Monday, October 27, 2008

R5 Paoli Local Never Smelled So Good

It doesn't matter how old and dismal a SEPTA regional rail car is,
when you are riding out on the platform between cars because it's
beyond standing room only there is plenty of fresh air. This morning
the trains were late and screwd up. I heard on KYW 1060 that it was
due to dew on the rails and wet leaves. All I know is that a few
commuters starting at Daylesford were surprised (and unsafely stunned
standing close to the tracks) when the late train from Paoli raced by
and skipped our platform. This evening platform 4B at Suburban station
was jam-packed. The 6:09 Thorndale express was way behind and the 6:12
Malvern local was at least 15 minutes late. THE LAST Phillies World
Series game was coming on soon and everyone wanted to get home. It's
7:20PM and I am halfway home. I left Comcast Center at 6:04PM.

I was almost the last person on the train boarding at Suburban. We
squeezed onto the platform cab. I felt bad for about one second for
the ton of folks at 30th Street Station who didn't stand a chance. One
tenacious lady budged in - she must have been from New York.

At Overbrook, the conductor insisted on dropping the hatch to the
stairs and we had to squeeze in further and then ridevin the air.
After a couple stops the capacity reduced enough to get folks off the
platform cab. And I grabbed a seat and took this shot over my head.

This commute is really going to get old this winter. Lesson learned
though. In this economy you either get a warm seat on SEPTA or you
get cool fresh air - never both.

7:36 ...

That iPhone application may earn it's keep yet ...

Tuesday, October 21, 2008

Day in the Life Washington Metro Style

When I get a chance to tag this entry it will be not far from the truth. Right off the early Amtrak into Union Station, I tried to move swiftly through the Metro station. I was heading over to the
Marriott Metro Center to attend Splunk Live! I purchased a One Day Metro Pass with a credit card which immediately f'd me in the turnstyle - "see station manager". Apparently, the "one day pass" only works after 09:30. WTF kind of day epoch is that? Actually, I will need another tag ...

I bought a second $2 Fare Card and finally got along.

How ironic that on my way to a Splunk technology briefing, , on my way to talk with experts who understand time epochs better than most, that a mis-apportioned time epoch screws me out of
eight bucks and 10 minutes.

(if you know what a log file is and you do not know about Splunk it is time to stop using grep and download Splunk - spend some more time with your kids)

There was a day, when I literally wore a bat belt for iPlanet (i.e. two SkyTel text pagers), that I was fed up with the virtual ( ha ha ) slap in the face of operational awareness by the Java VM engineers. If the paucity of information in the GC logging was bad enough, there was no time stamp in the GC log messages. If you system traced the VM more time() system calls would fly by than you could shake a stick at and these guys weren't generous enough to drop a LONG into a printf().

I asked for a timestamp. I bitched for a time stamp. About as much time went by as there were VMs signaled SIGQUIT. One day on my way to gaining weight

"hissss...would you like any fries with that timestamp?"

I left the Java DriveThru with my


and ... Huh? WTF? These guys don't print a UTC time stamp, they don't print a formatted time value. They print an unconventional duration since the epoch of the VM/GC runtime start. I guess if you live in a virtual world the epoch of the VM start means something (look everyone! The VM ran for 87,000 seconds!). In the real world it is almost worthless. When the rest of the real world treats time as a standard you might as well live in The Black Sun if you devise
your own time epochs.

Thursday, October 02, 2008

Tape n Scissors

One gauge of our systems is to track web resources (URLs) over the
time domain measuring cache hits, origin hits, response codes,
response time, total hits and total bytes. it all comes from
telemetry. Telemetry represents change in state over change in time.
In simple terms telemetry indicates events. Logs provide telemetry
because logs record an event - the state of a component at a point in
time. A web server access log record HTTP response message
exchange events - the final state of a dispatched HTTP request. At any layer
from browser to data source, telemetry from the logs anchor the
complexity of our architecture in the simplicity of time and
uniformity of web resources identifiers.

I am working on cGauge but took some time today to try and get more
eyes on the data.

There is something more powerful than web server access logs. Try adding an event for the initial state of a dispatched HTTP request!