Performance Acceleration Module (PAM)

Along with the introduction of the new FAS3100 series arrays  a couple of months ago, NetApp also made an announcement regarding a Performance Acceleration Module (PAM).

Since i haven't  seen any of the NetApp bloggers discussing it, although I may have missed it, I figured you'd probably find it more interesting than hearing a story about how every time I go on vacation I lose my luggage.

So what is a PAM?

The PAM offers a new way to optimize the performance of a NetApp storage system by improving Throughput and Latency while reducing the number of disk  spindles/shelves required as well as power, cooling and rack space requirements .

 

It is a an array controller resident, intelligent 3/4 length PCIe card with 16GB of DDR2 SDRAM that is used as a read cache and is integrated with DataONTAP via FlexScale which is software that provides various tuning options and modes of operation.

Why the PAM?

In every disk array, the processors can access data from two areas: from disk or the Cache. If the data is not in the Cache, this results in a Cache miss which means the data will have to be fetched from disk. Disks have been and continue to be the slowest part and the weakest link in a disk array. The time (latency) to fetch data directly from the disk is orders of magnitude greater than from cache. For read  intensive random I/O applications that are latency sensitive this requires the configuration of a high number disks in order to provide user and application acceptable latencies. Most Microsoft applications fall in that category as well as Server virtualization platforms and File services which tend to generate, primarily, these types of workloads.

In fact, for these types of applications, conventional wisdom says, you size for performance rather than capacity, which means additional spindles and shelves. However, while you'll get the IOPS and latency you're looking for, you are doing so at the expense of wasted, unutilized capacity, additional rack space and power and cooling costs.  

So the idea with deploying one or more PAMs is that a hit to the PAM's cache will considerably reduce the time it takes to fetch the data as compared to the same process occurring from disk. The additional benefit here is that it will also allow for a more reasonable amount of disks and shelves to be deployed to support the same application thus reducing costs.

As far as the power and cooling aspect of it, a single PAM card (16GB) requires at most 18W which is about 95% less than an FC disk shelf with 15K RPM drives. That equates to approximately 3000kWh/year of savings per shelf not deployed and 3U of rack space. With that said, I do not mean to mislead anyone by insinuating that for each PAM you add you'll can remove a disk shelf.

The PAM's performance contribution is really workload dependent. It could be that a single PAM can eliminate one, two, three or more disk shelves or zero. Again, its effect is workload dependent and that's why we have developed the necessary analysis tools and sizers to be able to determine its effect for each situation.

How many PAMs can I install?

It depends on the Storage system type. For the some of the midsize systems the minimum is 2 PAM modules (32GB) and scales up from there to 10 modules for the FAS6070/6080 series. 

I understand that Solid state Disks (SSD) offer another alternative to adding disk drives as a means of optimizing performance. Another similarity between SSDs and the Performance Acceleration Module (PAM) is that both contribute to storage system performance with workloads that are random intensive.

However, there distinctions to consider, and while there's a there is a bright future for SSD in enterprise storage arrays, cost has been and continues to be a large hurdle to overcome, to the point that SSDs are currently deployed primarily for niche applications.

In contrast, the NetApp Performance Acceleration Module (PAM) is available with mid-range storage systems and positioned for mainstream workloads such as the ones produced by most Microsoft applications such as Exchange and server Virtualization platforms like VMware.  

Comments

[Chad sakac]

Great post as always Nick - one thing I did want to point out - we now offer SSDs in midrange (CX4 and shortly the Celerra) and high-end (DMX) platforms. We're also starting to see a big knee in costs as the volume is ramping, and other entrants (Intel) are getting into the mass market.

Not trying to argue the PAM vs. SSD angle - the PAM is a perfectly valid (but different) approach.

[Patrick M. Slattery]

Is the RAM upgradable on the PAM board? Would it be possible to put 32GB or 64GB of RAM on a board?

[Nick Triantos]

Hi Patrick,

The PAM module comes in 16GB per card right now. The maximum than can be deployed per supported array is approximately 10 times the array's RAM.

[Nick Triantos]

I need to get my math right. It's not 10x. It's actually 2.5-4x depending on the platform.

[korman]

Will a PAM Card help with Netapp's dirty little deswizzling secret?

[korman]

Also... What are the implications on the over all performance of my servers after netapp cluster takeover?