written to the primary Exagrid array and subsequently replicated to the
disaster recovery site. Exagrid's architecture premise is based on a grid of "full servers" rather than a controller/shelf configuration. They use post-process deduplication. The grid architecture allows you to scale out by simply purchasing an additonal Exagrid unit and adding to the pool.
To read more about the Exagrid solution, check out their site here.
The proposed solution is to build two identical tier 3 storage arrays – one for headquarters and one for
disaster recovery. Each array will be comprised of a single controller and multiple drive shelves. Initially we will configure one drive shelf for each array. The drive shelf will hold 45 drives, and provide roughly 80TBs of usable storage in a RAID6 configuration. As capacity requirements increase, additional drive shelves can be purchased and daisy chained to the existing configurations without the need for additional controllers (up to 4 shelves per controller).
Actual Hardware Specs:
- Supermicro SuperServer 6027R-TDARF - 8x SATA - 8x DDR3 - 740W Redundant
- 2 x Six-Core Intel® Xeon® Processor E5-2620 2.00GHz 15MB Cache (95W)
- 8 x 4GB PC3-12800 1600MHz DDR3 ECC Registered DIMM
- 2 x Intel 520 Series Cherryville SSDSC2CW180A3K5 2.5" 180GB SATA III MLC Internal Solid State Drive (SSD)
- LSI SAS 9211-8i SAS 6Gb/s PCIe 2.0 8-Port Host Bus Adapter
- LSI SAS 9205-8e SAS 6Gb/s PCIe 2.0 8-Port Host Bus Adapter
- No Windows Operating System (Hardware Warranty Only, No Software Support)
- Three Year Warranty with Advanced Parts Replacement
- Supermicro SuperChassis 847E26-RJBOD1 - 4U - 45 x 3.5" SAS - Dual Expander - 1400W Redundant
- 45 x 2.0TB SAS 2.0 (6.0Gbps) 7200RPM - 3.5" - Western Digital RE SAS WD2001FYYG
- Supermicro System Cabinet Front Bezel MCP-210-84601-0B (Black)
- Quick-Release Rackmount Mounting Rails - 26.5" to 36.4" (Included) (MCP-290-00057-0N)
- 0.68M External SAS Cable (SFF-8088 to SFF-8088)
- Thinkmate® Three Year Warranty with Advanced Parts Replacement
So how do the two hardware solutions match up? Here's the best part -- both hardware vendors are identical. Exagrid sources SuperMicro components for all of their systems. Don't believe me? Take the faceplate off of your Exagrid unit and look at the chassis and drive caddies -- classic SuperMicro. So if the quality of the hardware is really the same then it boils down to horsepower. As far as horsepower goes, our standard EX10000E Exagrid has two quad core, middle grade, Intel processors and 8GBs of RAM -- certainly nothing to write home about. Our new solution has two 6 cores Intel processors and 32GB of RAM, additionally we will be running two Intel 520 Cherryville SSDs for caching purposes. In theory the hardware vote is going to the home built solution - later this week, I will get some real performance numbers.
Tomorrow I will introduce our OS of choice for the new backup repository and speak generally about our functionality requirements.