On the edge of deployment, it's time to create a new test environment for new PCIe technology.
With the upcoming release of PCIe Gen3 disk drive products using the SFF-8639 connector (due out this year), discussions at the University of New Hampshire InterOperability Laboratory (UNH-IOL) naturally turn to how these products will be tested.
The conversation focuses on whether there is a need to create a testing environment for the SFF-8639 specification beyond the current CEM specification.
When connectors, channel media, or channel lengths are changed, the electrical signal will behave differently. The more media changes you have, the more return loss will result. The energy loss will be significant after traveling from point to point, according to Josh Beaudet, the lead technical engineer for the UNH-IOL PCIe consortium.
"This weaker signal gives you a greater chance of bit errors," Beaudet said. "These bit errors can cause increases in latency, decreases in throughput, and the system is simply not going to perform as well as it could."
Early SFF-8639 implementers have been tested according to current PCI-SIG CEM tests. However, these tests are designed for the CEM connector, which is a slightly different environment from the electrical channel that will be used in many SFF-8639 implementations.
"For other high-speed serial interfaces, like SAS and SATA, we use a test fixture and test the signal right from the drive, without de-embedding the effects of the test fixture. It's not perfect, but with a high-quality test fixture, the effects are minimal," Beaudet said. "PCIe measurement relies on de-embedding the fixture to remove its effects on the signal during capturing. This can be done in PCIe, because the PCIe CEM test fixtures have been tested extensively. Their electrical characteristics are well known, and there is software for getting rid of their effects."
The PCIe CEM fixture has been heavily studied, and the community knows how it works, but that is not yet the case for SFF-8639. "We don't yet have a standard test fixture that they are going to de-embed out. So to compare apples to apples, on the SFF-8639, you need to remove the effects of the connectors and test fixtures."
With potentially longer cables and several additional connectors between the PCIe switch and the drive, the channels for PCIe signaling from SFF-8639 drives are different from the environment for which the CEM tests were originally designed. Research is under way to determine how different and reliable the CEM tests are in this new environment.
"Typically, these 2.5-inch PCIe disk drive implementations will need to route PCIe signals through an SFF-8639 connector and across a mid-plane, to another high-speed cable, then to the PCI switch," said David Woolf, senior technical staff member at the UNH-IOL. "While many of these components have been validated above 8 Gbit/s, testing them all together in system, or at least with a conformance test that mimics their behavior together, is important."
Companies are always looking to develop and deliver the highest-quality products possible; they want to do it right on the first try, eliminating issues before the products reach the market. "Demand for this testing exists, and we should see movement and development on this as more SFF-8639 products enter the market," Woolf said.