Evolving Storage to Satisfy Increasing Demand for Speed, Stability, and Capacity
High-performance cloud is not cheap and cheerful. You can’t cut corners and deliver stable, performant infrastructure at the lowest possible price. Although everyone must contend with financial constraints, we must invest in up-to-date equipment to stay competitive and contend with growing volumes of data. Otherwise, we lose ground to our competitors—even in a small market.
Sure provides residential broadband, connectivity, and telephony services to the Channel Islands and the Isle of Man. Our Sure Business division offers cybersecurity, managed services, and offshore data centre, cloud, and recovery Infrastructure as a Service (IaaS) to financial services companies. We host cloud colocations in Jersey and Guernsey.
As a senior cloud services engineer, I design, implement, and maintain our customers’ infrastructure. Our customers don’t see what’s happening under the bonnet. They just consume it, and in their eyes, their infrastructure exists as a web portal. I also serve as a technical account manager (TAM) for some of these customers, including a top-tier client that experienced serious growing pains.
Alleviating SQL Server Bottlenecks
This customer, a large B2C finance company, was faced with mushrooming SQL Server databases hosted on our HPE Nimble AF60 Storage arrays. Business was booming, and their data volumes were continuously growing. They began to experience hiccups, performance suffered, and their SQL Server infrastructure slowed. As a company that sells online to consumers, their primary metric is response time, and their executive team wasn’t happy with transaction turnarounds.
SQL databases are I/O- and memory-intensive. You need fast storage, and lots of it, for everything to run smoothly. To address this customer’s capacity and speed issues, we had to consider migrating their applications to more performant storage technology. The HPE Alletra 6000 was relatively new then, but it represented the evolution of what we already had, with the same physical controls and dashboards as our HPE Nimble Storage arrays. We weren’t testing unproven technology or starting from scratch, and we didn’t have to worry about new code because we were building on our existing foundation.
HPE storage solutions have a brilliant reputation for stability, quality, and performance, but we already knew this from experience. In all our years running HPE Nimble Storage, we’d never even had a drive fail. HPE Alletra 6000 reduced latency and increased performance—once we migrated our customer to this new storage array, performance improved dramatically, and customer transaction processing response times fell from over six seconds to less than two.
One of the chief selling points of HPE Alletra is remote management. I sit 30 feet from the storage array, so it’s easier to manage it locally, but HPE Data Services Cloud Console (DSCC) makes it easy to configure and deploy infrastructure from anywhere in the world.
We don’t manage our arrays like a typical enterprise user because our business is selling storage. When a customer buys storage from us, we create a logical unit number (LUN), allocate a portion of our array as a volume, and set a performance cap based on the quantity of storage purchased. That way, performance doesn’t suffer as we add accounts, and the 10th or 100th customer benefits from the same speed as the first, no matter how much capacity is purchased or how busy the array is.
Stability, Performance, and Preventing Catastrophes
When you sell Cloud as a Service, stability is critical. If you’re running a private environment and you have an outage, one company goes down, and one senior management team is upset. When hosting multiple customers, there are dozens of senior executives looking for someone to blame if they’re offline. For some of our accounts, a 40-minute outage can cause £400,000 in losses, so we must prevent storage downtime from ever occurring.
We have two ways of preventing such catastrophic scenarios:
- Additional headroom. We purchased more capacity than we use, so our HPE Alletra 6000 isn’t using 100% of the available resources all the time. As a result, a little wrinkle is unlikely to cascade through everything and cause a significant outage.
- Our HPE Alletra 5000 testing environment. Before we deploy any firmware patches or software updates on our primary cloud, we test them on this secondary environment, which serves as a surrogate for our production environment. The HPE Alletra 5000 uses the same software and is connected to the same HPE ProLiant servers as our primary cloud, so if updates and patches work well there, we know we can safely roll them out to our customers.
To further enhance performance and ensure consistently fast response times, we installed HPE Alletra Storage's optional storage class memory accelerator modules, which serve as a cache that speeds up read and write operations. Combined with HPE Alletra’s NVMe disc architecture and 100 GB Ethernet—four times faster than the HPE Nimble Storage networking hardware—these modules push the speed limits our customers can achieve. And our customers notice. In fact, one of our biggest clients was poised to move to Azure public cloud, but we retained them by showing that we used HPE technology to offer faster infrastructure than Microsoft. That would have been a small gain for Microsoft but a tremendous loss for us.
Cutting-Edge Technology, Product Continuity, and Lower TCO
Sure has mostly standardised on HPE hardware, including HPE ProLiant servers at our Jersey and Guernsey data centres. They’re high-spec HPE ProLiant DL380 Gen10. The cluster at each location supports a multi-tenant cloud, and we use a VMware stack to provide segregated access to that pool of machines. But these aren’t the only HPE servers we use.
Last year, we installed a dedicated cluster of HPE ProLiant DL385 Gen10 Plus servers using AMD EPYC CPUs for one of our customers. These chips are highly optimised for SQL Server use cases, so we handpicked the processors and server chassis for maximum throughput. It was the most powerful equipment we could purchase without moving into specialist hardware. Moving their virtual machines from our previous servers to the AMD-powered DL385 servers improved application speeds by 35% overnight, and the customer was extremely pleased.
We used HPE InfoSight to monitor software patching and update recommendations on our HPE Nimble Storage arrays and are now using it to keep our HPE Alletra 6000s up to date. We also occasionally use the forecasting tools within HPE InfoSight to recommend configurations for our biggest enterprise clients when speed, capacity, and latency issues can hobble their workloads. We don’t need it for smaller customers who want to configure a handful of virtual machines, but it comes in handy when we’re speccing enterprise-level cloud infrastructure.
Another crucial advantage of working with HPE is value. HPE products are either best-in-class or near the top of the heap, and the company offers well-engineered, stable solutions that are well-supported and receive timely and tested security updates. Everyone looks at cost first, but the day-one cost isn’t the whole story—over a few years, the TCO for HPE solutions tends to be lower than their competitors, and they offer better long-term support. It’s reassuring that HPE will replace a component or patch a vulnerability faster than the competition and invests in engineering to produce new platforms and keep them running longer.
Engineering the Future
If there was ever a doubt that HPE would remain a key partner for Sure, HPE GreenLake for Block Storage built on HPE Alletra Storage MP put that to rest. This latest iteration of HPE storage technology uses advanced storage virtualisation and multi-protocol support to keep pace with customer expectations. Our first HPE Nimble Storage arrays were roughly a quarter the size of our HPE Alletra arrays, and HPE GreenLake for Block Storage built on HPE Alletra Storage MP further pushes the boundaries and capacity of storage technology.
One of our customers currently operates a 35 terabyte VM, which wouldn’t fit on our early HPE Nimble Storage hardware, and is already testing the limits of our HPE Alletra 6000. We’ll soon need 250 terabyte arrays to keep pace with customer demand for increased capacity and performance, and HPE GreenLake for Block Storage built on HPE Alletra Storage MP is one step ahead of our current needs.
As I mentioned earlier, the best way to stay ahead of potential problems is to have extra headroom. HPE has the vision and expertise to engineer the next generation of storage solutions to give us that extra space to manoeuvre and ensure our customers stay out in front.