How do you spell “Lightning Speed” API Manager?
January 25, 2013 Leave a comment
Well, if you are a Star Wars junkie like I am and watched the Mel Brooks spoof movie Space Balls then you probably would spell this as ludicrous speed 🙂
One of the reasons why I want to write this is that I get asked this a lot by my customers. “Why is Intel in the API space” and “Why should we buy from you” but more importantly “What are your differentiators from ——— (you insert the name here of your choice). While the first two are addressed in some of my other blogs (such as Intel Data center software strategy, etc.), I want to talk about specific aspect of our differentiator.
We all know Intel does chips very well. What you may not know is that Intel does software well too. (Refer to my blog Intel does software here). When you are combining the fastest breed of Intel processors and the software intelligence built on top with direct access to the Intel Core we are miles ahead of our competition. One such thing is outlined below.
Latest versions of our gateway products (Service Gateway, Security Gateway, API Gateway, Tokenization Gateway, and Cloud Encryption Gateway) all have one thing in common: Tied to the Intel chips at the hip and perform at highly elevated speeds. The latest such enhancement is an addition for the optimization for cryptographic acceleration using Intel processors. This enabled us to outperform ASIC cryptographic accelerator boards such as Tarrari and Cavium by a wide margin.
The concept behind the cryptographic acceleration is that when you have a heavy cryptographic operation such as SSL terminations, WS-Security operations such as encryption, decryption, sign and validation of messages the processor is pressed to perform at higher loads thereby reducing the performance and the availability for other operations. So most vendors offload the cryptographic operations to a special add on board, such as Cavium boards, to do this operation while they can continue to operate in a normal fashion. But what is not known outside is that this offloading increases the latency on crypto operations as the information needs to be sent back and forth (and reduces the throughput of message processing as well). Also, this is only possible if you are purchasing a hardware/ appliance model from the vendors. In my last search, none of the vendors offer crypto acceleration boards on their virtual versions. This essentially means that your throughput will drop highly if you use the virtual / software versions without these special crypto boards.
When my engineers told me the following results I didn’t believe them. After all, how can you beat the performance of a specialized board that is specially made to speed up crypto operations? They ran the test and showed me the live results. I am speechless so I wanted to share the results with you. There is a reason why our API Gateway is the “fastest” performing gateway in the market and the results prove that. If you are interested reach out to us for more detailed test results or for an on-site test run. We would be more than happy to prove our point. Looking at the below, essentially we are 2x-3s times faster than someone with a crypto accelerator card and probably 20-40x faster than someone with no acceleration card (most virtual machine vendors don’t have the acceleration card so they tend to stack you up with more licenses than you need to provide the capacity you might need).
|Intel CPU based gateway vs 3rd party crypto accelerator||Small messages||Medium messages||Large messages|
|Throughput increase in software||88%||82.5%||68%|
|Latency Increase with Third-Party Accelerator||65%||67%||57%|
This is because we don’t process the security / crypto operations at the application stack. Instead we send it to the processor core directly and have it perform that for us. The patent for those processing emanated from my group but it is embedded in every chip that is produced since the Sandy Bridge and Westmere architectures. Those processors expose AES-NI and SSE 4.2 instructions sets that we leverage. While this is patented yet publicly available, none of the other vendors are using these technologies today and based on the development time involved I don’t see any of them using for another 2 years at the earliest if they start today.
So in the future if someone tells you they are the fastest API solution in the world ask them to prove it and see what happens. I can’t publish the actual test results here but if you reach out to us we would be more than happy to share those results with you under NDA and if needed show the live tests for your specific messages as well.
Don’t forget to check out Intel API Manager