Whitepaper: Storage Strategies For AdvancedTCA Platforms

Jan. 6, 2009
In order to reduce costs, a growing number of companies are trying to limit the number of different platforms on which they base applications. Each platform type requires a different tool chain for development; a different supply chain to manage; and diff

In order to reduce costs, a growing number of companies are trying to limit the number of different platforms on which they base applications. Each platform type requires a different tool chain for development; a different supply chain to manage; and different processes to deploy and maintain in the field.

Increasingly, telecom equipment manufacturers (TEMs) are implementing common platform strategies, where their goal is to minimize the number of platform types, yet still support the breadth of elements and applications required by the service providers. AdvancedTCA (ATCA) technology has been established as one of the key components of this approach—an industry managed open standard with the required performance, manageability, and availability characteristics required by the telecom industry. This should not be surprising, as all the major TEMs were involved in the specification of ATCA, and now most (if not all) top tier TEMs have some form of ATCA-based common platform initiative underway.

Adopting a common platform approach benefits the industry by increasing the number of applications that can be developed for that platform.

For example, initial deployments of ATCA based elements were focused on signalling and control-plane applications such as system-in-package (SiP) based softswitches and radio network controllers. Improvements to ATCA switch-blade technology have increased the application scope to include data-plane and packet-processing elements. In addition, we are now seeing a trend to push the common platform into more server-based applications such as billing engines, authentication databases, service delivery platforms, and IPTV content caching. These applications are much more diverse than other network element classes, and require a more flexible approach to data storage.

Storage can be complicated. It seems that every application has a different mass storage requirement and each development organization has different ideas on what topologies are acceptable. This means that there is no “one size fits all” storage solution. The approach adopted by storage providers for the enterprise server market is to provide a tiered storage strategy where the customer can choose the right capacity, performance, topology, and price point for any given application.

This white paper explains that although there are many different applications for systems based on ATCA server blades, it is possible to simplify them into a small number of storage application classes with similar needs, and propose a tiered storage strategy that can be employed. Finally, it gives recommendations for narrowing the choice for each storage class. First, though, it provides a quick primer for the various terms used in storage discussions and identifies their specific relevance.

Storage Primer
The purpose of any storage product is to reliably store and to provide access to such data that is required by an application in a timely and cost effective manner. This storage is normally implemented using hard drives (rotating magnetic media), although solid state drives (SSD) using flash memory technology are becoming rapidly more popular as their capacity increases and cost decreases.

Scalability can be discussed in two balancing dimensions: capacity and performance. You can scale capacity by increasing the number of drives in the storage array. Increasing the speed of the drives and/or number of CPU blades that access the storage array scales application performance. The performance of the storage array is usually governed by the type, interface, and spindle speed of the drives used as well as the amount of cache that the application can utilize.

For example, for certain applications, it may be more beneficial to fit more drives of smaller capacity than fewer drives of a larger capacity. The application characteristics determine how capacity and performance are related. Click here to read up on drive options.

Application characteristics also dictate how reliable the access and retention of data must be. A typical telecom application requires no single point of failure in the overall architecture and requires data to be accessible at all times. Of course, it would be best if no drive failures ever took place, unfortunately a hard-disk drive (HDD) is usually the most unreliable component in a system. Therefore, steps must be taken to against possible breakdown within a storage array.

This is usually achieved by means of redundant array of independent disks (RAID) technology coupled with the ability to hot-swap failed drives to get the storage array back to full operation as soon as possible. RAID can be implemented in software or hardware using dedicated RAID controller chips. In general, more sophisticated RAID implementation are best done using hardware RAID controller. Click here to read up on RAID.

Continued on page 2. ATCA Applications Requiring Storage
The challenge then is how to extend the common platform in a cost-effective way that still expands the applicability to these new areas. One answer is to better classify the applications, and here it is possible to break them down into fundamental storage application classes (i.e., on-blade storage, direct attached storage, and shared storage).

These storage application classes have a narrower range of storage capacity, connection, and performance criteria, and therefore can be used to narrow the selection of storage technologies for the ATCA common platform.

System Controller Storage Application
The purpose of system controller storage is to hold and provide boot images, host the root file system and provide storage for status and log data. The system drive also hosts configuration data that is subject to change from time to time. Capacity needs are usually quite limited, and performance is usually not critical.

A special challenge for the system controller is logging data. These records are expected to be continuously up to date in order to provide good post-mortem data after a system failure.

Database Storage Application
Typically used with a database application like Oracle, a database storage application is used to store customer and session data, and can be used to create billing engines. Capacity needs are moderate (up to 4 Terabytes), but random access read/write performance and particularly write latency are critical since the system must guarantee the security of any data written to disk.

Content and Streaming Storage Application
The content storage application is used to provide access to media files and other related user storage. Examples include video mail, audio ring tones and local cache files for IPTV. A special challenge for this application would be to predictably deliver large digital media files to customer endpoints.

Very large capacity needs are common, and read performance must be consistent to avoid stuttering and glitches of video and audio files. This application could benefit from active data management where the least-accessed data is treated differently from the most-accessed data. In the case of a streaming application, there will be a lot of parallel shared access to the storage array. ATCA Storage Topologies As previously stated, the enterprise storage industry approaches the problem by providing a tiered storage solution set, capable of scaling between very small and very large, and the same approach can be taken for ATCA based systems.

A challenge for ATCA platform, not present in the enterprise, is the environment in which they operate. The requirements for central office equipment include the requirement to run for 96 hours at 55°C without failing. This can impose additional limitations on storage options. Another specific choice to make relates to keeping all components in-shelf versus connecting to separate external (off-shelf) reliable storage arrays, such as those available from EMC. In-shelf storage keeps everything within the same reliable, managed enclosure as the blades it serves, sharing power and cooling and minimizing cable connections. However, the bladed form factor of ATCA imposes scalability limitations if reliability is to be maintained, and the small drives used lag behind the larger enterprise serial attach SCSI (SAS) and Fiber Channel drives in both performance and capacity. However, both implementations have their place in the tiered storage strategy.

There are several ways to set up the storage topology for ATCA based platforms.

On-blade Storage
On-blade storage is characterized by ATCA CPU blades with at least one drive. Blades targeting server applications with larger storage requirements can be configured with two drives and potentially an associated SAS controller. If required, additional storage can be added via AMC expansion. By using hot swappable drives, it makes it easier to perform field service and replace defective drives. It is also possible to add a storage drive directly to the blade by using a rear-transition module (RTM), a companion blade fitted into the rear of the ATCA platform.

ATCA CPU blades can use serial ATA (SATA) and SAS drives as well as SSDs. As previously stated, SAS drives may run into thermal challenges given the environmental constraints within which most telecom equipment is subjected. Direct Attached Storage (In-Shelf) Directly attached storage makes it easy to significantly increase overall data availability. By using ATCA bladed drive carriers, developers can attach additional drives (usually 2.5-in. form-factor), via a cable or by utilizing the ATCA update channel.

It is possible to attach up to four individually hot swappable 2.5-in. drives to the front blade. By using a rear transition module (RTM), two additional hot swappable drives can be attached as well. By using a JBOD (just a bunch of disks) configuration, these drives can be daisy chained to other blades as well. In the direct attached storage topology, the CPU blade would handle RAID management. It is also possible to configure the JBOD blade to enable dual CPU blade access, but no more than two blades can have access to the JBOD storage array.

Direct Attached Storage (External)
For applications that require large storage capacity, yet access from only a one or two CPU blade, a simple cost-effective external storage array can be utilized. This can be implemented using either external SAS or Fibre Channel network connections.

Shared Storage (In-shelf)
By using in-shelf, shared storage developers can provide access to common storage from multiple CPU blades using the ATCA fabric interface via iSCSI protocols. The overall storage capacity is the same as using a JBOD blade—with up to four hot swappable drives plus the capability for adding two more via RTM. It is also possible to extend the storage capacity by daisy-chaining a JBOD blade off a shared storage blade.

One advantage of using a shared storage architecture is the ability for multiple host CPU blades to access storage data without complicated, external SAS cabling. In addition, shared storage blades usually provide sophisticated RAID controller and iSCSI accelerator chips that off-load the CPU blade processing element.

Shared Storage (External)
External shared storage provides shared access to common storage data via either Fibre Channel or Ethernet/iSCSI. The advantages include the ability to support additional CPUs by adding a Fibre Channel switch. It also enables developers to use larger 3.5-in. form factor drives. By using extremely sophisticated RAID controllers and management software, external storage arrays can divide up large storage capacity and share it among many CPU blades. Using this architecture, external storage capacity is almost infinite.

Storage Application Solution Matching
If a system controller application is being developed, an on-blade storage solution using SAS, high temperature SATA or SSD is usually more than adequate. For large database storage applications, direct attached or shared storage solutions are recommended depending on the number of CPU blades that require access to the storage array. JBOD blades can be used for applications requiring single/dual CPU blade access, shared storage blades can be used for applications requiring multiple CPU blade access. For extremely large content supply applications, the best option is an external, shared storage solution that can use multiple Fibre Channel connections and/or large capacity, high RPM SAS drives.

The Future of Storage
With the increased popularity of video and digital graphics, combined with a never-ending requirement for data storage and records, the demand for storage in the enterprise and telecom markets is exploding.

Storage is one of the most exciting and dynamic spaces in the IT, telecom, and communication industries. Every year, there are increases in overall capacity; hard-drive speeds, and the number of bits that can be stored per square millimeter. Reliability is improving—MTBF numbers are increasing. As a result, both hardware and maintenance costs are dropping. In addition, with the nonstop evolution in SSDs, continually improving capacity and speed, while cutting costs, also offers new and exciting opportunities for integrating storage with the high performance and availability provided by ATCA based systems.

Summary
As ATCA platform developers expand beyond telecom and begin to target new markets such as enterprise and other server-based applications, data storage, and data management becomes more and more important. ATCA platforms offer a wide range of options for incorporating storage—from moderate, on-blade solutions to large-scale, external storage and SAN networks.

Sponsored Recommendations

Comments

To join the conversation, and become an exclusive member of Electronic Design, create an account today!