On Tuesday, December 18, 2012, Mat Wallace, Director of Cloud Architecture at colocation and IaaS provider ViaWest, dropped by the Peer Incite meeting to discuss his company’s experiences and specifically how SSD array maker SolidFire provides a unique solution to the “noisy neighbor” problem.
ViaWest was originally a colocation service provider that over the last several years has evolved to also offer IaaS services in the geographic U.S. Western states. It has major data centers in four Western states and is just opening a data center outside Las Vegas.
As a result it offers an unusual mix of services. It remains a major co-location service provider. It also provides traditional outsourcing services via the Cloud in which it runs dedicated hardware/software combinations in its data centers for individual customers. It also provides a full IaaS service based on multi-tenancy of shared hardware, software, and network infrastructure.
It uses a variety of hardware including, on the storage side, EMC DNX dedicated arrays for individual customers, NetApp and 3PAR. Its internal networking is 10 Gig Ethernet using Cisco and Arista gear with multiple short links to provide very high speed data access. Its data centers are highly virtualized using mostly VMware with some other hypervisors in specific areas mostly for individual clients.
Over the last few years a lot of effort has gone into developing co-location technology, Wallace says. But one major problem remains. “Noisy neighbors” are IaaS users that use the lion’s share of key resources, such as IOPS, degrading service for other customers. This is a problem inside data centers, when busts of traffic for one application slows reads-and-writes in the array for others and is the that traditional, unvirtualized data centers run each application on a separate, dedicated server, often with its own dedicated disk array. Even then, in many cases the arrays used by high-performance transactional applications have to be vastly over-provisioned to spread out the reads and writes over a large number of disks. And even then, QoS can degrade at moments of very high demand.
In an IaaS environment, where the economics are based on colocation, this approach is untenable. ViaWest has pretty much solved the problem of network overloads with 10 Gig Ethernet. However, it realized the impossibility of guaranteeing consistent QoS for high-performance, high-volume applications when using traditional HD arrays, even in a dedicated environment. That sent Wallace on a quest for a better answer among the emerging SSD vendors.
He talked to several, including WhipTail and Pure Storage. Their answer, he said, was to provide large numbers of IOPS and basically hope that things sorted out so that each individual customer got the performance it needed. That was not good enough.
SolidFire, he said, stood out because its tool set allows him to allocate a specific number of IOPS to specific customers or applications regardless of the demands of other customers. “With SolidFire the granularity is unique. Nothing else allows you to deliver SLAs in this way. We can guarantee that a certain number of IOPS is always available to you. That is the killer app.”
This eliminates the need for massive over-provisioning, which can actually save money in some situations over using traditional HD arrays. For instance, he said, he was talking to a ViaWest engineer working with a customer that needs a guaranteed 7,000 IOPS for a 2 Tbyte database. With SolidFire that can simply be dialed in, without the need for over-provisioning to spread reads and writes out. In that kind of situation, it may well be cheaper to use SolidFire than traditional disks both in total initial cost of the array and in power and cooling and other operational costs.
ViaWest is still testing the SolidFire solid-state array. He said that so far 2-4 milliseconds is the worst performance they have seen on long, high-demand runs. Typically, the array delivers 400-700 microseconds for 75%-90% of all reads and writes. And when the array fills up or slows down, “the way they distribute blocks across the entire array gives amazing scalability by taking advantage of increasing number of nodes.” Increasing capacity is as simple as plugging in more storage or adding a node – the system automatically starts using the increased capacity without human intervention or any interruption of operations.
ViaWest is in the process of upgrading to CloudDirector 5.1, which supports storage tiering, as part of its VMware architecture. Wallace anticipates using this capability to create a tier 1 of SolidFire with two lower tiers using traditional HDs, one for all operational data not on SolidFire and the other for archiving.
“The Cloud has many advantages. You avoid CapEx expenditures and can burst usage to support sudden demand increases or grow more slowly as your business grows, without having to guess today what you will need in two or three years. You can react swiftly to sudden changes in the business such as new business opportunities. And now with SolidFire you can have the same kind of scale-on-demand and guaranteed QoS in your Cloud storage that you have with CPUs. The big noisy-neighbor problem that was the major trade-off is eliminated.”
Latest posts by Bert Latamore (see all)
- Wikibon view: IBM legitimizing Spark to compete with Hadoop - June 19, 2015
- Red Hat flying high as big customer event opens next week - June 19, 2015
- Wikibon: Key ingredient of future storage architecture is leadership, not tiering - June 11, 2015