A must-read blog for IT professionals, solution architects, and storage administrators. The time to make your applications run faster is now.
Applications are one of the most mission-critical elements in today’s digital economy, and poor performance results in bad user experiences leading to productivity and revenue losses for your business. More times than not, sluggish and unresponsive applications can be blamed on either underpowered processors, network bottlenecks, or storage inadequacies. In this blog, we shall focus on storage – the least understood and arguably the most critical piece to the application performance puzzle. Some who dabble in storage will tell you flash solves all. But that is only a fraction of the solution, and only addresses the endpoint.
Beyond Shiny New Objects: Flash, NVMe, and 3D XPoint technologies all have a role in shaping the performance envelope of storage infrastructures and ultimately the response and throughput of the applications running on top of them. Yet anyone who follows the industry recognizes them as temporal components soon replaced by even faster, cheaper media and more advanced methods of connectivity.
The four best practices presented below contemplate not only the endpoint where data rests, but the route it takes there and back, and how its value changes over time. We will discuss data placement, path redundancy, fault domains and access speeds, as well as factor in cost tradeoffs, business continuity imperatives and ongoing modernization challenges to help you keep your applications running at peak performance.
Best Practice #1: Match Storage to the Value of the Data
In the world of applications, data is king. The faster the data processing, the faster the applications run. With increasing I/O activity, even your most performant storage devices tend to become overloaded and consumed by competing demands from low priority workloads and aging data. For most organizations, adding more expensive storage gear each time this happens is not a viable option. What if you could ensure only the most valuable – in other words, the most actively used – data gets stored on your premium storage, and infrequently used data is only stored on secondary and tertiary storage? Then I/O demands on your primary storage will reduce and so will capacity consumption, which in turn will keep the device running faster.
This is where data tiering comes to your aid. Tiering is the process of moving data between storage devices based on how frequently the data is accessed.
- If the data access temperature is high (hot data), keep that data on your fastest tier.
- When data access temperature cools, move that data to secondary storage so you can free up space on your primary storage.
DataCore SANsymphony is a software-defined storage (SDS) solution that leverages built-in machine learning to automatically detect data access patterns and determine when and where to place data. Without causing any disruption to the application accessing the data, SANsymphony automatically tiers (moves) hot data to primary storage and warm and cold data to secondary and lower-cost storage. This allows you to save more space on your premium hardware, avoid I/O bottlenecks and capacity overload, and improve overall responsiveness. Additionally, this also saves you money as you do not have to keep throwing hardware at your capacity and performance problem. Provision your valuable data the fastest storage it deserves.
Best Practice #2: Ensure an Alternate Data Path for Uninterrupted Access
Storage technology upgrades occur quite frequently out of necessity. The most common one is replacing slow storage hardware with modern, high-speed gear. Such storage refreshes typically require planned outages and prolonged periods of data migration. Both have major repercussions on data access, either interrupting it or slowing it to the point of being unusable. It will require disconnecting the old device from the application’s data path and decommissioning it, then integrating the new storage, and then pointing it to the application’s host server. Also, you need to factor in the manual oversight of the lengthy data migration project that needs to happen between the old and new device.
To overcome this disruption, DataCore SANsymphony provides you with the best practices of creating path redundancy and fault tolerance for your storage infrastructure by:
- Leveraging multi-path I/O techniques such as ALUA (Asymmetric Logical Unit Access) from the operating system or hypervisor
- Synchronously mirroring your data between suitable storage devices even from different manufacturers
- Automatically copying application data in real time to the mirrored device for redundancy
- Automatically failing over to the mirrored copy when it’s time to change/refresh the old storage device
- Ensuring there is no data access disruption for the application during the process of hardware upgrade
- Transparently evacuating data from the old device and moving it to the new one (or balances it across other storage in the pool) once the new and upgraded device is in place
- Automatically switching back the data path and pointing to the new storage device from the failover device
During the course of this entire project, there is no application downtime experienced. Injection of a new technology and data migration happens in real time without applications or users realizing any change happening at the back end. Modernize your datacenter while ensuring no performance degradation or downtime.
Best Practice #3: Leverage Parallelism for I/O Processing
The rise in popularity of big data analytics, IoT and, in general, burgeoning digital business transactions, places heavy demands on the storage backend. Regardless of the high-speed storage device being used, traditional serial I/O processing will have its limitations. Even when using multi-core processors, I/O operations typically happen in a sequential fashion – one core at a time. This leads to compute workers waiting on I/O workers to process their data. This also results in CPUs being wasted as they are not put into use simultaneously. The bottleneck multiplies in effect in a virtualized environment where many apps/VMs run on the same host and wait to get their threads processed serially.
DataCore SANsymphony uses a patented parallel I/O processing technology wherein the variable number of cores in a multi-core processor are leveraged at the same time to process I/O threads in parallel. In this way, compute workers get serviced much quicker with much faster I/O processing (up to 5X more than serial processing as seen in our customer environments). This also reduces the cost and complexity of spreading the load over many servers/virtual hosts as a single multi-core server can now process multiple I/O workers simultaneously.
This results in faster application response, money saved from not having to add additional hardware to increase throughput, and higher workload consolidation ratio allowing you to run more VMs per server and improve operational efficiency. Speed up data processing significantly with parallel I/O.
Best Practice #4: Cache in RAM Where Possible
Well, we know that throwing cash at the problem is not going to be a practical solution in the long run. Let us try to throw cache at the problem and see what happens.
Using flash is great; RAM is even faster than flash. Employing RAM caching the right way in addition to flash will add to your performance gains. RAM cache is memory that acts as a buffer for disk I/O. Consider it as volatile memory that holds data temporarily for faster read and write access by the application without having to wait until the data is written to the non-volatile disk and read back. When RAM is used for caching, it increases the speed of data access by orders of magnitude in comparison to spinning disks and even flash arrays.
When using DataCore SANsymphony (present in the data path between the application and backend block storage) as L1 cache, you can leverage the DRAM available in it as a caching device and increase the speed of I/O reads and writes.
- SANsymphony uses write coalescing technology to reorder random writes that are held in cache and writes them to the disk in sequential stripes. This avoids waiting on disk for every write operation and speeds up the overall writing process.
- Read processing is accelerated by SANsymphony by pre-fetching anticipated blocks from the disk (based on earlier reads from the disk) into the read cache. This makes reads much faster.
DataCore SANsymphony supports up to 8TB of RAM per node for caching, which increases the I/O processing exponentially, thereby facilitating faster application response. Cache in on it and accelerate read and write processing.
As we saw with these four best practices, there is a lot you can do with your existing infrastructure to improve application performance. By optimizing your hardware resources and employing the right tools and techniques, you can ensure your applications run faster and users do not experience delays. In addition to these best practices, DataCore SANsymphony software-defined storage solution offers many other capabilities to further help you improve performance. Whether it is balancing loads across devices or setting Quality of Service (QoS) limits on I/O traffic to regulate throughput, you can do a lot more with SANsymphony to turbocharge your applications. Take a test drive of SANsymphony in your environment today!