R/V Ewing Communications Cost Analysis
Val Schimidt
January 2003

The R/V Ewing is a global oceanographic research ship, owned by NSF and operated by the Lamont Doherty Earth Observatory of Columbia University. The ship maintains primary communications to shore via two satellite phone and data systems. These are: INMARSAT B service via a Nera Bm antenna, and Fleet-77 INMARSAT F service via a Thrane & Thrane (of Denmark) antenna. The Ewing is currently using a combination of services from the two antennas for data transfers.

At the moment, data connections are made to shore to support email and file transfers in an automated way roughly 4-6 times a day.

On the “B” the Ewing's current primary data connection type is “Low Speed Data” LSD, which has a maximum connection rate of 9600 bps connection. The interface for the B LSD is a modem interface and calls are placed to a modem bank either at Columbia, LDEO or an ISP. These data calls are charged at full “High Speed Data” HSD (64 kbps) rates.

On the “F” the primary connection type is the Mobile Packet Data Service which is a variable bandwidth connection, for which one pays “per megabit” rather than per connection duration. Also available via the “F” are analog modem calls to either the LDEO modem banks or to an ISP modem bank. These calls, in theory should support the full modem capabilities over a standard 64kbps link.

Below is a cursory analysis of the R/V Ewing's connection trends and costs in the month of September 2002 based on statistics collected by our system administrator.

A few notes on the Calculations:
· As stated, the calculations cover the month of September ONLY. Inspection of other months of data show September to be relatively representative.
· Zero's in the original data indicated failed connection attempts. These were removed prior to this analysis to ensure correct calculation of statistics.
· “Lamont96” and “acis” refer to connections to two alternate modem shore-side modem banks via the Low Speed Data link on the B. These modem banks are the LDEO campus and Columbia campus computer services respectfully.
· Data from “B (acis)” and “F (Analog Modem)” have been disregarded. They contributed to neither the total data transferred, nor the total number of connections in any significant way.
· There were a few connections during the month when files were transferred in addition to the normal email traffic resulting in abnormally high transfer sizes (2-4 Mebytes vise ~120 kbytes). Prior analysis of these large data transfers indicated a significantly higher data throughput rate than is “typical” over an MPDS connection. Average throughputs and bytes transferred have been calculated both with and without these transfers to show the effect they have on cost.
· You'll find the term “Mebibyte” and "Kibibyte" used in this discussion to indicate 2^20 bytes or 1024 kbytes and 2^10 (1024) bytes respectively. These are new terms to most, but are technically correct. The term megabyte, has traditionally been used for this amount of memory by hard drive manufacturers but is no longer correct when referring to 2^20 bytes. The prefix “Mega” is reserved to indicate 10^6 of something hence “Megabit” is 10^6 bits, as has always been the case. For more information, see
www.ldeo.columbia.edu/~vschmidt/notes/Data_Units.htm

Number of Connections Per Connection Type in September:
160 MPDS
92 B (Lamont 96)
2 B (acis)
2 F (Analog Modem)

For the ensuring calculations, data from “B (acis)” and “F (Analog Modem)” have been disregarded. They contributed significantly to neither the total data transferred nor the number of connections.

Just disregard this, it's a work around to support odd units:

First we need to define some units:

Remember that we're defining what is traditionally thought of as a Megabyte with the more technically correct term "Mebibyte. Don't get confused or bent around the axel.

Loading the data:

Manipulating the Data:

Since we are concerned only with successful connection attempts we need to remove any "zeros" in the throughput and bandwidth data files. To do this I'm extracting the data values, sorting them and then removing the "zeros" at the top. Unfortunately, behind the scenes I'm inspecting the array manually to determine the number of zero's to chuck. There is most definitely a more elegant way to accomplish this..

On a few occasions, large data files were transferred with the regular mail shipment. This can be seen in the "Bytes per Connection" plot below.

Plotting Index:

A histogram is plotted to show that these large data transfers are an infrequent occurrence. (Inspection of other months of data shows this to be true).

Prior experience has shown that larger data transfers usually result in considerably faster transfer rates than "nominal" via MPDS. In an attempt to understand how these transfer rates affect the bottom line, "nominal" data sets have been created with outlier data points removed. [Again, this was done by manual inspection rather than any fancy math - an inelegant method.], as well as a "TotalBytes_large" matrix comprosed of those removed data points.

Similarly, the throughput rate data has been sorted and the "zeros" removed, such that only successful connections are measured in our statistics. This process is repeated below for both the Low Speed Data (9600 bps) via the B dialed in to Lamont, and the MPDS connections via the F below.

Calculations :

Average number of bytes transferred per connection:

For all connections:

Number of connections of each type:

Average effective throughput:
(Effective bandwidth based on data sizes and connection times collected by the system admin.)

Total bytes transferred for the month for MPDS:

Cost of MPDS connections:

The cost per Mebibyte:

Low Speed Data via the B - Costs :

Total Time on the B is calculated using both the lower estimation of data rate and the higher estimation of data rate, however it is not clear that the increased throughput seen for large file transfers occurs for LSD links on the B. Nonetheless, the comparison is instructional:

Time on the B - using mean bytes transferred minus those for large file transfers:

Cost of that Time:

Current rate structure is provided below for COMSAT B Telephone. For these calculations a nominal rate value is chosen to provide a best guess based on typical regular usage for the Ewing.

Average Cost per Mebibyte for transfers over the B:

Summary:
Clearly, with the current data sizes and usage, MPDS is vastly more cost effective. Costs per Mebibyte via MPDS are a full 1/3 the cost of LSD via the B. Rules of thumb can be created for connection costs:

MPDS = $35 / Mebibyte
LSD = $150 / Mebibyte

Clearly the cost effectiveness of one connection method versus the other is completely dependant on the throughput obtained for LSD calls. One might calculate what effective LSD throughput is required to achieve the MPDS rate.

or

This rate is larger than is possible via the LSD 9600 bps data connection.