Hubbert Smith introduces the complex topic of storage performance and engineering for storage performance.
- [Instructor] So, now we're going to discuss storage performance. This is a big, interesting, fascinating topic with lots of moving parts, so here we go. Storage has two big vectors. There's bandwidth, analogous to speed limit on the roadway, and there latency, basically a fancy word for a wait, which is analogous to a stop light on a roadway. Now, for stop lights, for latencies, it's not a big impact if there's only a few cars or a few storage workloads on the wire. Latency is a big deal when you have lots and lots and lots of cars because they can backup and fail to get through the traffic light or the waiting component of the storage area network.
The other dynamic that makes this fascinating is that storage is not a steady state thing. The storage workload fluctuates over time, so here we see on the X-axis, the Time vector and on the Y-axis, the Workload, and the dotted line shows that every system has a max and the storage workload fluctuates with time. There's holiday shopping or end of quarter types of bookkeeping activities and so forth. So, this makes the workload on the storage system also administrative tasks.
Backups create these workload peaks or images, or there's a number of other things that can cause these peaks. The thing to be avoided is when you exceed the system maximum and you create too much demand which results in a system slow-down, which impacts your business. A system slow-down can be caused by not enough sequential wire speed, the sequential performance which is measured in megabytes per second. Alternatively, the slow-down can be caused by not enough IOPs, not enough operations per second getting through the wire.
And in order to dig into that and understand root cause and corrective action and do the right level of engineering, it's important to understand that it's not just about the storage array, it's not just about the SSDs or the hard drives, it is truly an end-to-end performance system where the weak link can cause the whole thing to slow down and in this example, we see there's PC Clients that create a workload, there's a Network Switch, there is a Server DAS in this example. In another example we'll show a storage array.
The Server DAS has finite resources, it has a finite throughput on the network. The CPU is finite, the RAM is finite, and then the performance off of the RAID group, the hard drives or the SSDs, is finite. These workloads tend to compete for the same resources and part of the job of a storage administrator is to organize your resources and virtualize your resources such that one noisy neighbor doesn't compete with and interfere with the resources of a high-demand or a high-value, time-sensitive workload.
We're going to cover some SNIA Exam tips related to end-to-end performance. This is the storage array example that I mentioned a moment ago. So, here we see an application server and it obviously has some sort of a database or some sort of an application in an operating system. It has wires, in this case Fibre Channel would connect the server to the switch, connect the switch to the array. Now, storage array, in this case, I'm suggesting NetApp and EMC, or EMC, or there are others. The storage array tends to level out the performance peaks and valleys with a thing called caching, but caching can mask slow storage, particularly hard drives.
Another thing to gate end-to-end performance is wire speed and competing traffic. Fortunately we have Fibre Channel zoning to allow you to parse off high-value traffic onto their own dedicated networking and keep that away from lower-value traffic that could interfere. Important to note, there are Fibre Channel wire speeds. Older generations of Fibre Channel are one gigabit, two gigabit, four gigabit per second, these have been around for a while. This is what you're probably going to encounter. Newer generations of Fibre Channel include eight gigabit per second and 16 gigabit per second.
- Storage administration as a career path
- Fitting storage building blocks together
- Storage today and how we got here
- Consolidation and service level agreements
- Implementing backup and recovery SLA
- Capacity planning
- Spinning drives
- Hard drive interfaces
- Fibre Channel
- Solid state drives