HYBRID ARCHITECTURE IN THE CLOUD

Introduction Global storage was approximately 130 Exabytes in 2010 and is predicted to reach 40,000 Exabytes by 2020. This equates to 5.2 Terabytes (TB) for every human on earth. Data storage needs are expected to double every two years. It is predicted that nearly 40% of all data stored will be in the cloud. Much of the global cloud storage needs are a result of mobile and personal device usage, such as smart phones, tablets, and digital televisions. These devices have small internal storage and rely heavily on information pulled from the cloud. 1 This information must be online and quickly accessed, but what about the remaining data in the cloud? Is it accessed frequently enough to justify the cost to store it? There are many sources of cloud storage in the market currently. This paper will compare some of the most popular cloud storage options, as well as give a cost efficient option to tackle the problem of data being stored and not being used. A special type of customer will utilize cloud storage rather than onsite storage. Today, the bottom line is return on investment and, for those who use cloud services, economies of scale. There are many ways to go about deciding on the best internal schematics of the service and then also just as many options for a customer to decide on the right service. There seems to be a consistent bottleneck in two different areas: one, the access of data via network and, two, efficiency of the tiered architecture. There will be a consistent issue of access speed over a network slower than that on the customer s intranet. The second issue of a need for tiered data is a requirement to have economies of scale. A study by Carnegie Mellon makes a proficient claim that data does not need to be online at all times; in fact, it rather acts under Pareto principles and about 80% of the data is considered sleeping data, or data that is rarely accessed and/or changed. 2 The other 20% of data is the real time data that must be constantly and instantaneously accessed. To illustrate how that affects the cloud service industry, an EMC expert defines the time line for what is sleeping data: 80% of the world s data does not change after it is 90 days old. 3 So, in the cloud, what happens to data that is 90 days old? Will it stay online in an expensive chassis and then be readily accepted by the customer the instant the request is made? The logical answer is no. Business best practice for the cloud computing service and the majority of the IT world is to utilize a tiered architecture and strive for a hybrid media choice model to exploit the best possibilities available. Defining Terms Terms used through this paper are those that can be found within the dictionary provided by TechTarget, an independent and generic source for IT definitions. 4 One major defining term to be acknowledged is the different tiered architectures. For the purpose of this paper we will focus on the 1 Gantz, John and Reinsel, David. (December 2012). The Digital Universe in 2020: Big Data, Bigger Digital Shadows, and Biggest Growth in the Far East. Retrieved August 1, 2014 2 Schroeder, B., Gibson G, (February 2007) Disk Failures in the Real World: What does an MTTF of 1,000,000 hours mean to you. Retrieved August 1, 2014 http://www.cs.cmu.edu. 3 Herzog, E. (2010). Cloud Tiering Appliance. (EMC, Interviewer) 4 TechTarget. (2012). Find a Tech Definition. Retrieved June 18, 2012, from WhatIs.com: http://whatis.techtarget.com/ P a g e 1 Hie Electronics, Inc. Copyright 2014.

generic version of the differed tiers. Tier 1 will be considered online and instantly available with a window of about a 2 second delay. Tier 2 will be defined as nearline data and has a longer access speed ranging from 7 seconds to 2 minutes. Tier 3 will be considered as offline data and not readily accessible with a window ranging from 10 minutes to 3 days until accessed. Further definitions that cannot be found in this paper should be directed to the point of contact found at the end of the paper. Utilizing the Cloud The cloud service must use all the tiers to provide the customer the best cost option for their needs. Providing many options to best fit the customer s needs may overwhelm them, though many industry reports conclude that instant access to all 100% of their data is no longer a common need or desire. As we all know, there is the balance of cost versus performance. How do we approach such a complicated issue that is so personalized per customer? The answer is simple. As cloud providers, we provide them the guidelines necessary for all major requirements to be met. These include the following top 5 basic, yet must-have, requirements: Instant access to data accessed and added within the past 90 days Guidelines to access data that is older than 90 days Speed of access and the access privileges between the different clouds Disaster Recovery Data Model that matches their needs, minimum 2 copies available Disaster Location Management Plan, minimum of 2 different unique and independent sites Providing this information to the customer will allow for the cloud provider to deliver a service that meets the needs of the customer and meets their internal requirements to deter legal issues. Economies of scale are the purpose of the cloud. Economic feasibility can only be found by utilizing a tiered architecture. Having a massive front end will require a substantial and even exponential back end. This back end is typically found in Tier 2. Typically, Tier 2 can have a wide array of options: virtual libraries, slower disk arrays, or a large tape library system. Why not use an automated system to manage the movement of data from active data to sleeping data and back, i.e. Tier1 to Tier 2 and back to Tier 1? Nearline Data in the Cloud Each of the common cloud storage offerings is missing one possible option for its customers. Some do not need access to their data early and often when stored in the cloud. Hal R. Varian, dean of the School of Information Management and Systems at the University of California at Berkley, has found that 75 percent of all new digital data is fixed content. 5 What if this data could be stored on an energy passive media, still available for access within a minute or two? This type of option would still offer data availability but with increased data access time. Customers that would choose this type of option would 5 Salamone, Salvatore. (May 2002) Increasing static data prompts storage changes. Retrieved August 1, 2014, http://www.techrepublic.com/article/increasing-static-data-prompts-storage-changes/. P a g e 2 Hie Electronics, Inc. Copyright 2014.

be using the cloud for purely backup and disaster recovery options or do not need instant access to their data. Hard drives have only a 3-5 year life span, resulting in a large maintenance expense as well as colossal power consumption costs. Hard drive storage would not be utilized for storing this fixed, occasionally accessed data. This would allow for economies of scale to be seen by both parties. The TeraStack Solution would tier this infrequently accessed data while still allowing for access in a relatively short time. The Hybrid Solution Hie Electronics is a leader in the Active Archive data storage system technology industry and the manufacturer of the TeraStack Solution, an Active Archive processing, data backup and archiving system. It allows for application hosting, 50-100 terabytes of data to be nearline accessible on Blu-ray optical media, and up to 64 terabytes of data on online hard drives. The company has been recognized by Frost and Sullivan with the American Video Surveillance Product Innovation of the Year Award and the Data Storage Technologies Green Excellence Award in Technology Innovation, for its ability to reduce energy costs with its storage technologies. A leader in Sustainable IT technology, the Hie Electronics TeraStack Solution product line delivers up to a 90 percent energy cost savings when compared with that of current technology. The TeraStack Solution operates on 425 watts of power compared to up to 7,000 watts with other current storage options. Hie Electronics is an Energy Star Small Business Network participant and a stakeholder in the Energy Star Enterprise Storage Initiative. Comparison of Cost and ROI The following example (Figure 1.1) shows different providers that meet all the basic requirements: have a stable internet access with practical speeds (approximately 2TB/day), multiple data storage facilities, two thirds of the storage uses network egress, ten percent uses network ingress, normal Get and Head requests, and normal service requests. The pricing is a generalization for customers and based on customer experience or published pricing. It may not include all associated network costs. Service Provider Back-Up Required (TB) $ / Month $ / GB Amazon S3 50 $5,515.00 $0.11 (Amazon, 2012) Atmos (Value) 50 $7,500.00 $0.15 (Peer1 hosting,, 2011) Barracuda 50 $12,500.00 $0.25 Google (World Wide) 75 $13,968.12 (Google, 2012) $0.19 Figure 1.1: Example of Cloud Pricing Options When calculating total solution cost and determining ROI, data migration must be a fundamental part of the considerations. Figures 1.2, 1.3, and 1.4 graphically show the costs associated with data migration, the categories within the migration costs, and the required frequency of data migration depending on the type of storage media used. P a g e 3 Hie Electronics, Inc. Copyright 2014.

Figure 1.2: Cost of Traditional Storage Arrays, including Maintenance and Migration Source: Wikibon, 2014 According to Hitachi Data Systems 6, the overall average cost of enterprise storage migration easily exceeds US$15,000/TB. The following chart defines the categories of migration and their cost. Figure 1.3: Cost of Data Migration per TB Data Migration Source: Hitachi Data Systems (HDS) Figure 1.3: Data Migration Frequency vs Data Storage Media Source: Panasonic 6 HDS (2014, March) Reduce Costs and Risks for Data Migrations P a g e 4 Hie Electronics, Inc. Copyright 2014.

Deciding on what solution to provide the customer and how to achieve that must also require an internal hardware solution. As a cloud provider, this is an investment that cannot be a year to year decision or reaction process. Rather, it is best to look at the total cost of ownership. Let s take, for example, a comparison of the TeraStack Solution to a hard drive set up such as Dell s DR200v, combined with the annual software license to manage such an appliance (e.g. Dell s AppAssure Service). This comparison can be found below in Figure 1.4. In this comparison, factors such as the infrastructure, user application software, and management are nulled. Rather, it takes into account assumptions such as 24/7 operation with 99% uptime, average national electricity costs with no increase, costs to buy the product upfront, and maintenance/migration costs. Notice that there is a need to buy the hard drive setup every 4 years due to the industry standard failure rate for HDD. The comparative return on investment is noted immediately after the first data migration that is required by the cloud provider in Year 4. Putting the major components of TCO together: 1) HW acquisition and 4 year replacement, 2) plus energy to power and cool, 3) plus annual maintenance agreements, and 4) plus migration costs every 4 years. The TCO picture emerges as time progresses; using the hybrid architecture found in the TeraStack Solution will allow a substantial decrease in hardware costs and running costs. ASSUMPTIONS: Dell DR200v and DR6000 HW, coupled with AppAssure SW, 24/7 day operations, $0.12/kWh, installation + training + integration + shipping costs are the same for all, 15% maintenance, buy the product upfront, no tax, electricity does not increase per year, excludes depreciation, does not take into account future value or present value, there is need for some cables such as Ethernet port converter, every 4 years data migration for HDD. Management software like AppAssure is priced by front-end TB, the Backup & DR Suite features a capacity-based licensing model that enables customers to access the specific data protection capabilities they need without having to overpay for those they don't. Users can classify the criticality of their data and applications, and then use the specific product that best matches the recovery requirements of a given asset. Dell customers use the suite in any combination they desire. For example, users might initially leverage AppAssure to perform snapshots of missioncritical data every 15 minutes, and once that data ages, use NetVault Backup to move it to archive. P a g e 5 Hie Electronics, Inc. Copyright 2014.

Dell tiered pricing for the suite starts at just $5,000/TB for up to 5TB of front-end capacity, and, through volume discount pricing, reaches as low as $2,250/TB for users protecting more than 250TB of data. The TeraStack Solution does not require additional management software, such as AppAssure. The management software is a component of the complete system offering. Year Initial $ 1 2 3 4 5 6 7 8 9 10 HDD Single Instance SAN Turnkey (150TB Online System) K$ 150 TB Dell Hardware Cost 4YR Replacement (Incl. Cost Learning) DR2000v DR Server ($13,500/4TB) + 2xDR6000 (2x$59,772/9TB+8x36TB @$3695) $155.2 $93.1 $55.9 Training, Install, Integration, Cables, & Shipping $33.4 Training Costs $15.0 Installation Costs $15.0 Integration $1.5 Cables & Shipping $1.9 Maintenance Agreement $31.0 $31.0 $31.0 $31.0 $31.0 $31.0 $31.0 $31.0 $31.0 $31.0 $31.0 Migration Cost 32% $49.7 $29.8 $17.9 Dell Backup & DR Suite License $2,250/TB Software $337.5 $337.5 $337.5 $337.5 $337.5 $337.5 $337.5 $337.5 $337.5 $337.5 $337.5 Estimated Power Usage @ 7,870 w/ hr $8.3 $8.3 $8.3 $8.3 $8.3 $8.3 $8.3 $8.3 $8.3 $8.3 Estimated Cooling Costs $8.9 $8.9 $8.9 $8.9 $8.9 $8.9 $8.9 $8.9 $8.9 $8.9 Total $606.8 $385.7 $385.7 $385.7 $508.6 $385.7 $385.7 $385.7 $459.4 $385.7 $385.7 Total Accrued Costs (150TB Online System) $606.8 $992.5 $1,378.1 $1,763.8 $2,272.4 $2,658.1 $3,043.7 $3,429.4 $3,888.8 $4,274.5 $4,660.2 Cost per GB/month (150TB Online System) $0.55 $0.38 $0.33 $0.32 $0.30 $0.28 $0.27 $0.27 $0.26 $0.26 Year Initial $ 1 2 3 4 5 6 7 8 9 10 8.14.42 TeraStack Solution (142TB Online & Nearline Solution) K$ 142 TB Hardware Cost 4YR Server Replacement (Incl. Cost Learning) $241.6 $20.0 $12.0 Training, Install, Integration, Cables, & Shipping $33.4 Training Costs $15.0 Installation Costs $15.0 Integration $1.5 Cables & Shipping $1.9 Maintenance Agreement $25.8 $34.4 $34.4 $34.4 $34.4 $34.4 $34.4 $34.4 $34.4 $34.4 $34.4 Migration Cost 32% $77.3 $6.4 $3.8 Estimated Power Usage @ 425 w/hr $0.4 $0.5 $0.5 $0.5 $0.5 $0.5 $0.5 $0.5 $0.5 $0.5 Aditional Facility Engery x 2.84 (Includes Cooling) $1.3 $1.4 $1.4 $1.4 $1.4 $1.4 $1.4 $1.4 $1.4 $1.4 Total Costs (142TB Nearline System) $378.0 $36.1 $36.2 $36.2 $62.6 $36.2 $36.2 $36.2 $52.0 $36.2 $36.2 Total Accrued Cost (142 TB Nearline System) $378.0 $414.1 $450.3 $486.5 $549.1 $585.3 $621.5 $657.8 $709.8 $746.0 $782.2 Cost per GB/month (142TB Nearline System) $0.24 $0.13 $0.10 $0.08 $0.07 $0.06 $0.06 $0.05 $0.05 $0.05 Figure 1.4: Cloud Comparison Cost per GB over 10 years Source: Hie Electronics For a modest 150TB ten year set of data without deduplication, the TCO for conventional HDD based technology like that offered by Dell starts at $0.55/GB per month for a single instance in year one. Assuming 4 year replacement and a HW cost learning curve, this evolves into $0.26 per GB per month by year ten (including maintenance and migration costs). The initial investment of the TeraStack Solution made by the cloud provider will allow for far cheaper costs per GB per month in year one of $0.24 and by year ten of $0.05 per GB per month compared to other cloud options; investing in the TeraStack Solution is a win. P a g e 6 Hie Electronics, Inc. Copyright 2014.

Speed of a Download What performance is required by your customer? Let s take a look at a customer example with three downloads of the same 15MB file copied to each cloud storage service provider. After the download was complete, the customer cleared the Gladinet cache and then initiated the Command Prompt for the next downloads via specific cloud provider. 7 Upload (sec) Download # 1 (sec) Download # 2 (sec) Download # 3 (sec) Average Download Amazon S3 83 18 20 17 18 AT&T Synaptic 84 37 49 45 43 Storage Google Storage 94 15 17 17 16 for Developer Peer1 CloudOne 85 20 16 19 18 Windows Azure 86 48 72 52 57 Blob Storage Figure 1.5: Example of Customer Experience to Speed of Access to Downloaded Data Noting the time difference and the way the study was done, the difference can be drastic. But who won? Was it Google or Amazon? The question has to be answered on how the customer set up each service and if it was identical. If all is equal, then the real winner is the provider that gave the basic requirements and got the greatest profit based on performance. More than likely Azure won the battle by using their tiered architecture and automated data retrieval system. Using an automated solution, such as the TeraStack Solution, there is a win-win that provides the data to the customer well within the specs and at the lowest cost to the provider. Other Cloud Provider Concerns Application agnostic is another concern to some cloud providers; can the cloud provider use any application desired for internal results and fluidity? The answer should be yes and explains why larger cloud solutions typically have sections of their infrastructure designed and separated based on purpose and applications installed. The data environment is always changing. It is time to take an approach that can mold with how data acts, and a hybrid solution that recognizes the Pareto principle for 80% of the data that is sleeping. It is time to counteract the problem that all companies face: how to keep 100% of the data alive as long as possible or as long as best practices are achieved. Summary The cloud is now facing what all data centers face today; there is a must to have a tiered architecture of data storage for the most efficient use of resources. How to go about this is always the balance between costs and performance. The cloud provider must assess what the best solution is to view the life of the 7 Du, Haikun, Huang, Zhihui Huang, Xu, Gang, Gladinet, Inc. US Patent US20100241731 A1. Filed March 16, 2010. P a g e 7 Hie Electronics, Inc. Copyright 2014.

customers data and take this window of time as the basis to depreciate the storage costs. Based on industry average, a typical document (excluding emails) will last between 7-10 years. Depreciating the cost to store this normal data should be accounted over a 10 year period. The most proven method to support the lowest total cost of ownership is to use a hybrid architecture that exploits online, nearline, and offline capabilities. The solution that is there for cloud providers that look for these economies of scale and economic efficiency can be found in the superior hybrid architecture TeraStack Solution. Consider testing this innovative key for your cloud or cloud provider for the data of your future. For more information about Hie Electronics and the innovative TeraStack Solution, visit the company s website at www.hie-electronics.com. For further questions, please contact Hie Electronics at info@hieelectronics.com or (972)542-2327. P a g e 8 Hie Electronics, Inc. Copyright 2014.

Works Cited Amazon. (2012). Amazon S3 Pricing. Retrieved June 17, 2012, from Amazon Web Services: http://aws.amazon.com/s3/pricing/ Du, Haikun, Huang, Zhihui Huang, Xu, Gang, Gladinet, Inc. US Patent US20100241731 A1. Filed March 16, 2010. EMC. (2010, April). Retrieved June 17, 2012, from EMC Centera Contect-Addressed Storage System. : http://www.emc.com/collateral/hardware/data-sheet/c931-emc-centera-cas-ds.pdf Gantz, John and Reinsel, David. (December 2012). The Digital Universe in 2020: Big Data, Bigger Digital Shadows, and Biggest Growth in the Far East. Retrieved August 1, 2014 Google. (2012, June 16). Google. Retrieved June 19, 2012, from Pricingandterms: https://developers.google.com/storage/docs/pricingandterms Herzog, E. (2010). Cloud Tiering Appliance. (EMC, Interviewer) Peer1 hosting. (2011). CloudOne Storage powered by EMC Atmos. Retrieved June 18, 2012, from Ping & People: http://www.peer1.com/sites/default/files/pdf/datasheets/cloudone_storage_consumer.pdf Salamone, Salvatore. (May 2002) Increasing static data prompts storage changes. Retrieved August 1, 2014, http://www.techrepublic.com/article/increasing-static-data-prompts-storage-changes/. Schroeder, B., Gibson G. (February 2007) Disk Failures in the Real World: What does an MTTF of 1,000,000 hours mean to you. Retrieved August 1, 2014 http://www.cs.cmu.edu. TechTarget. (2012). Find a Tech Definition. Retrieved June 18, 2012, from WhatIs.com: http://whatis.techtarget.com/ Wikibon (2014, Apr) The Cost of Storage Array Migration in 2014, David Floyer: http://wikibon.org/wiki/v/the_cost_of_storage_array_migration_in_2014 HDS (2014, March) Reduce Costs and Risks for Data Migrations: http://www.hds.com/assets/pdf/white-paperreducing-costs-and-risks-for-data-migrations.pdf Dell Software Revealing Backup and DR Suite at $2,250/TB to Protect 250TB Including AppAssure, NetVault Backup and vranger with single capacity-based license. This is a Press Release edited by StorageNewsletter.com on 2014.08.18: http://www.storagenewsletter.com/rubriques/software/dell-software-revealing-backupand-dr-suite-at-2250tb-to-protect-250tb/ Dell, DX6000 and DR2000v Launch of Backup Appliances (2014) http://www.dell.com/learn/us/en/uscorp1/press-releases/2014-07-15-dell-software-appassure-dl1000- dr2000v-backup-appliance http://searchdatabackup.techtarget.com/news/2240220506/dell-gives-netvault-backup-a-facelift-addsenterprise-dr-disk P a g e 9 Hie Electronics, Inc. Copyright 2014.