What's the Best Choice?
by Michael J. Cannavo
With an alphabet soup of available storage and archiving options, facilities that know
the pros and cons of each have the most to gain.
 Very few components in PACS have changed as dramatically in the
past few years as archives. Once the domain of very large, very expensive jukeboxes
holding multiple high-cost platters that took anywhere from 2 to 4 minutes to retrieve a
single prior study, current PACS archiving strategies almost universally employ the use of
storage area networks (SANs) that use compact, high-speed, rack-mounted disks controlled
by a single fibre channel switch that can store and retrieve data in near real time. The
time it takes to prefetch a prior study has been reduced between 20- and 100-fold, with
the cost of storing the same data reduced almost equally as much.
Numerous articles have been written about the migration from direct attached storage
(DAS) and network attached storage (NAS) to SAN. Currently, acquisition, deployment,
management, and operational costs all seem to favor SAN (and NAS to a lesser degree), with
much lower total costs of ownership, even though DAS has a slightly lower initial
acquisition cost. Unfortunately, this cost savings with DAS often is negated due to the
need to replicate data on the disks. In addition, the continual downward price spiral
associated with SANs-as data is packed more densely onto disks and the yield of these
rather small disks increases, not to mention better data transfer performance-makes
it a medium of choice in archiving for the foreseeable future.
Breece Hill's iStoRA 4000 (top), Data Distributing's Ciprico DiMeda 3600
(middle), and Array Corp USA's 2905 Laser Film Digitizer (bottom). |
The discussions about DAS versus NAS versus SAN application can take up volumes;
however, here are the basics.
A load balancer is at the front end of each application (DAS, NAS, and SAN). With DAS,
there is no coordination or data sharing between the servers, and files are replicated to
maintain consistency. Servers in a NAS configuration share some data, but there is no
coordination among servers. A SAN allows all servers to share the same data and to
coordinate access to the same data pool.
DAS requires no interface between the servers and the disks-hence, the meaning of
direct attached storage. NAS uses single or multiple filers as the interfaces between the
servers and the disk farms, with an Ethernet connection between the server and filers.
SANs have a small single-fibre channel switch as the only point of failure between the
server and disks.
Pros and Cons
Naturally, people question the different types of networked storage specifically
relating to NAS versus SAN. Basically, those who need file I/O (input/output) have
typically chosen NAS; those needing block I/O have typically chosen SAN. What are the
differences? In a nutshell, file I/O is what is used to talk to files, and block I/O is
what is used to talk to disks.
Usually, file I/O uses either a network file system (NFS), which is a protocol
developed by Sun Microsystems (Santa Clara, Calif) that allows a computer to access files
from a network as if they were on its local disks, or other access protocols running over
transmission control protocol/Internet protocol (TCP/IP)-based Ethernet network. File I/O
also has locking mechanics, so data may be shared between different operating systems.
File I/O usually uses TCP/IP as the transport mechanism and, thus, must conform to the
seven-layer "stack." In other words, each piece of data needing transport must
traverse the entire IP stack before going "over the wire," which causes overhead
and slows things down. This latency can be minimized by a device that off-loads the CPU
cycles needed to transmit the data through the IP stack from the server CPU to the network
interface card itself, but allows for another point of failure in the system.
Block I/O is the basic mechanism for disk access using the SCSI protocol as the command
set. Block I/O is fast, and data can be transmitted in various block sizes from 2K upward.
Block I/O also can be done over various transports, including SCSI cables, copper fiber,
optical fiber, and even encapsulated and then transmitted over IP networks. These reasons
are why it is used in SANs.
Rorke Data's Silverline CWDM (top), and StorageTek's BladeStore (bottom) |
Advantages of file I/O are ease of implementation and the ability to "share"
files, but the disadvantages relating to speed and latency-and the fact that many
applications cannot be "installed" on a network share-minimize its use.
Advantages of block I/O are speed, minimal latency, and high availability, all areas that
are key in a PACS deployment.
Network latency (the delay between sending and receiving data that increases as the
amount of traffic on the network increases) and throughput (the total rate at which data
can be sent across the network) typically renders NAS too slow for relational database
management systems (RDBMS), which are the systems used most by PACS providers. NAS also
has developed a slightly broader set of applications in recent years that have allowed it
to migrate to a more general-purpose networked solution. That said, most PACS vendors
still have migrated to a SAN solution, favoring cost and simplicity over expense and
complexity.
Both DAS and NAS have fairly significant drawbacks. DAS requires software to
synchronize the data pools, has difficulty in maintaining coherence between targets that
are unavailable or change frequently (as with PACS), and have high administrative and
storage costs. NAS requires more storage, as each node has its own copy of the same data,
and there is no guaranteed data integrity-something that is critical in a healthcare
environment. Performance and throughput with NAS also are lower than SAN, and network I/O
can be expensive. Finally, the use of NAS filers allows for multiple points of failure as
well as bottlenecks and management challenges, as the load and data must be redistributed
and partitioned across the multiple NAS filers.
SAN avoids many of these barriers, allowing data sharing around the servers,
consolidation around a single back-end network, and centralized management of storage
resources. Most importantly, SANs also offer guaranteed data integrity, high availability
with no single point of failure, better throughput and performance than NAS, and, as a
bonus, lower costs per GB versus DAS.
Talkin' Ca¢he
It is extremely hard to find a PACS that still uses technologies like WORM (write
once, read many), MOD (magneto-optical disk), or similar technologies. This is not to say
that other technologies aren't being used for archival in radiology, as many CTs and
MRIs still archive on low-capacity (1 GB–2 GB per platter), 5.25-inch MODs or WORMs.
Still, at least 75%–80% of all PACS archives today are SAN based.
Advances with DVD technology have allowed disks to increase to 9.4 GB per disk at a
current cost of less than $10 per disk. With 80-platter jukeboxes priced under $20,000,
DVDs have come into play as the medium of choice in low to moderate volume storage
applications. Also, DVDs are extensively used for online backup (eg, disaster recovery) by
almost all PACS vendors.
From top to bottom are DeJarnette Research's PACSware Migration Gateway 2.0,
RADinfo System's PowerArchive, EMC's CLARiiON, and Kodak's VIParchive |
The cost effectiveness of DVD is far beyond that of MOD or WORM, which both store less
data at nearly 10 times the cost. From a speed standpoint, DVD competes exceptionally well
with older medium, allowing access to images in about 30 seconds (versus 2–3 minutes
or more by MOD and WORM technology). DVD is used in about 10%–15% of all PACS today,
primarily in small to midsize applications, like a 250-bed hospital or large-scale imaging
center. Other facilities employ various tape strategies, including linear tape open (LTO),
advanced intelligent tape (AIT), and digital linear tape (DLT).
The cost for 1 terabyte (TB) of archival storage on a complete SAN solution (load
balancer, server, fibre channel switch, and disk farm) has dropped below $20,000, with
additional TBs costing less than $8,000 each-just one more reason for the widespread
acceptance of SAN.
Although SAN prices are cheap when purchased directly from a variety of vendors in the
medical marketplace (eg, EMC, Hewlett-Packard, IBM, Dell, and Toshiba), it does not mean
that SANs themselves will be cheap when purchased in conjunction with PACS from the major
vendors. The same can be said with DVDs.
A 5 TB SAN with an LTO tape backup can cost $100,000–$300,000 from a PACS vendor,
with an 80-slot DVD-RAM jukebox alone easily exceeding $75,000. Why the huge price
disparity from open-market pricing to PACS-vendor pricing? There are a couple of reasons.
First, a SAN is not necessarily a SAN. The system design (from both a redundancy and
availability standpoint), back-end server architecture, expansion capability, throughput
(both inbound and outbound hits), connectivity (fibre channel or SCSI), and other items
all factor into the pricing of a vendor's SAN. Some vendors also build the costs of
the archive management software in the PACS server, and others fix the costs in a
dedicated archive manager. Still others design a system with a dedicated server for the
tape backup as well. Also adding to the cost of the SAN are the software to run it, the
database management, the archive manager, and other components inherent in a PACS design
that ensure data integrity. This area also is one that provides the vendor with margins.
Even a single DAT or DLT backup server will be priced 5–10 times that of its cost on
the open market, if not more.
To that end, many vendors are now allowing facilities to purchase their own SAN and are
just charging a software application and systems integration fee to allow the facility to
integrate its SAN into the PACS design. This situation is becoming the standard in PACS
implementations, allowing the facility to distribute the SAN costs over multiple
departments instead of a direct charge back to radiology-as one would find with a
dedicated PACS archive. This purchase plan also allows the facility's IT department
to maintain the SAN and control the storage medium that will address the data from
multiple clinical systems that are likely to reside on it (electronic medical records,
pathology, cardiology PACS, and more).
Making It Compact
The last area regarding changes in archiving that needs to be discussed is data
compression. Since the adoption of the JPEG 2000 standard for data compression in the past
year or so, both lossless and lossy compression can now be used in long-term archiving
scenarios. Because the cost of SAN storage is so low, more than 95% of all images on SANs
today are stored either uncompressed or, most typically, using lossless compression, which
stores images at a compression ratio of less than 3:1.
There are trade-offs, however. Compressed images add a half second or so to the first
image display time to allow for image decompression. For the most part, though, the
benefit of having three times the storage capacity far outweighs a virtually nonexistent
reading delay.
While JPEG 2000 compression allows for significantly higher compression ratios (20:1
and higher), the fact that data could be "lost" even if it can be considered
"visibly lossless" has some facilities concerned about the medico-legal
ramifications of compressing out an artifact that was there when the original uncompressed
image was read. Once double-blind and receiver operator characteristic (ROC) studies have
been performed to compare lossless JPEG to JPEG 2000 compressed images and the public
becomes more comfortable using lossless compression for applications other than Web
viewing, then the industry will see a decade of image storage for a 500-bed facility
housed in an area about the size of a single PC.
Archiving strategies have changed for the better, and no doubt will continue to change
as technology advances in leaps and bounds. It is up to each healthcare facility to
determine the best approach and strategy to take regarding archiving, balancing price,
performance, and growth requirements for today, tomorrow, and beyond.
Michael J. Cannavo is the president of Image Management Consultants, Winter
Springs, Fla, a PACS consulting firm.
Storage Suppliesby Rick Romano
Given the overheated volume of diagnostic pictures and
patient text records generated nowadays by most imaging enterprises, it's a wonder
their informatics systems-employing yesteryear's wheezy storage
devices-don't simply burst into flames while straining to keep pace.
Manufacturers of data-archiving technology aren't keen about letting anything even
remotely like that happen, so they've been working hard at developing futuristic
solutions to help facilities and institutions gain speedy, cost-efficient control over the
electronic files building up within their PACS, RIS, HIS, and more.
"Today, with the advent of new imaging and workstation
technology, often the only thing standing in the way of the substantial cost savings
possible with filmless diagnostic operations is the archival, management, and distribution
of digital images," says Roger Swigart, director of business development for ODS
Medical, the healthcare division of Objective Data Storage (Columbia, Md). "In fact,
this might be the only 'demon' remaining to be conquered to achieve the
substantial economic, medical, and marketing benefits of filmless imaging."
 And conquer it they are. Need
convincing? Just check out this small sampling of what's currently available in data
archiving:
Array Corp USA (Brentwood, NJ). Unveiled
just this year, the Array 2905 Laser Film Digitizer can scan to 2K resolution in just 7
seconds and auto feed up to 100 individual 14- x 17-inch films at 50-µm pixel spacing.
Its variable sampling pitch, capable of changing spot size from 50 µm to 500 µm in
single micron steps, also allows a single sheet of multi-formatted images to be broken
into individual DICOM images. This feature is a huge benefit in terms of productivity for
radiologists who wish to compare patient studies captured directly from modalities against
the relevant prior films by means of a soft-copy workstation's stack and cine modes.
The 2905 also offers moiré-reduction filters to eliminate the hindrance of grid lines.
Breece Hill (Louisville, Colo). Contending
that even small organizations can benefit from the sophisticated data protection and
storage-management capabilities found in high-end disk and tape products, Breece Hill now
offers the iStoRA 4000. This integrated storage, retrieval, and protection appliance
combines disk, tape, and data protection software in a single rack-mountable turnkey unit.
It delivers seamless disk-to-disk-to-tape functionality, including staged backup, rapid
restore, and HIPAA compliance, all in a plug-and-play package. Features include a 1.5
terabyte (TB) serial-ATA disk array and a tape autoloader with a choice of tape formats
and capacities up to 13 TB. The iStoRA 4000 also integrates an Intel-based server with
data protection software to permit policy- or rules-based data traffic and storage
management.
Data Distributing LLC (Santa Cruz, Calif).
Hottest among Data Distributing's current product lineup is the Ciprico DiMeda 3600.
This RAID-based network attached storage (NAS) system is seamlessly integrated for PACS or
medical records archiving. Affordable, scalable, and easy to use, the DiMeda 3600 promises
99.999% uptime and fast access speeds for everything from file-sharing to archiving of
images small and large, even in multi-platform, multi-vendor environments. Features
include nonlinear editing, medical imaging PACS, and global command and control.
DeJarnette Research Systems (Towson, Md).
Because transplanting archived images from legacy PACS into new, state-of-the-art
replacement systems can be such a significant challenge, DeJarnette has introduced
PACSware Migration Gateway 2.0. A Windows-based tool kit, the solution supports multiple
migration strategies, including study date-based migration, platter volume-based
migration, priority RAID migration, prefetch-based migration, ad-hoc migration, and
double-pitch operation. PACSware Migration Gateway 2.0 also supports functionality, which
allows for the generation of premigration work lists (based on numerous criteria) and
postmigration incremental migration work lists. Further, it permits migration scheduling
(to minimize interference with daily clinical operations), multi-migration engine
operation (to speed migration), DICOM data cleanup of the legacy data, and data
verification.
Eastman Kodak Co (East Rochester, NY).
Efficient management of information stored across various platforms is the idea behind
Kodak's DirectView VIParchive (Versatile Intelligent Patient Archive) software. The
system features centralized, enterprise-wide management of images and information
associated with clinical systems-including radiology and cardiology-as well as
back-office systems, such as purchasing, inventory, and other records. VIParchive also
offers dynamic and transparent life-cycle management of information, IS media, and
technology; plus, it's storage-vendor neutral to protect against obsolescence. The
VIParchive platform is incorporated within the DirectView PACS System 5, and is a feature
of Kodak's Enterprise Information Management (EIM) Services.
EMC Corp (Hopkinton, Mass). Storing the
right data at the right time on the right storage medium helps facilities gain maximum
value from patient information, simply and cost effectively. That's why EMC has
partnered with leading PACS and electronic patient record (EPR) application providers to
deliver solutions that leverage storage investments across multiple applications and, in
so doing, provide true clinical information continuity. EMC solutions support both
short-term "live" viewing and long-term archiving across the enterprise. For
example, EMC CLARiiON and Symmetrix networked storage systems manage, protect, and share
short-term active files, while EMC Centera content-addressed storage (CAS) offers
flexible, long-term archiving of medical images, patient records, and other unchanging
digital assets. Also, EMC Healthcare Solutions for PACS and EPR are integrated with
radiology, cardiology, and hospital information systems to automate clinical information
workflow and optimize patient information life-cycle management.
IBM (Armonk, NY). IBM recently rolled out a
near-line disk storage for lower cost, long-term retention of infrequently accessed data.
It's an entry-level disk storage system named TotalStorage FAStT100 Storage Server
employing SATA (serial advanced technology attachment). When paired with IBM's
TotalStorage FAStT Storage Manager software, the FAStT100 forms a common management tool
that can help simplify the addition of new servers and applications. Meanwhile, the FAStT
Storage Manager software can be used to create, on board the FAStT100 Storage Server, as
many as 16 strategically allocated partitions in order to increase the use of storage
space while reducing storage management costs.
ODS Medical (Columbia, Md). ODS, a division
of Objective Data Storage, has announced a fresh concept in DICOM digital image
storage-the PACSstor Image Archive Appliance. This product brings together
high-performance RAID storage and advanced image-management technology in a single
platform. With it, DICOM studies can be stored directly from modalities or from PACS and
then accessed by departmental workstations. Administrative and management functions are
Web based, providing the ability to support departmental operations and meet HIPAA
requirements. The new system provides an advanced solution for redundant image storage and
disaster recovery, and it can support filmless operations in a variety of imaging centers
and smaller diagnostic imaging departments.
RADinfo Systems (Herndon, Va). RADinfo
Systems' PowerArchive is a HIPAA-compliant way to archive medical images for off-site
storage. It can be configured to produce multiple CDs or DVDs on a scheduled basis, or on
a resource basis whenever the CD or DVD becomes full. One copy of the completed disk
labeled for off-site disaster recovery can be stored with other vital facility data, such
as RIS, HIS, or office data backups. Users can view the CD or DVD on any Windows-based PC
without loading additional software. An FDA-cleared DICOM viewer is included.
RadVault (Hayward, Calif). SmartStor is the
name of RadVault's top-notch data-storage service, which features secure, off-site
retention of a copy of each patient's medical images for that patient's
lifetime. Ushered into a military-grade colocation facility and redundantly backed up,
SmartStor data are received using 128-bit encryption over secure-socket layer (thus,
ensuring image integrity and HIPAA compliance) and then are monitored for security around
the clock (data remain encrypted even inside the data center). As a bonus, RadVault
provides its SmartTrac data-tracking service free of charge so that users need only log in
via RadVault's secure Web site in order to monitor stored studies.
Rorke Data (Eden Prairie, Minn). For remote
SAN connectivity and disaster recovery, Rorke Data offers the Silverline-Coarse Wave
Divisional Multiplexing (CWDM) solution for use in applications relying on fiber-optic
cable (chiefly dark fiber or SONET). Enterprises so equipped gain bidirectional, 4:1
utilization over their existing single optical cable infrastructure or strand in a radius
of up to 60 miles. Silverline-CWDM complies with all SONET, SAN, and Ethernet
specifications and delivers complete protocol independence. No encapsulating or other
data-affecting applications slow down throughput. Since CWDM is an "optical
extender" for diverse data networks, users can mix and match other connections, such
as GigE or SONET, over this same single dark fiber. The result is CWDM pays for itself
quickly because multiple types of data are aggregated over one strand.
StorageTek (Louisville, Colo). Significant
enhancements are scheduled to be made to StorageTek's BladeStore, a cost-efficient,
"mezzanine-level" advanced technology attachment (ATA) disk-storage product that
reliably enables fast access and high performance to multiple years of archived studies.
Also from StorageTek is the newly introduced StreamLine SL500 modular library system. It
provides small and midsize facilities with scalable capacity and performance by using
mid-range tape-drive technology, which makes it ideally suited for deep archive and
disaster recovery of images.
TDK Medical (Garden City, NY). Looking for a
way to create CD copies of diagnostic images in DICOM format? Try the TDK CDRS-1100AUTO TP
Medical CD Recording Station. Software required to view the images is stored on each CD,
allowing users to view patient studies on DICOM-compatible review stations or virtually
any standard PC. The software suite includes TDK Medical's proprietary CDRS Image
Viewer, which allows viewing motion images from such modalities as ultrasound and
cardiology and includes eFilm Lite viewing software from Merge eFilm (Milwaukee). Images
can be exported into standard PC applications, such as Microsoft PowerPoint and Adobe
PhotoShop. Like all TDK Medical products, the CDRS-1100AUTO TP Medical CD Recording
Station is manufactured to ensure error-free, bit-accurate recording and management of
radiological images and results.
Rick Romano is a contributing writer for Medical
Imaging. |
An Eye on Opticalby Edward M. Smith, ScD, FACNP
A healthcare institution's medical information, such as
images from all clinical departments and clinical and business databases, is contained in
a healthcare enterprise archive (HEA), which is managed and supported by the
institution's IT department. The HEA contains both fixed content files (FCF), which
include images, wave forms, and structured reports, and variable content files (VCF),
which are the clinical databases located in a centralized data center. Rather than using
disparate storage silos that are difficult to manage and secure, each clinical department
stores its images and data in the HEA, which provides an economical storage solution that
is secure, accessible, and in compliance with current regulatory requirements. A variety
of ways exist to store data in this manner, one option being optical media, which provides
a secure and accessible alternative to magnetic media for the storage of FCF in the HEA.
What's Required?
Medical images must be stored in a manner that satisfies both clinical-operational and
legal requirements. From the clinical perspective, images must be immediately available
anywhere, anytime; the current standard is about 3 seconds for the first image. This
requirement implies an uptime requirement for the HEA of at least 99.9%, and preferably
99.99%-hence, a fast, redundant, and highly available magnetic disk system is needed.
Typically, 1–24 months of online storage is required, depending on the choice of
media and management software deployed for the balance of the archive.
From a legal perspective, images must be stored in a secure
and accessible environment for the duration specified by state and federal law, typically
7 years for adults. Effective April 21, 2005, the Health Insurance Portability and
Accountability Act (HIPAA) will require that two uncorrupted copies of all images be
retained at two distant and secure sites. That way, if the primary site becomes unusable,
the images can be obtained from the disaster recovery site (DRS) in both a secure manner
and a reasonable period of time.
HIPAA requires at least two tiers or levels of storage at the
two locations. At the primary data center, facilities should employ fast magnetic disk and
possible slower magnetic disk for online storage and magnetic tape or optical media for
long-term storage. The remote storage site would require a write once, read many (WORM)
type of media or a software solution to ensure that the data on the magnetic media cannot
be altered. Also required is hierarchical storage management software, which manages when
the images are moved from one media to another.
In it for the Long Haul
Storage technologies used by the HEA include magnetic disk, magnetic tape, and optical
disk. Each of these technologies has its advantages and disadvantages at different periods
in the information management life cycle.
Magnetic disk is designed for rapid access and uninterrupted
data availability. Because data on magnetic disks can be altered or corrupted,
sophisticated hardware and software technology is available to protect data; however, the
total cost of ownership (TCO) will increase significantly. Since rapid access is not a
primary criterion for the long-term archive or the DR archive, magnetic disk is not the
most cost-efficient solution for the total HEA solution. A combination of magnetic disk
and a lower cost media will result in a lower TCO.
Magnetic tape can provide a low TCO, excellent storage
capacity per tape, and flexible scalability; however, magnetic tape is inherently
rewritable, making it easy for the tape to be altered or corrupted. To ensure these data
are secure and not corrupted, two or more copies of the tape would have to be made and the
tapes periodically verified to ensure that data have not been altered.
Optical storage devices include CDs, DVDs, and ultra density
optical (UDO) disks. Data are stored on optical media by burning data onto the recording
surface using either a red or blue laser. Optical storage media are not affected by
magnetic fields; thus, data cannot be altered. Both CDs and DVDs provide portability of
patient studies. Currently, DVDs are being used for long-term archiving applications.
A blue laser burns data onto the UDO disk using optical phase
change technology. At present, the WORM UDO disk from Plasmon (Englewood, Colo) can
accommodate 30 GB in a standard 5.25-inch cartridge that can be used in many vendors'
robotic systems. Data are stored and retrieved from sectors with UDO, allowing rapid data
retrieval. Sony Electronics Inc (Park Ridge, NJ) has a competitive blue laser optical disk
that can store 23 GB per disk that is contained in a nonstandard cartridge.
During the initial phase of managing the information, rapid
access, high data transfer rates, and uninterrupted availability are most important;
therefore, fast magnetic disk storage is the optimal choice. To conserve cost, some HEAs
can use slower media, such as slow magnetic disk, magnetic tape, or optical media, after
the information is further into its life cycle-roughly 3–24 months. The storage
solution at the DRS also must ensure the information will not be corrupted or altered in
any way; plus, it must be accessible and result in a low TCO. At the DRS, however, rapid
access is less important.
The TCO of the various components of the HEA is a critical
issue. At its current storage capacity per disk, UDO is about twice the cost of tape and
about the same as DVD. However, UDO storage capacity is scheduled to increase to 60 GB in
2005 and to 120 GB in 2007. When compared to protected magnetic media, the cost of UDO is
lower by a factor of approximately 4:1.
Currently, the TCO per GB of 20 terabytes of UDO archive for
5 years would be approximately $11.15 per GB of native storage. Let's assume that a
typical study requires 40 MB of uncompressed storage. Assuming a lossless compression
ratio of 2:1, the cost to store a typical radiology study in the long-term archive for 5
years would be approximately 23 cents.
UDO has significant advantages in the long-term components of
the HEA compared to other storage media technologies. In addition to being significantly
less expensive than magnetic media, once UDO is written, the data cannot be altered, which
is the primary criterion for maintaining a secure and unadulterated HEA. UDO is being
supported by Hewlett-Packard, IBM, and Plasmon, as well as being used by Dell, GE
Healthcare, and Siemens. UDO media is manufactured by both Plasmon and Verbatim and is
compatible with robotics that support 5.5-inch format media cartridges.
Edward M. Smith, ScD, FACNP, is a professor of radiology
in the department of radiology at the University of Rochester School of Medicine and
Dentistry, Rochester, NY. |
 |
|
|
|
|
 |
|
Resources
