Intelligent QoS: Live Data Migration in the Fast Lane

We did a head-to-head comparison of data migration methods, and the competition choked. Again. In fact, choking is a step up for most of them.

You see, very few data migration/mirroring/replication tools have the ability to throttle their performance. For example, mirroring tools that are borrowed for migration typically silver the mirror at full speed. On the surface, that might seem like a good thing, except that data migration doesn’t happen in a vacuum. It shares the same host servers and the same storage bandwidth as your mission-critical applications, thereby choking the performance. To use a different analogy, it’s like opening many busy onramps to your SAN’s highway. The result is obvious: slowdowns during periods of heavy traffic and occasional gridlock that compromise your SAN’s quality of service (QoS).

Some migration tools do have throttling capability, but it’s limited. In fact, choking might be the better word. They can choke the performance of their data migration—say, limiting their application to 100 MB/s (mega-bytes per second) in a 1000 MB/s bandwidth pipeline, and leaving the other 900 MB/s for application SAN traffic. True, this method does reduce gridlock most of the time, but this is analogous to adding those onramp traffic lights that allow only one car to pass every few seconds. It works but it’s not efficient. There are times when the main highway is not busy at all but the onramp traffic lights are still in effect. Anyone who’s been stuck in a busy onramp with zero traffic on the main highway knows how frustrating that can be. Can’t they put some sensors to monitor the main highway traffic so that the onramp traffic lights are turned on only when needed?

At Cirrus Data, we took a completely different path to the problem of SAN traffic with our intelligent QoS (iQoS) approach. Because our Data Migration Server (DMS) appliance monitors the data path, we can see what the real-time traffic conditions are and pause data migrations when the host application traffic is congested. But once there is a gap that can be used, DMS automatically resumes migration at FULL SPEED, knowing that it won’t negatively impact the main traffic. This is like having the above mentioned “traffic monitors” on the main highway, thereby letting cars enter the highway at full speed whenever there is a gap, ensuring that every bit of “bandwidth” on the main highway can be fully used (filled in by the onramp traffic) without causing a slowdown.

You might think a stop-and-start method like this would take longer, but it’s just the opposite. In analyzing statistics from our real-world customer engagements, we’ve found that SAN I/O traffic spikes often. There are lots of small gaps that could be used for migration at full speed. By making use of these gaps, it is possible to maximize migration rate with little or zero impact to application traffic. When you analyze the I/O on a particular LUN, you will see that it is typically only at 20% duty-cycle at the sub-second level. That means, 80% of the time, these other tools that choke migration performance (limiting to 10% bandwidth use) are leaving nine out of ten lanes of highway completely unused. Do the math: 100 MB/s x 100% versus 1000 GB/s x 80% = 8X faster data migrations with Cirrus Data DMS.

What does this have to do with QoS? With DMS, you can efficiently use all available bandwidth without guessing. You’re getting much faster data migrations and the highest possible quality of service on your SAN, all of the time.

When you’re ready to move into the fast lane without the dumb onramp traffic light, Cirrus Data DMS is the only way to get there. Check out this white paper if you’d like to know more!