Storage Architecture Options in ATCA Designs

The latest advancements in Internet Small Computer System Interface (iSCSI) hold some exciting advantages over the Direct Attached Storage (DAS), Network-attached Storage (NAS), and Storage Area Networks (SAN) in the large datacenter environments. iSCSI is designed to transport large blocks of I/O over an IP network instead of files from a file system. While it’s clear that Ethernet has won the ATCA backplane bus war, it appears that a new battle is emerging over the best storage architecture for ATCA.

ATCA is ideally suited for IP Multi-media Subsystem (IMS) applications and as such, provides an ideal front to explore the storage architecture battle. The IMS network elements, such as media gateways and the media resource function, must access large amounts of storage to evaluate subscriber profiles, feed media streams to client devices, or process audio streams for voice analysis and recognition. iSCSI provides a very cost effective and scalable interface between the network elements and media / data. A dedicated external storage solution such as a Fiber Channel (FC) SAN can also provide high-speed access to media with arguably the best reliability and security.

As we explore iSCSI solutions in the ATCA architecture, some key concerns surface around security, reliability, and throughput. By embedding iSCSI software into hardware devices such as an ASIC on a mezzanine card, the concerns begin to disappear and what emerges is a very cost-effective and scalable storage solution. One company that is taking an early lead in this approach is Astute Networks and their new Caspian 10Gb iSCSI ATCA storage blade.

The Caspian blade provides a Dual-Redundant 10Gb iSCSI (ATCA Option 9) using an Athens 10Gb iSCSI ASIC and a Single Fabric with no cabling or storage fabric to manage. To get more insight regarding this design approach, I spoke with Fazil Osman, Astute Network CTO and presented some questions about the new Caspian blade.

Q: What makes your iSCSI ATCA blade better than what has been attempted by other manufacturers?

A: The requirement for an iSCSI ATCA blade is for it to be shared by all the CPUs in a chassis. This means that the performance has to be in the same range as an external array. We have seen two companies attempt to make an iSCSI blade. One manufacturer had a low performance 1G blade that was not good for sharing and another added a 10G connection but did not solve the performance problem. They use a very low-end processor to do the iSCSI termination. In Caspian, we use our own 10G iSCSI termination engine called Athens that can do iSCSI termination at > 8Gbps. This allows us to support a bulk throughput of 300-400 MBps. Because of the low power dissipation of Athens, we can put a RAID 5 engine where the others only support mirroring.

Q: Have you obtained any benchmark data on throughput per processor?

A: Most CPUs we have tested against have 1G support and they can support the full bandwidth of that 1G link. Most customers have told us that they are not looking for 10G bandwidth from a single CPU. Our feeling is that you will need an iSCSI accelerator to support that bandwidth from a single CPU.

Q: Will iSCSI meet the reliability demands of ATCA in the Telecom space?

A: iSCSI is as reliable as FC as a protocol. The limitation of iSCSI has been performance on the target side, and iSCSI accelerators such as Athens solve that problem. Our Caspian board is dual-homed to support any failures. Additionally, iSCSI utilizes the ATCA backplane to connect servers with the storage array, eliminating the reliability and logistical issues associated with using the FC cables that are required with external arrays. The end result is a solution with equivalent protocol reliability and superior physical reliability than FC.

As an example, the IMS application ATCA platform illustrated in Figure 1 shows each iSCSI RAID blade connected to both switches. The data is mirrored to the blade at the application level and each blade runs RAID 5 (750GB of protected storage with 150GB drives and 2-drive RTM on each). This provides two levels of data protection in the system (effectively RAID 50).

Final Score

The latest approach by Astute Networks is “on target” as an iSCSI architecture. The inherent fear of latency and security risks are clearly mitigated with this embedded design. The remaining question will be the speed of market acceptance and pricing to compel designers to move away from the more traditional storage architectures in ATCA. IT

Jeff Hudgins is VP of Engineering at Alliance Systems. For more information, visit the company online at http://www.alliancesystems.com.

» Internet Telephony Magazine Table of Contents