The Silk Platform is designed to optimize database workloads by virtualizing native cloud resources to offload and accelerate applications. Since Silk presents as if we are volumes, we are often compared to native data storage options.
Misunderstandings about cloud IO often revolve around the notion that a lower initial price tag signifies the best bargain. However, as with many technological aspects, a comprehensive evaluation is essential to discern true value. A case in point: a preliminary cost analysis might suggest that Silk’s pricing in Azure is like some of its high IO counterparts, as detailed below. Yet, there is more to the story than meets the eye.
In the following table, you’ll find the most common high IO solutions for relational workloads on Azure IaaS (Infrastructure as a Service) for a 10TB/TiB challenge.
|High IO Solution||Details||IOPs||Throughput||Total Monthly Cost|
|Ultra Disk||5*2TB each @ 100k IOPs & 2GB/s||500K IOPS||10GB/s*||$29,542.63|
|ANF (Azure NetApp Files) Ultra||100 TiB Capacity Pool||640K IOPs||10GB/s**||$40,216.97|
|Silk||3 c.nodes and PV2||1m IOPS||10GB/s***||$24,696.00|
*High IO limit posed by NIC at VM (Virtual Machines) level and limit dependent upon the VM sku.
** 100TiB of ANF ultra must be provisioned to achieve 10GB/sec
*** A single Silk POD can sustain up to 26GB/sec and 2m IOPS with 8 c.nodes. 3 c.nodes can match the maximum performance of a single Azure instance
All costs east us region presuming zero discount from Vendors with 1 Year commitment.
Although the throughput claims are similar, all hovering around the 10GBps limit, Ultra Disk is not really a competitor due to how limits are placed at the attached storage layer vs. the other three in the list.
Understanding VM vs. Solution Limits
Microsoft’s VM documentation serves as a valuable resource for understanding the IO throughput egress limits of your selected VM. For Silk, Egress limits only affect Writes, while Read performance to Silk often hits the limits of the physical hardware. In the subsequent example, I will highlight the upper limits of the Eds v5 series:
|Size||vCPU||Memory GiB||Temp storage (SSD) GiB||Max data disks||Max temp storage throughput: IOPS/MBps||Max NICs||Max network bandwidth (Mbps)|
When evaluating storage and network capabilities, it is crucial to differentiate between throughput measures. Attached storage throughput is measured in Megabytes per second (MBps), whereas network egress limits are denoted in Megabits per second (Mbps). Since a byte comprises 8 bits, converting Mbps to MBps becomes straightforward:
Mbps ÷ 8 = MBps
Take the Eds v5, for instance. This highlights the formidable IO throughput of the E104id_v5 VM SKU in terms of network attached storage. Though the attached storage caps are at 4000 MBps, the network bandwidth rockets to an impressive 100000Mbps—equivalent to 12500MBps! However, the Ultra Disk does not fully harness this potential, being restricted to the 4000MBps limit of attached storage.
Compression and deduplication offer significant storage savings and improved performance without compromising data quality. Although Ultra disk does not support this feature and ANF makes no mention of it, its inclusion in Silk reduces storage needs. Specifically, a deployment model with a size of 10TiB benefits from a 2:1 data reduction, leading to substantial cost savings—bringing the total down to just $17,350.00.
Furthermore, depending on the workload(s), Silk often provides even greater data reduction rates, further increasing the cost efficiency of the solution.
Volume snapshots are now widely embraced by businesses, offering application-consistent snapshots without the added burdens of conventional backup approaches. Beyond serving as a defense against ransomware, snapshots can also function as tiered disaster recovery solutions, cloning tools, and more. Integrating this feature into Silk eliminates the need to buy an additional service or product to achieve application-consistent volume snapshots for relational or other application workloads.
Many high I/O solutions offer cloning features, and several tout their capabilities for “thin cloning.” However, only a handful truly delivered on this promise. For numerous relational workloads, creating multiple copies of multi-tiered environments—spanning databases, applications, and more—often constitutes the bulk of monthly IT resource consumption. Alongside the drain on technical resources, there is a tangible cost associated with storage. Without the option of thin cloning, resorting to “thick” clones can cause storage costs to skyrocket, often quadrupling or even more.
Silk offers an efficient solution with its thin provisioning, allowing for read/write clones that cater to a variety of needs, from development and patching to testing and reporting. By utilizing thin clones through this provisioning, users can achieve all these functions while consuming minimal additional storage, leading to significant cost savings.
For our sample deployment involving a 10TB/TiB workload, we anticipate a need for 4 monthly clones stemming from the 2TB/TiB production workload. This would necessitate doubling the allocated space for Ultra Disk, compelling the use of traditional backup/cloning methods. Additionally, ANF would require thick clones derived from their own volume snapshot technology. Conversely, Silk can harness all tiers using thin clones, effectively obviating the need for supplementary storage, thereby reducing associated monthly expenses.
|High IO Solution||Ending Storage||Initial Cost||Ending Cost|
|Ultra Disk||5*2TB each @ 100k IOPs/2GBPs
Azure Files for Azure Backup 5TB
|ANF Ultra||100TiB Capacity Pool
Additional Capacity- 10TiB Premium required, 32TiB available.
|Silk||3 c.nodes and PV2 with 2:1 data reduction and thin provisioned snapshots||$24,696.00||$17,350.00|
The Total Cost of Ownership (TCO) of Silk for migrating workloads to the cloud is a comprehensive assessment of all costs associated with deploying, operating, and managing workload resources, not just initial entry. Often, when organizations first look at migrating to the cloud or selecting a cloud service, they might be swayed by upfront costs alone. However, to truly understand the financial implications, one must delve deeper into the TCO. It accounts for not only the visible costs but also hidden expenses such as data reduction, backup, and cloning solutions. Therefore, like the age-old adage “never judge a book by its cover,” organizations must never judge cloud offerings based purely on initial prices. By focusing on the TCO, businesses can identify and leverage opportunities for significant end savings in cloud costs, ensuring they obtain the best value for their investments.
One final way of reducing TCO in the cloud is to commit to longer usage/reservations, this is ideal for large database workloads that typically have a longer residency. Unfortunately, there is typically near zero cost benefit for reservations on high-performance cloud data platforms. Silk, however, can benefit from the discounts available to Azure Compute and software licensing. With three-year reservations, the total cost of the Silk configuration above reduces further to $12,852 / month.
Looking to Future-Proof the Cost of Your Cloud?
Discover the TCO in the cloud of SQL databases at scaleWatch Now