Friday October 27, 2006

Today at Sun we are experiencing the joy of a spam email having slipped through our defenses and made it onto a large, internal email alias. It doesn't happen very often, but when it does I'm starting to feel we really miss an opportunity to reduce our staffing in a functional area I like to call "Clueless Idiots." Those, of course, being the employees who are stupid enough to Reply-All with demands to be removed from the alias, sending yet more unwanted email to thousands of their colleagues.

At least we have not yet seen a barrage of email to the alias telling people to stop sending email to the alias. That's a sure route to a veritable email blizzard.

I suppose I should add the following. I have no knowledge of any future reductions in force at Sun and hope, in fact, that based on our recent results over the last several quarters, we will not need to take any such actions in the future. But if we do, I have some nominees! :-)

Get a clue, folks. Hit the Delete key and keep movin'.

Thursday October 26, 2006

For those following the continuing saga of my 1.83GHz Mac Book Pro, I report that it is boxed up (again) in preparation for being sent back to Apple tomorrow for repair, because two days ago it started making a very noticeable, regular, fast, ticking noise in the area of the Delete key. Web searches indicated there is probably something wrong with one of the fans, and the Apple Care person agreed.

While Apple has the unit, I am hoping they will also look at three dark spots on the display that I've been living with since I got the laptop. I finally realized these are too big to be dead pixels--they look more like dirt behind the glass. Due to a mis-communication with the Apple Care person, the display didn't get listed as a return reason, so I've included a note with the laptop, hoping the techs will look at this problem, too. If not, I may actually send it back again.

I reported last time that my battery was one that was recalled prior to the big Sony battery fiasco. While I must say this new battery doesn't seem to last nearly as long as the original, on balance I'm happier if there is less chance that my laptop will spontaneously combust.

Last month, AT&T launched an automated, network backup service for consumers called AT&T Online Vault. It transparently encrypts and backs up files from a user's home PC and incrementally transfers them to AT&T's secure, professionally run datacenter.

I've been thinking lately about the consumer desktop experience and how bad it is. And how much worse it will get as consumers continue to generate larger and larger piles of data that they actually care about with no good way to protect that data from loss. We're talking about digital photos and, increasingly, digital video--stuff people really care about, stuff they want to keep safe for a long time, and stuff they want to migrate forward as they upgrade to new machines.

Most consumers lack the skill and patience needed to back up their machines. And this is even more true now that local disk drive sizes far outstrip the size of DVDs and CDs, making the entire process all that more ungainly. If you are thinking, "shame on them for not backing up their data if they really care about it", then shame on you for your techo-arrogance. It's precisely that attitude that has lead to the crap exerience our industry offers the home computer user. But I digress.

Online Vault looks like an excellent and much needed step towards improving the home computer experience. The service costs $2 per gigabyte per month with a $17.95 per month maximum. Available for Windows 2000 and XP. No Mac version currently, but Mac users have .Mac, arguably the best overall service for safely storing desktop state in the network where it belongs.

Scott Adams, the creator of Dilbert, has a condition called Spasmodic Dysphonia. For the past 18 months he's been unable to speak, except in certain contexts. There is no known cure for this problem.

A recent post on his blog includes both a fascinating description of this condition as well as an encouraging account of something he discovered recently. Scott had a Good News Day.

Tuesday October 24, 2006

Yesterday was National Mole Day and I missed it! It is, of course, observed from 6:02am to 6:02pm on October 23rd (10/23). I should have celebrated with some guaca-mole. Made with some nice ripe...avogadros?

Hemlock Gorge is a small preserve nestled into the intersection of Route 128 and Route 9 in Newton, Massachusetts. It's a surprisingly tranquil refuge given its location and size. And, for me, it is very convenient since I pass it on my commute to Sun each day.

In addition to the gorge itself, there are three manmade structures of interest in this small area. The old silk factory, a wonderful semi-circular waterfall, and Echo Bridge. I've included photos of the latter two below.

Echo Bridge is part of the Sudbury Aqueduct, which was built in the 19th century to bring water from west of Boston to the Chestnut Hill resevoir. It is currently not in active use, but is kept in reserve as an emergency water supply. I've read that echoes from the platform at the base of the bridge are marvelous. If so, I may post a recording at some point.

Waterfall at Hemlock Gorge

Echo Bridge

Tuesday October 17, 2006

Today's Project Blackbox announcement got me thinking about big mobile datacenters. I'm thinking really big--like Suezmax big. What would be the total computational power of a container ship fully loaded with Project Blackbox units? I assume the ship would need a sizeable power plant for datacenter operations and propulsion. And that sea water could be used for heat exchange.

Estimates are that there are ships currently under development in the 15000 TEU (Twenty-foot Equivalent Units) range. Project Blackbox uses 1 TEU containers, so our Suezmax datacenter could include 15000 Project Blackbox units.

One could fit each container with a variety of payloads, but for the sake of our thought experiment, let's assume we use x64 servers and that 250 four-core servers will fit in a 1 TEU container. Further, assume conservatively that these are 2.5GHz Opteron processors. Doing the math to find the floating-point capability of our Suezmax datacenter, measured in FLOP/s (floating-point operations per second): Each core is capable of 5.0 GigaFLOP/s. Each container is therefore capable of 1000 x 5.0 GFLOP/s--let's call that 5.0 TeraFLOP/s. So our datacenter would have a peak performance rating of 15000 x 5 TFLOPs, or about 75 PetaFLOP/s. That's well over 100 times the performance of the world's current largest supercomputer. Not too shabby.

If instead we loaded each container with 250 Sun Fire CoolThreads T1000 servers, we'd have 8000 threads per container, for a total of 8000 x 15000 = 120 million simultaneous threads available in our Suezmax datacenter. Now that would be a web presence!

Of course this is not how one would build a datacenter of this size. But at smaller scales the virtues of mobile deployment, modularity, and customizability start to look more interesting for temporary datacenter expansions, for dropping IT infrastructure into zones of need, and for getting computation closer to data...

Monday October 16, 2006

Remember watching the odometer on your parents' car roll over to the next thousand or ten thousand? There was something fascinating about seeing all those nines disappear to be replaced by a string of zeroes.

According to US Census Bureau estimates, the resident population in the US will pass the 300 million mark tomorrow. You can watch those eight nines turn into a splendid row of zeroes on the Census Bureau's POPClock, here.

Thursday October 12, 2006

Today I noticed my Mac Book Pro (running 10.4.8) was doing a lot of disk activity with the system sitting idle. I could hear the disk accesses and the system was running noticeably slower when I tried using it.

I ran the Activity Monitor (in Utilities) and could see that two processes, mds and mdimport, where doing a lot of processing and their activity also seemed synchronized with the disk activity. With a little web searching, I found these processes do indexing for Spotlight, Apple's spiffy new search tool in MacOS.

I also found this site that suggests the continuous disk activity could be related to a corrupt Spotlight index. The page also describes how to turn off indexing, delete the spotlight index, and restart indexing. After following these directions and waiting for the index to be rebuilt (mouse over the spotlight icon at the top right of your screen to see whether the indexer is running), my Mac is back to being its happy and reasonably quiet self (as quiet as a Mac Book Pro can be--you may have heard the stories.)

Wednesday October 11, 2006

Early Access for Cluster Tools 7 has started and the bits are here. This is the first version of CT to include an MPI library based on Open MPI. As mentioned previously, we are in the midst of a transition from our old, proprietary implementation of MPI to an open source based model in which we contribute our resources and expertise to the Open MPI effort to leverage innovation and expertise outside of Sun. During this transition, we are willing to take some steps backwards in functionality in order to get to a better and stronger MPI position in the longer term. One such example is Open MPI's lack of robust thread safety, a feature which our proprietary library implemented very effectively. We'll get there, but it isn't the highest priority feature.

The big news is that this first EA drop includes an initial implementation of uDAPL support to enable MPI to be used with Infiniband on Solaris. This first version uses send/recv semantics, with a plan to continue improving uDAPL support over the course of the Early Access program.

Remember that Early Access is more like an alpha release than a beta release: our intent is to share the early bits with you as soon as possible to allow for early experimentation and testing. If you are a Sun field person involved with HPC or a Sun customer interested in MPI, it would be great if you would put this EA release through its paces so we can identify problems you care about as early as possible.

Congratulations to the Sun ClusterTools engineering team and to the Open MPI community for getting to this milestone. Much work has gone into improving the quality and robustness of the code base, on integrating Open MPI with Sun Grid Engine, and on adding support for Infiniband on Solaris. There is a lot more work to do, but this is a great start.

Tuesday October 10, 2006

A Solaris engineer and I got to talking last week about whether an operating system needs a wear leveling file system if it is going to run on a flash memory device. There are several such file systems for Linux that are used when that operating system is run in a small, embedded environments that use such media. But is this necessary or might it be the case that modern flash memory cards perform wear leveling internally?

For those not familiar with the term, a wear leveling file system is used on media like flash memory that has a limited service lifetime--parts of the memory can wear out if read or written too many times. Wear leveling works to spread data across the memory device to reduce usage hotspots and extend the useful life of the device.

The answer to this question--whether flash memory devices perform wear leveling internally--should be of direct interest to consumers. Think about your digital camera. If neither the camera nor the flash memory card are doing wear leveling, are you prematurely aging your memory card if you don't fill it completely before transferring your photos?

I was interested enough to write to three memory card manufacturers and ask them whether their products support wear leveling internally. I wrote as a consumer and not as a Sun employee. One answer was surprising.

Lexar and Kingston both told me their products support wear leveling. Lexar even sent me a short writeup describing several additional methods they use in their controller to increase reliability and product lifetime.

The odd man out was SanDisk. Their technical support people told me "that information is proprietary and cannot be disclosed." I asked them to check whether this was really their response and confirmed that this was indeed their answer. Personally, I'll take that as a "No" when making future memory card buying decisions.

Update as of April '07. Thanks to GlitchC for posting a pointer to a 2003 SanDisk whitepaper describing their wear leveling technology. As a further update, I'm pretty sure I know why SanDisk was being evasive last October. According to this press release, SanDisk completed its acquisition of msystems in November, about a month after my posting. Among other capabilities, SanDisk seems to have acquired a more sophisticated wear leveling scheme than that described in their 2003 paper.

Monday October 09, 2006

Hey, cool. Kai Ryssdal's interview with Scott aired tonight on NPR's Marketplace in the View from the Corner Office segment. Usually reserved for CEO interviews, they made an exception for Scott.

The segment was short, but the full interview was much longer and more interesting.

Only Scott would do a Sunray demo over the radio, but it worked! And Kai asked exactly the right question; "How come everyone on the planet's not buying this?"

Saturday October 07, 2006

My wife and I visited Watch Hill, Rhode Island, on the same day we saw Sandy Hook, New Jersey. There's a certain symmetry in that since Sandy Hook is the northernmost point of New Jersey's shore, and Watch Hill is the southernmost point of Rhode Island's.

Watch Hill, Rhode Island (facing west)

We had dinner at the Olympia Tea Room and enjoyed both the ambience and the food. If you go, do try the Avondale Swan for dessert. Yum.

Friday October 06, 2006

Pakistan has obviously been a frequent topic in the news over the last five years. But I was surprised to learn recently it isn't quite the small country I had assumed. Think about your own assumptions, and then answer this question: Where in the top 100 most populous countries does Pakistan fall? Once you have an answer, put your mouse on the graphic below. Were you surprised?

Tonight's full moon is the Harvest Moon and, with the moon almost at perigee, it is about 12% bigger than the typical full moon.

NASA has some entertaining facts about moonlight here. They mention several experiments that would definitely be fun for kids or kids at heart. Purkinje shift. Fun with your fovea, etc. I'm off to try reading a book in the moonlight...