Author: David Woolf
UNH-IOL Senior Technical Staff
Since 2004, UNH-IOL and SCSI Trade Association have teamed up to offer six major Serial Attached SCSI (SAS) plugfest events, the most recent one occurring in April, 2006, at the UNH-IOL facility in Durham, New Hampshire. Plugfest events have proven to be excellent opportunities to test emerging SAS products and prove the viability, durability and reliability of this new storage I/O technology.
Large System Build
The main event at this plugfest was the Large System Build, in which individual SAS hosts were connected to a standing configuration of SAS expanders, JBOD enclosures, and tape drives. Connected data was running through the configuration while various disruptions were introduced, such as hot-swapping drives. With more and more SAS products being announced and shipping, the expectations were that this most recent plugfest would show that SAS is a robust technology ready for the real world. Defining the test program for a large system build was challenging.
The Test Plan
The SAS plugfest team spent several weeks working on a test plan for the event that defined how devices would be connected and what sort of testing would be performed. An initial challenge was how to test the many host controllers that would be present at the event. Each controller needed to be tested in exactly the same manner, as opposed to testing each controller with a different downstream configuration. It was decided that the controllers would each be given a time slot in a series of identical tests and that the downstream configuration would remain static throughout the week of testing.
The test plan for this event would be aiming higher than at past events. Instead of doing simple throughput tests over many hours, this plugfest needed to show that SAS could work in non-ideal conditions. This test plan would introduce disruptions and irregularities to the SAS fabric to prove its durability, such as staggered power-ups, drive swaps, cable disconnects, and unexpected power-downs.
The simple, single initiator standing-storage configuration allowed for fast and efficient testing. During earlier plugfests, many hours were spent constructing the initial setup of the large build and chasing down configuration issues that took up test time. This latest plugfest was vastly different. The initial setup of the large build was complete by the end of the first day and testing was ready to begin on the morning of the second day. Additionally, the test plan, which the Plugfest Technical Team had prepared, included SAS protocol analyzers stationed throughout the SAS build for “on-the-fly” debugging. This allowed test engineers to observe each SAS initiator discover each disk drive and expander in the build, and to solve problems quickly.
One of the more interesting tests performed was to create fabric disruptions during a tape backup simulation. For this test a host controller began I/O to three tape drives, one of which was at the end of the SAS chain through a SAS expander and two SAS enclosures populated with SAS and SATA disk drives (see Figure 1).
Figure 1: Diagram of Tape Backup simulation during SAS Plugfest.
A second host controller began I/O to the disk drives only. During the tape test, I/O drives were removed and reinserted (hot swapped), from the upstream enclosures. The simulated tape backup continued without interruption despite the configuration changes within the SAS fabric. This was an important step in demonstrating the robustness of SAS products, especially in the multi-vendor environment created at the plugfest.
Another important test was hot swapping of a drive during I/O to other drives within the same disk enclosure. The ability to pull out a malfunctioning drive and plug in a new drive without disrupting I/O to other parts of the system, and without power-cycling the server, is essential for IT staff. In this test, one drive was removed from an enclosure while I/O was running to some of the disks in the enclosure. Then a new drive was added in place of the old one. It was verified that the removal and adding of the new drive did not disrupt I/O, and that the server saw the removal of the old drive and addition of the new drive with nothing more than a software rescan–no power cycle necessary. The test finished by verifying that I/O could be run to the newly added drive. These simple, but essential features, must work correctly for a wide deployment of SAS (and its many applications) to be successful in the market.
The testing at this event was far more comprehensive than at previous events, when a lot of available time had been spent in building the SAS fabric. Two key observations coming out of the event were the demonstrated robustness of SAS fabrics when I/O disruptions were introduced, and the increasing support for SATA disk drives. Each of the four major SAS disk drive companies also brought SATA drives, and these were tested in the enclosures with and without active-active multiplexers.*
*NOTE: The active-active multiplexer is an interposer card, attached to a SATA drive inside the enclosure, that allows multiple hosts to simultaneously send data to a single SATA drive, a feature not available with native SATA, but made possible by the active-active multiplexer.
UNH-IOL Test Engineers were able to gather a great deal of useful data during the event, which will be used by the attending companies to debug products in their own labs, and also in driving the test plan for future SAS plugfest events. As the SAS Standard continues to be refined, plugfests will continue to be the proving ground for new products. Looking ahead, recent additions to the SAS specification regarding Zoning and management of SAS expanders are key items to be tested at future SAS plugfests. The smooth nature of this plugfest, combined with its focus on testing emerging features of SAS, demonstrates that the technology itself is healthy and poised for continued growth.