FibreSpy is a weird term. Everyone knows that there are no spies in storage. The industry is made up of only good people and good products, right? OK, maybe that's going too far. But rest assured that FibreSpy is a good technology for
Before we jump into the details, however, let's talk about the state-of-the-art in modular storage.
arrays have always been divided into two groups: modular storage boxes (EMC Clariion, HDS Thunder, IBM FastT family are examples); and monolithic storage boxes (EMC DMX, HDS Lightening, IBM Shark). Each comes with different feature sets, software and scalability. Often, they use different management software. You couldn't buy an entry-level storage box at a competitive price and simply add to it to increase performance or capacity beyond a certain limit. Lately, however, we have seen some signs of breakthroughs. Emulex's InSpeed and its latest offering, FibreSpy, are examples of this.
InSpeed is a
(SOC) that allows a designer to connect 14 to 20 FC drives in such a way that, while they continue to communicate with each other and the controller via FC-AL protocol, they obtain full 1, 2 or 4 Gbit switching bandwidth per port and better reliability.
Practically every storage array vendor I know has incorporated (or is incorporating) this technology within its boxes. Here's why: In a traditional RAID array (modular or monolithic), the back end is made up of drive trays of up to 16 FC drives implemented in a JBOD fashion. That means the drives are all connected in a FC loop and all drives speak FC-AL protocol. This presents little problem when the configurations are small, but once you have several trays connected on the back end, performance goes to hell in a hand-basket.
The problem lies with the way
Fibre Channel Arbitrated Loop
(FC-AL) works. Being an arbitrated loop protocol, it doesn't allow the nth drive to read or write until it is given its turn to speak. So while the drives are ready, able and willing, data has to wait. This is OK if there are three or four drives, but becomes a major issue if there are, say, 100 drives on the loop (max allowed by FC-AL is 126 devices). Now you know why no one puts more than, say, 30 or 40 drives on a loop, even if they are allowed to per the configurability rules.
The other major issue with standard FC loops is the inability to isolate a drive when it is malfunctioning. In many cases, the vendor would isolate the whole tray if there were a single malfunctioning drive. Can you imagine isolating 2.3 terabytes (TB) simply because one of the 16 drives in a tray with 146 GB drives was acting up? Yet this has been the standard practice until recently.
InSpeed SOC basically converts a JBOD into a Switched Bunch of Disks (SBOD). In other words, the FC disks are each connected to a switch port and the SBOD trays are in turn connected to other SBOD trays, either directly or via external InSpeed chips. The drives essentially gain in two dimensions: performance and reliability. They gain in performance because each drive is now available for immediate communication on full FC speed per port. They gain in reliability due to additional monitoring and diagnostics afforded by the chip.
There are two major implications. One, you can add a much larger number of drives on the backend, possibly all the way up to 126, the max allowed by the FC-AL spec, before running into performance issues. Alternatively, you can enjoy much better performance with the drives you do have. Two, the reliability improvements means you only isolate the drives that are malfunctioning and the management software knows more precisely what is wrong with the drive.
Emulex, via the Vixel acquisition, has recently introduced a complementary technology to InSpeed, called FibreSpy. While InSpeed allowed the designers to reach the 126 drive maximums per FC loop, the FibreSpy chip allows the designer to build powerful systems with thousands of FC drives on the back end, while maintaining the performance integrity and maintainability. In a subtle way, it also enables the creation of tiers of storage while staying within the FC standard. In addition, it enables the creation of simpler, cost-effective FC-based SAN solutions for the small and medium-sized business market.
FibreSpy is a five-port
application-specific integrated circuit
that boasts four FC-AL ports and a SPI-4.2 port. When attached to the back end of a RAID controller, one FC-AL port is connected to the RAID controller side and the other three ports can each attach up to 125 FC drives of the same speed. The chip itself is capable of negotiating 1, 2 or 4 Gbit speeds but only at a port level. The back end can be expanded by connecting a second (or third or fourth or fifth) FibreSpy chip via the SPI-4.2 bus. A five-chip chain would have 19 ports available for the drive side, allowing up to 2,375 drives, or over 700 TB of capacity, using 300 GB drives. All the niftiness of the InSpeed chip and more is built into this FibreSpy chip for reliability and diagnosability. Tiers of storage can be built using, say, 1 GB drives on the first port, 2 GB drives on the second, 4 GB drives on the third. Rotational speeds of drives can be mixed in any way one wants. If you wanted to have a tier equivalent to SATA, you would attach FATA drives from Seagate (these are SATA drives with a FC-AL front-end). Don't forget each tray of up to 16 drives can be SBODs, in order to maintain the balance between capacity and performance.
In another incarnation, the FibreSpy chip can be used in front of a RAID controller, allowing what would otherwise be a single FC port to host three FC-AL ports. If the RAID controller was designed with four FC ports, you can now connect up to 12 hosts. Since the ports speak FC-AL and not fabric, the configuration is a lot easier to implement and manage. Since the chip would be implemented internally, a 12-host configuration would require no external FC switch. And while the per-port pricing on FC switches has been dropping dramatically over the past year, it would be hard to compete with an integrated FibreSpy chip. I would expect these low-cost FC solutions to compete effectively with iSCSI-based solutions coming to the market.
In my view, modular storage will grow dramatically over the next three years, partially due to technologies like InSpeed and FibreSpy.
About the columnist: Arun Taneja is the founder, president and consulting analyst of the Taneja Group, an analyst and consulting group focused on storage and storage-centric server technologies.