IVS 2000 General Meeting Proceedings |
|
|
|
Haystack Observatory Technology Development Center
Alan Whitney
MIT Haystack Observatory
Abstract:
Current work in VLBI technology at Haystack Observatory includes further development of the
Mark IV correlator system, plus a new initiative to develop a low-cost high-performance Mark V
VLBI record/playback system, that will utilize mostly commercial-off-the-shelf (COTS)
technology and most likely incorporate magnetic discs as the storage media.
Mark IV correlators have been operational at USNO, MPI, JIVE and Haystack for about a year
and have now completely replaced the Mark IIIA correlators. Haystack Observatory maintains
the software for the USNO, MPI and Haystack installations.
Before the changeover from the Mark IIIA correlators to the Mark IV correlators, extensive cross
comparison tests were done to ensure that the Mark IV results were of the highest integrity. In
fact, the first published results from the Mark IV correlators occurred less than 2 months after the
changeover from the Mark IIIA correlator.
The Mark IV correlator software has been continuously improved over the past year to include
new operational capabilities and efficiency improvements, and all Mark IV correlators are now
operating at an efficiency exceeding that of the Mark IIIA, with still more improvements to be
made. Some of the features that are now supported are:
- Processing of Mark IIIA, Mark IV and VLBA tape formats
- GUI control interface
- Simultaneous auto and cross-correlation
- Support for 1 and 2-bit samples
- Full phase-cal extraction
- Bar-code-driven tape library
- Transparent file-version control
- Full configuration specified by a set of VEX-format files
- HOPS post-correlation package
Soon to be implemented will be:
- Multiple-stream processing (multiple simultaneous independent scans)
- Faster tape synchronization
- `Double-speed' tape playback, followed by `quadruple-speed' tape playback
With support from NASA, JIVE and NRAO, Haystack Observatory is now developing the Mark
V VLBI record/playback system based on the requirements for the next generation of VLBI
systems:
- Minimum of 1 Gb/sec data rate
- Economically upgradeable/expandable to 8 Gb/sec over the coming decade
- Design based primarily on unmodified off-the-shelf subsystems and components
- Modular, easily upgradeable as better/cheaper technology becomes available
- Robust operation, low maintenance cost
- Easy transportability
- Conformance to VLBI Standard Interface (VSI) specification
- Flexibility to support electronic transfer (`e-VLBI') and computer processing of recorded
data
- Minimum of 24-hour unattended operation at 1 Gb/sec
Though this list of goals may appear idealistic, we believe they can be achieved in a system
which can be designed today, and which will evolve in the following years to even higher data
rates and additional storage capacity with minimal additional engineering effort.
Though both magnetic disc technology and magnetic tape technology have made great strides
over the past few years, the pace of magnetic disc development has been so great that it is very
likely that disc storage will become cheaper than magnetic tape storage by 2004. This trend can
be clearly seen in Figure 1, where the $/GB for both disc drives and magnetic tape media
(including projections for LTO tape media, which we focused on in the original COTS concept
proposal) are plotted as a function of time. In addition are shown two 1998 projections for disc-
storage costs, one from IBM and the other from the National Storage Industry Consortium
(NSIC); the latter suggests a tape-to-disc crossover in the latter part of this decade. However, it
is clear that the cost decreases of magnetic disc storage have dramatically accelerated over the
past two years, far exceeding industry projections of only 3 years ago, and show no signs of
ceasing. Current disc industry predictions suggest that in 2004 the cost per GB will be $.30 for
low-cost `dimestore' IDE (aka ATA) drives, but even these predictions may well prove
conservative. As you can see from Figure 1, by 2004-5 the cost of disc drive storage is
expected to fall below that of magnetic tape. Magnetic tape industry predictions have
historically been pretty much on target and are expected to continue to be so, provided magnetic
tape can even survive in the face of magnetic disc progress! Furthermore, note that the costs
shown in Figure 1 are for complete disc drive units, while the tape costs are for the magnetic
tape media, ignoring the substantial cost of tape drives.
The reason for this happening is quite clear - the flow of money into magnetic disc development
is perhaps two orders of magnitude greater than that for magnetic tape! The disc industry sees
no apparent obstacles to the current breakneck pace of development for at least several more
years. Magnetic disc area-bit-densities on modern discs now (in early 2001) are 15 Gb/, and
are expected to rise to at least 100 Gb/ by 2004. This is even far outstripping optical
recording technology, which appeared to have such a bright future just a few years ago but has
not become a serious competitor to magnetic discs for high-data-rate, high-density storage.
We believe the VLBI community should begin to prepare now for a changeover to magnetic
discs, which is the front-runner technology for the new Haystack Mark V system.
Figure 1:
Disc/Tape Price Comparisons
|
A `Mark V' VLBI data system based on magnetic disc technology is now being designed
at Haystack Observatory. Based on a standard PC platform with standard commercial
busses and interfaces, it will support a continuous data flow of at least 1 Gb/sec to an
array of up to 16 magnetic discs. The only custom hardware will be VSI-H formatter
and deformatter interfaces.
Because the Mark V system is based on a standard PC platform, the data stored to discs
will also be available for local analysis by the PC or for transmission over a network for
e-VLBI.
The cost of either the 1 Gb/sec Mark V recording or playback system (without discs) is
expected to be $25K. Multi-Gb/sec systems will be constructed simply by using Mark
V systems in parallel, though future versions of the Mark V may have standalone multi-Gb/sec capability.
The Mark V system will be explicitly based on disc drives with IDE interfaces, which are
the most cost-effective consumer discs that can be procured. Furthermore, the IDE
interface is so pervasive and so successful that any successful follow-on to magnetic
discs, such as (possibly) optical discs, will almost certainly also be available with IDE
interfaces, in which case they can be adapted to the Mark V system almost seamlessly.
It is important to note that the Mark V data system is only a data recording and playback
system and will require the use of a separate set of back-end electronics (IF converters,
BBC's, samplers, etc).
Disc industry expectations are that in 2004 a terabyte disc will be available for $300. The
recording of a continuous data rate of 1 Gb/sec for 24 hours will consume 11 TB of storage, so
that an array of only 12 such discs for a total cost of $3600 will comfortably operate unattended
at 1 Gb/sec for 24 hours. This compares to today's cost of $20K for VLBA/Mark IV media of
the same storage capacity.
Disc drives will be mounted in carriers, holding either single or multiple discs, made for multiple
insertion/removal cycles. When modern disc drives are powered down, they are quite robust to
external handling forces and can be shipped easily in padded containers. Including the carriers,
the shipping weight per disc should be 1.5 kg, so that the shipping weight of 12 discs
containing an aggregrate of 11 TB will be 18 kg.
There are a number of other obvious benefits of magnetic discs over magnetic tape:
- Rapid random access to any data
- Essentially instant synchronization on playback into a correlator
- Self contained; expensive tape drives are not needed
- Can always buy and use the latest and cheapest `dimestore' IDE disc drives, while still
using older ones
- The entire host platform, including the computer and interface cards, can be easily and
inexpensively upgraded as newer models become available
Haystack is now beginning the development of the Mark V system based on magnetic disc
technology. Early demonstrations of feasibility may occur as early as the end of 2001, with a
deployable system ready as soon as 2003.
|
|
|
IVS 2000 General Meeting Proceedings |