A Comparative Economic Analysis
See Part One of this Series: How Google Apps Threatens Microsoft.
Contents |
Research Premise
Exchange 2003 was a very I/O intensive solution that typically necessitated the use of a large number of lower capacity high performance disks. Poorly designed storage for Exchange 2003 often left organizations vulnerable to hardware failures and unpredictable performance. As a result, many practitioners chose to deploy SAN-based storage for Exchange 2003. With Exchange 2007, and the recent introduction of Exchange 2010 (E2K10), Microsoft has made significant enhancements to its flagship messaging solution. This is critical because Exchange is an integral part of the business in many companies, often supporting revenue-generating activities.
Many of these improvements affect storage and I/O to the point that Microsoft is now strongly recommending users no longer deploy SAN for Exchange but rather default to direct attached storage (DAS) as the preferred infrastructure.
Combined with Microsoft’s Exchange enhancements, the benefits of DAS are that it is less complex to install than SAN and for smaller shops it will be less expensive on both a CAPEX and OPEX basis. However the story is not that simple. While E2K10 delivers substantial benefits to clients, particularly by simplifying its clustering approach, we believe Microsoft’s recommendation to deploy DAS on Exchange is misguided for many larger clients.
In comparing the choice of DAS versus SAN practitioners should consider three primary factors:
- Can the SAN be shared across multiple systems in the application portfolio? The more a SAN is leveraged the lower the cost overall.
- What are the availability requirements of an organization’s email application? Is it mission critical?
- Is virtualization a fundamental organizational strategy?
As summarized in Figure 1, users have several options with regard to storage deployments. Larger Exchange installations should use caution when considering Microsoft’s advice to use DAS for Exchange because DAS is best suited for smaller installations that wouldn’t leverage sharing across applications. In larger environments, DAS will increase costs, create information availability exposures, and reduce operating leverage.
The two most important financial considerations for Exchange customers pertaining to the choice of DAS or SAN are:
- The size of the installation (e.g. number of seats, number of Exchange servers, amount of storage capacity).
- The asset utilization that can be achieved by sharing SAN capacity across multiple applications.
Our analysis shows the crossover point where SAN becomes less expensive than DAS is just under 1,000 Exchange seats and approximately 15 TBs of shared storage; where the same SAN infrastructure supports at least two or more applications. For installations with more than 1,000 seats, DAS costs are often noticeably higher than SAN for both OPEX and CAPEX.
This research note is the second in a two-part series and presents an economic model to help practitioners quantify the financial and business value considerations for choosing DAS or SAN in an on-premise Exchange environment. Part One of the series explored how Google Apps is impacting Microsoft’s on-premise and Cloud-based Exchange strategies with implications for storage decision-makers.
Background: Exchange 2003 – 2010
With Exchange 2003, decisions about storage were straightforward. For enterprise deployments with availability and performance SLAs, SAN was the only viable storage choice. Exchange 2007 introduced a new 64-bit architecture that reduced IOPs by 50 percent. Also, E2K7 includes a continuous replication capability which provides higher levels of resiliency and gives users more options.
Exchange 2010 extends the E2K7 functional roadmap by further reducing I/O requirements another 50 percent (or 85% relative to E2K3). In addition, E2K10 introduced two new storage-related functions:
- Database availability groups (DAGs), which provide redundant copies of mailbox databases with continuous replication and automatic recovery;
- Database-level failover, which removes the need for clustering and improves overall uptime.
These enhancements have provided more options for users and support Microsoft’s broader Exchange objectives, which are to simplify infrastructure and have all Exchange databases on DAS.
The question for organizations is will the enhancements in E2K10 and a DAS strategy provide the lowest cost of ownership?
Cost Analysis: Comparing DAS and SAN for Exchange
To answer this question, Wikibon constructed a model to evaluate the cost of DAS versus SAN for Exchange. To accurately assess the financials, we developed a return on assets (ROA) model. An ROA model differs from a classical ROI model in several ways as shown in Figure 2.
- An ROI model evaluates how quickly and by what magnitude benefits accrue to offset initial startup costs combined with ongoing operational expenses. It’s an excellent approach for evaluating a one-time investment that has relatively small ongoing CAPEX.
- Classic ROI models are not as useful, however, when an investment is constantly taking on new injections of capex over time—such as with email infrastructure where upgrades of technology are regular over time. Also, email infrastructure is often shared, so isolating a piece of the investment (e.g. the Exchange infrastructure) doesn’t tell the whole story.
- The primary objective of an ROA model is to provide a budget view and understand how investments impact currently installed assets and how future rolling investments will affect costs and business value over time. It attempts to model a real customer with myriad assets installed over a number of years. Importantly, an ROA model takes into account the fact that there exist assets on the books with value that will not be de-installed until their useful life has expired.
- An ROA model allows clients to compare a strategy of installing cheaper infrastructure with one that may appear more expensive initially but leverages the value of existing assets over time. An ROA model can differentiate between investments that have no impact on installed assets and those that do.
An ROA model for example assumes a mix of installed arrays that vary by age. Each system has different initial costs, OPEX, power/cooling, staffing, etc., and over time specific assets are replaced by new assets. Understanding true costs requires modeling a complicated series of interrelationships that more accurately reflect real-world installations.
In summary, the benefit of an ROA model is it provides a way of assessing the profile of an installed customer and understanding the impact on the customer over a period of time.
We have modeled the full Exchange server, storage, server and client software, archiving software, maintenance, environmental, and staffing infrastructure required to deploy 2010 but chose only to expose the storage-related aspects of the model for this research note.
Figure 3 shows the cost per user per month across a wide range of Exchange installations (scaled by number of seats). The model shows that in smaller installations, DAS is cheaper than SAN, but as the number of Exchange users grows SAN, which can be leveraged across multiple applications, becomes significantly less expensive than DAS.
- The chart shows a budget view (i.e. depreciated) which includes all storage components of hardware, software, maintenance and staffing required to support the number of users indicated on the chart.
- It includes older assets depreciated over a 48 month period using a straight line method.
- Average age of an array is 4.7 years based on Wikibon analysis of installed base data.
- SAN equipment analyzed: iSCSI 200, 1,000 seats; EMC CLARiiON (FC) 5,000, 10,000 seats; EMC V-Max (FC) 25,000, 60,000 seats.
Figure 4 depicts detailed cost breakdowns by line item for the storage budget in an Exchange 2010 example using 10,000 users. As indicated, the ROA model provides a view of the impact a technology has on installed assets and includes, for example, power, cooling, and space costs for older equipment, which tend to be higher than newer equipment. We feel this is a more realistic view of total costs in Exchange environments than a ‘greenfield’ ROI model would show (i.e. a model that assumes assets that still have book value are retired).
- The ROA model assumes SAN infrastructure supports three or more applications.
- Utilization assumptions for DAS and SAN are shown in Table 2.
- Power cost assumptions $.09 per kWh.
- Space costs assume $150/sq. ft/annum.
- Capacity start/end; DAS 200TB/643TB; SAN 200TB/380TB.
Also, practitioners in our survey indicated that the improved availability may warrant the extra cost of a SAN in situations where DAS would be less expensive. We believe arguments that suggest DAS is less exposed than SAN because loss of a SAN means loss of an entire data store whereas loss of DAS only means a partial outage of the data store are specious. A well-designed SAN will provide higher availability than DAS. A poorly designed SAN should be avoided at all costs.
The bottom line is that for smaller configurations where SAN infrastructure cannot be shared across the application portfolio, DAS is less expensive than SAN. The crossover point between DAS and SAN is around 1,000 seats, and as the number of Exchange seats grows, DAS becomes more expensive than SAN. Organizations should consider availability as an additional factor in the equation.
The CAPEX Myth
One of the common misconceptions in the DAS-versus-SAN debate is that DAS is less expensive on a CAPEX basis. While it’s certainly true that the street price of a DAS TB is cheaper than a SAN TB, this simple calculation ignores the practical realities of customer installations. Two factors make SAN capex cheaper than DAS at scale:
- Because SAN capacity can be shared across multiple applications, its costs will often be less from a CAPEX perspective because additional capacity needs can be met from the existing SAN, thereby avoiding incremental capital expenses over a period of time. Because DAS is not a shared resource, its capacity is not fungible, and an environment with several applications using DAS inevitably needs a higher total storage capacity than one in which the same applications share a SAN.
- Storage capex costs must include additional server horsepower required to manage storage, including caching, copy services, software RAID, and replication.
Further supporting this trend is virtualization. Microsoft is recommending organizations avoid virtualizing Exchange 2010. However our survey respondents indicated that they typically are virtualizing existing Exchange infrastructure with a shared SAN model and plan on virtualizing E2K10 using SAN.
Using the ROA methodology, we can model the actual CAPEX cost for an Exchange 2010 installation over a four-year period. Figure 5 shows that in larger shops that can leverage SAN capacity in more places than just a single Exchange application, capacity acquisitions can often be met by utilizing existing SAN resources. The ROA model normalizes the scenario so that we can compare the two alternative storage architectures on an apples-to-apples basis.
- Existing installed SAN capacity can often meet the needs of new Exchange deployments.
- Assumes utilization rates shown in Table 2.
- Assumes multiple applications can leverage the SAN at 5,000 users and above.
- Assumes iSCSI SAN at 5,000 seats and below.
- Includes additional server capacity required to manage storage.
Recommendations for Practitioners
Blindly heeding Microsoft’s advice to only use DAS and avoid SAN for Exchange 2010 installations can increase costs significantly for larger installations—both OPEX and CAPEX. Users should carefully consider the parameters that will affect cost when evaluating DAS vs. SAN for Exchange 2010 including:
- The size of the installation in terms of users and capacity,
- The degree to which infrastructure can be shared across multiple applications,
- Expected growth rates,
- The degree to which server virtualization is a fundamental strategy.
To the extent that installations are large (more than 1,000 seats) and multiple applications can share a storage infrastructure, users will typically find SAN to be cheaper than DAS on an OPEX basis. Also, to the degree that an organization is growing and a one-time purchase of infrastructure is unlikely to meet demand over a five-year period, organizations will find that SAN will be cheaper than DAS on a CAPEX basis (for installations of more than 1,000 Exchange 2010 users).
While E2K10 has brings substantive improvements relative to E2K3, these enhancements do not negate the economic realities of shared infrastructure. Smaller shops however will be well served by Microsoft’s improvements and by DAS.
Additional important considerations not assessed in this study are:
- Availability: Exchange 2010 continues to use IP-based replication which is ‘lossy’, and users should understand their RPO and RTO requirements before investing in infrastructure.
- Politics Often organizations prefer that a single group (e.g. the Exchange group) manage the entire hardware and software stack. These groups may not want to go to a separate shared SAN organization for storage deployments. While these are important considerations, users should understand the cost implications.
- Virtualization: Microsoft is strongly urging users not to virtualize Exchange 2010. In Wikibon’s view, Exchange 2010 is a perfectly viable application to be virtualized over the next five years and will gain further cost and utilization advantages when combined with SAN.
Finally, iSCSI is gaining momentum as a protocol and, increasingly, smaller environments can combine the simplicity of iSCSI with Exchange 2010 and continue to receive the benefits of SAN.
Action Item:
Footnotes: Related Research