Fast Cache and the Power Path

What’s the connection between server Flash and storage Flash?

Recently, there’s been a lot of talk about server Flash here at EMC.

The EMC implementation of PCIe server cache or EMC VFCache makes for a very good application performance story. It augments EMC storage-based Flash (i.e. EMC FAST) for optimal data storage just nicely.

These technologies will most assuredly help you deal with the challenges that come with the significant server virtualization and storage growth projected.

But, focusing on server and storage optimization alone is like reading only the beginning and the end of a good story and missing all the excitement that happens in between. It’s the total package that makes for the most rewarding experience. You need a comprehensive approach to optimizing server, storage—and network.

Fortunately, there is a way to optimize the connections between your servers and storage to get the most out of these new technologies.

Hyper-Consolidation – Too Much of a Good Thing

Increasing virtual machine consolidation or hyper-consolidation results in I/O bottlenecks.

Sure you’re able to get more out of existing server resources, but often at the risk and expense of application availability and performance. Adding the performance boosts made possible with Flash technologies may solve some of the problem, but not all of it.

Performance Boosts – More than a Flash in the Pan

Storage-based Flash solutions like EMC Fully Automated Storage Tiering (FAST) bring resource optimization to storage, providing the ability to assign critical workloads to more expensive, powerful Flash drives and less important processing to less expensive, lower performing SATA drives in a single array.

Additionally, FAST for Virtual Pools (FAST VP) promotes only the active portions of a volume/LUN to Flash drives. This gives you the best utilization from limited Flash drive capacity.

Built-in intelligence moves inactive portions of a volume/LUN to SATA drives, so you can use both Flash and SATA together in Tier 1 environments.  You can also use this Flash as cache to buffer reads instead of going back to spinning spindles every time.

Benefit: good performance and scale at the storage level.

Server-based Flash solutions like the new EMC VFCache extend the benefits of storage-based Flash tiering to the server; basically, it off-loads some of the processing from storage and moves it to the server.

These benefits include better performance (especially for read-intensive applications like Oracle), tuning intelligence (now not just for storage but the server too), and protection (application-writes mirrored to the back-end storage leverage the high availability and disaster recovery capabilities inherent in the array).

VFCache automatically adapts to changing workloads by determining which data is most frequently accessed and placing it within its cache; more highly read data stays local while least-used stays back on disk.

The result is better performance (i.e. up to a 300 percent improvement in input/output operations per second (IOPS)), and lower latency (i.e. as much as 50% reduction in latency). Deploy VFCache on multiple servers and you can scale the gains across your data center.

Benefit: good performance and scale at the server level.

Great.  You’ve now got powerful performance engines at both ends of your network.

But, what gives you the performance boost in the middle?

Too Many Tools – Too Little Time

Historically, for the most part, managing the connections or data paths between servers and storage was left to the native multipathing technologies bundled in with and specific to operating systems.

Though some offered some flexibilities, more often than not these native multipathing technologies were implemented using default settings whether or not the most beneficial to the data center.

There was a time before virtualization though when underutilized server, network, and storage capacity could take up the overall performance slack and many times nobody noticed the less than efficient handling of the data paths.

Who’s got the time?

Correcting path failures and poor configurations interrupts operations and uses up valuable staff time. At the root of the problem is this multilayered, operating system-specific approach to managing data paths.

Most native multipathing I/O (MPIO) solutions in operating systems use a round robin approach which does not take into account unstable or less than viable data paths for the traffic. Add to this the lack of a complete view of data paths across the virtual and physical environment and the net result is less than optimal data path utilization and not so great performance.

Automate and Standardize the Connections

Much like virtualization and server-based Flash optimize servers, and technologies like FAST optimize storage, you need a data path management solution that optimizes your data paths.

This data path management technology should provide:

  • A single tool set across physical and virtual
  • End-to-end I/O visibility for the entire data center
  • Automated path management, failover/recovery, and load balancing
  • Improved application performance and availability
  • Lower costs and the ability to scale

Possibly the only technology to provide all of these capabilities today is EMC PowerPath.

A server-based solution (just like VFCache), PowerPath is intelligent data path management for the entire shared SAN.

It combines multipath I/O capabilities and automated failover/recovery and load-balancing with support for heterogeneous operating systems and storage. These capabilities provide the ability to address the impact of not only virtualization but also the day-to-day challenges of the physical environment, as well.

Patented algorithms optimize connections to EMC storage, as well as many non-EMC arrays.

In virtual environments, PowerPath constantly checks and rebalances workloads across all available paths to ensure optimal performance to support shifting virtual machines.

In physical environments, it not only monitors and adjusts to workloads but gives you the ability to segment bandwidth per application giving priority to more important apps; you can even dedicate data paths to certain priority applications.

The benefit is a performance boost of up to 3x. This performance benefit was covered in a recent post here. The results cited were for Windows environments, but I can tell you the results for other operating systems are similar including VMware vSphere. You’ll be hearing more about these results in the not too distant future.

Benefit: good performance and scale for data paths.

The recent VFCache announcement did not make reference to any multipathing solution and is agnostic to the underlying data path management technology. While all well and good from the server perspective, keep in mind that a successful virtual data center is predicated on a comprehensive server, network, and storage strategy.

You not only want to optimize your servers with VMware and EMC VFCache and your storage with EMC FAST, but you should also optimize your data paths with EMC PowerPath.

It all makes for a great performance story from server to network to LUN!

About the Author: Mark Prahl