H.K. Jerry Chu's Weblog

Monday May 09, 2005

Hello world!

Hello, world! This is Jerry Chu from the Solaris Kernel Technologies group at Sun. Since this is my first welog, a brief introduction of myself is in order.

First off, my official name is Hsiao-keng, a direct (phonetic) translation from my Chinese name. But most people know me by Jerry.

Next, a bit of my past work at Sun. I have also done work outside of Sun. But that will be the topic for some other time.

I started the OS kernel work from SunOS 4, a BSD based UNIX system that was used in early years at Sun. Yes, it has been a long, fun, and sometimes bumpy ride. For the past ten years, my work has mainly been in the networking area, first as the main TCP caretaker for Solaris 2, followed by a short sting in the clustering networking area. In recent years I have turned my passion to the performance of the Solaris networking stack. As a senior staff engineer of the Solaris Network Performance Team, we have cranked out a lot of good work in Solaris 10, from the famous FireEngine project, to a number of less known, but equally important projects including packet chaining (a.k.a. MDT), various hardware offload technologies (all flavors of checksum offloads, Large Segment Offload, packet classification), in-kernel HTTP caching (a.k.a. NL7C), copy avoidance..., etc.

In my spare time I also participate in IETF, first in the TCP Implementors' Working Group, and now as a co-chair of the IP over InfiniBand Working Group.

I will leave a lot of fun details for later. Suffice to say that the work in the system and networking area at Sun has been cutting edge and challenging. The system and networking industry continues to evolve from the advance of both the software and the hardware technologies. Today a critical design decision lies on finding the proper abstraction, at the right level in the stack, from both the cost and performance perspectives, to draw the boundary between the software and the hardware, and to realize the Network is the Computer (TM) paradigm. This requires an accurate analysis of the cost and the state of both the software and the hardware technologies at hand. The End-to-End Principle from more than two decades ago, to focus the lower layers of the stack on performance, while leaving correctness to the higher layers towards the applications still applies today.

More later,

Jerry May 09 2005, 07:26:59 PM PDT Permalink