deWindt Research

Past and Current Projects at deWindt Research



Project Wormhole -- 1995-2005

Beginning in 1995 remote connectivity to workplaces and educational institutions was becoming a necessity even as the Internet was still in it's infancy. At this point in time it was mainly focused on employees and especially those responsible for maintaining campus networks and servers. While single node PPP connections worked for some, it was truly necessary for others to be able to bridge their home network to their work network as if they were physically sitting at their desk. This is where Project Wormhole found it's roots. Using custom built Linux 1.2.8 based machines with a very stable 28.8 modem this was thought to be a possible solution to the "bridging" problem that faced users. Ultimately the solution was simple: 2 386SX/25 Linux machines, each connected to their respective networks with an AT-1500T 10Mb/s ISA card, connected to one another over a single POTS line using a 28.8 Hayes modem at each end. Once the PPP connection was established over the serial line, each machine would then run a custom shell script to use /32 routing table entries and proxyarp entries to route small networks (less than 12 nodes) from the remote location to the campus network. This allowed the satellite locations to use small slices of a /24 CIDR without changing routing tables on any campus routers. After inital testing was performed in the lab, the project went live in early 1996. After some time in production the setup was enhanced so that each endpoint would run a heartbeat to the other end and when the connection became unstable or fully dropped, the connection was automatically restored. This proved extremely helpful for remote location where the quality of the telco plant copper was aging and was cluttered with signifigant line noise. With the release of Courier 56k modems all remote locations were upgraded and the campus site was consolidated to a single Pentium 150 server with a Cyclades Cyclom-Y multiport serial card to handle the inbound connections. Ultimately these connections grew to a total of 6 remote sites each with a second POTS line connected for a remote console port in the event system upgrades were needed. The project remained in continual use until 2005 when the setup was replaced with Ascend Pipeline 50 routers for 128kbps ISDN connectivity to 2 locations more than 4 miles from the NOC and NET-TO-NET 2.3Mbps SDSL network extenders for 9 locations within 23,000 feet of the NOC.

Project PPPserv -- 1997-2003

By 1996 dialup access was becoming a necessity for virtually everyone. Unlike in prior years when dialup connections were mainly used for IT staff and remote facilities, the need for end-user PPP and SLIP services was on the rise. Higher Education was beginning to embrace the technology and began deploying devices such as the well known (at the time) TribeLink8 which provided seemingly seamless network connectivity for both MacOS and Windows clients to campus networks and the Internet. However as the demand grew, the capacity of the TribeLink hardware became an evident problem as more and more units were needed to handle demand and costs were rising at an astonishing rate. The other unfortunate issue that was encountered was that the TribeLink8 had, unknown to many IT professionals of the era, a fairly high failure rate which caused increased maintenance costs and manhours to continually travel to campus to reboot a faulted unit. A solution was proposed that was dubbed Project PPPserv. The project was a very simple and logical one inspired by a local BBS that ultimately became a regional ISP. The proposal was to build a single Pentium 166 system running Linux kernel 1.2.13 with 2 Cyclades Cyclom-Y multiport PCI cards and a single AT-2450 10Mb/s PCI network card for remote to local client connectivity. A low end Trident PCI VGA card was specified due to stable kernel support of the chipset and the server was finished off with 32MB of RAM and a 250MB ATA hard drive. The system started with 16 USRobotics 33.6 modems connected to POTS lines and the heavy lifting was all done by mgetty, ppp and custom scripts to enforce idle timeouts and time limits. The two largest pieces of the project was allowing the system to run in parallel with the TribeLink8s which used RADIUS for authentication and it was necessary for the server to have a web GUI that matched that of the TribeLink8 as closely as possible. These features were accomplished with freeradius, Apache httpd, and PERL. As the original Windows NT RADIUS server that was used for the TribeLink8s stored passwords in cleartext, migration was easy and the password entries for local authentication was scripted. When a user account was added, it was added to the local password database and included in RADIUS via a PERL script. When the system was released into production not only did the staff accept the new GUI as satisfactory, they were also pleased that feature requested were easily satisfied in-house unlike with the TribeLink software. The Tribelink8s were retired shortly after the PPPserv project was complete and dialup service continually expanded flawlessly exceeding original expectations by handling 38 POTS lines and over 400 accounts. Ultimately this project remained in continual production use until 2003 when it was replaced with a PRI PPP server from Cyclades.

Project Stream -- 1998-Present

In 1998 the groundbreaking concept of real time video streaming over ATM networks was being tested in the Higher Education sector in Virginia. As a result of this endeavor, what eventually became known as deWindt Research started testing low resolution video and audio streaming techniques over both Ethernet and Dial-up circuits. Initially, testing was done using Cisco IPTV on campus Ethernet networks and over the point-to-point ATM connection between several HiEd sites. On private networks at the disposal of deWindt Research, various different technologies were tested. Initially CU-SeeMe was used for video conferencing over 33.6 analog modem lines between Farmville and Bedford Va which proved reliable for basic conferencing tasks. As production testing on the ATM networks expanded to include NetMeeting this too was tested on the Farmville/Bedford connection and proved too resource intensive. As time progressed the focus changed from video conferencing to video and audio streaming of real time events and prerecorded content. The first obvious, and ultimately the most stable, platform was determined to be RealServer by Real Networks. Using the same 33.6 modem circuits it was found that live audio and video as well as recorded content were reliably trasnmitted. Services such as real time audio transmissions of local emergency services and sporting events were soon easily streamed to remote clients over dialup. Other server types such as Darwin were tried with inferior performance results, leaving RealServer as the primary platform for streaming for the period. As streaming technologies matured the decision to change platforms was made and deWindt Research concluded that Icecast and VLC were the dominant winners. Icecast for its ease of deployment and VLC for its video quality. These platforms are the two used currently at deWindt Research.

Project GuestReg -- 2000-2015

Starting at the turn of the century mobile computing was starting to take shape in the form of laptop computers being more and more mobile. As a result it became common for houses, hotels, colleges and universities to have guests traveling to their location who desired network connectivity. Due to some sites expressing concerns about letting just anyone connect to their network the concept of allowing guests to use a network with limited or carefully monitored access became an issue. The solution that was formulated was a project named Project GuestReg so as not to be confused with the similar project named NetReg. GuestReg was complex internally but was operationally simple. When a MAC address unknown to the dhcp server was seen broadcasting a DHCPDISCOVER packet the server would respond with a lease that was allocated from a specific IP pool and was given a blackholed gateway and DNS server that limited its access to only a single server. This server could also be configured to blackhole the MAC address at the gateway router as well to prevent a user from trying to configure a static IP address. The client was then presented with a welcome screen asking them to register their identity at which point they were given a valid, non-blacklisted, address with a lease valid for 24 hours. The system administrator had the option to have all registrations send a notification via email or pager at which point any suspect registrations could be revoked either temporarily or permanently. The project ran on a standard configuration using ISC BIND, ISC dhcpd, Apache and PERL. The project was released as a beta in late 2000 with the last version with enhancements to take advantage of Apache2 and wireless access point ACLs being released in 2015.

Project Skynet -- 1997-Present

Project Skynet is by far the longest, continually active, project that has ever been undertaken by deWindt Research. Starting unoffically in 1996 as an intellictual conversation between colleagues regarding 802.11 wireless technology, it became a completely active project in 1997 with the availability of Lucent/ORiNOCO 802.11b PCMCIA cards. Using ORiNOCO Gold cards and external antennae, throughput and reliability tests for point-to-point connections were conducted for over 4 years. From direct line of sight connections to omni-directional point on campus to multiple point at differing distances the technology was tested for over 2 dozen use cases. Until approximately 2001 all of the antennae were manufactured by hand by the researchers with significant time spent in the field "war driving" with laptops and Garmin GPS systems to map results. The research continues even in 2020 as different vendors, standards and applications are tested. Given the drastic drop in the prices of legacy gear it has become easier to research what older, lower bandwidth, connections can be brought up in a stable environment to reach farther distances with drastically cheaper hardware. At the time of writing dewindt.us utilizes, and performs testing with, hardware from the following manufacturers and standards respectively: Cisco, Netgear, Lucent/ORiNOCO, Linksys, Broadcom and Intel utilizing 802.11a, b, g, n and ac.

Project WhiteRose -- 2015-Present

In 2015 a small research project was started named Project Whiterose with two Raspberry Pi2s to evaluate what realworld production use such small units could provide given their small size and low power consumption. Initially they were setup as just a 2 member cluster pair running some small calculations for hypothetical uses which provided a baseline of their usability. After this was completed they were returned to a Day0 configuration and setup as a Master/Slave MySQL pair which was put into production use for other development projects. As these 2 units functioned in this role, evaluations of the complexity of enterprise deployments of EqualLogic and NetApp network storage devices were conducted with startlingly disappointing results given their price and ongoing licensing costs. Given this knowledge, the focus of Project Whiterose was changed to network storage. An additional 2 nodes were added to the cluster as well as additional storage turning the cluster into the framework for a full fledged NAS. The final configuration replaced a pair of 2 Intel Core servers with 2 Pi4s, 3 Libre Renegades and 2 AMD nodes which were attached to 6 external spinning disks, 2 external flash drives for storage and a single Intel Core2 node for Archive use. All nodes were attached to a dedicated 1gb/s switch with the controller node being connected to the core production network switch. All backups are automated to live disks, offline RW ROMs and long term storage of system configurations are stored on LTO. The project was brought into full production use for dewindt.us in late 2019 and has been running with no unexpected errors ever since. The controller node exports all of the disks via NFS and CIFS for easy accessability. Currently the cluster contains over 16TB of SATA storage with the flash storage used for temporary storage, logging and file transitions. The cluster has been proven to be as reliable as a drobo Pro system at a much smaller price and has the advantage of increased capacity and fault tolerance. Current details of the Whiterose NAS may be found via the main dewindt.us homepage.

During 2022 a need arose to expand not only the capacity but also the speed of the WhiteRose NAS. The first step was to add a new filer node to replace the aging Intel Atom system. A Pi 4B was chosen and during the replacement process 2 4TB USB3 disks were added to the NAS. As the year progressed the 2 existing Pi2 files and the Pi3B+ controller were also beginning to to pose speed concerns with the ever growing amount of data being stored. Due to the huge price increases in Pi hardware due to supply chain issues, substitute hardware was researched and ultimately 3 Libre Computer Renegade ARM nodes were purchased along with an additional pair of 4TB disks. With these nodes up and running it was decided to replace the Pi3B+ (controller0) with a 8GB Pi4B already on hand. Controller1 was upgraded to a Core2 Duo system with internal and external SCSI for disks and an LTO4 drive and 2 new management nodes were installed. The NAS network was upgraded to a fully managed AT-9924T-EMC Layer 3 switch with a redundant AT-FS724L switch in hot standby. The FLASH storage was upgraded along with the main work/logging disks boosting the system to 1/4TB of true flash storage and a RAID mirror running on SSDs.

Project Blackhole -- 2020-2024

A recent project that is now an active piece of the dewindt.us network backbone is Project Blackhole. During a routine router upgrade to the demarc router at dewindt.us it was decided that in lieu of routing all firewalled data to the bitbucket it would be better used as ITSec data and routed to an internal null address for analysis. The first step was reconfiguring the Netgear edge router to perform all of the port forwarding necessary to keep external services operational (http, ssh, dns, etc.) while sending all other, potentially hostile, data to a null internal address. After extensive testing of the configuration, and with no problems encountered, a dedicated hardened Linux machine was configured to reply to all inbound packets that were not destined for a legitimate service offered by dewindt.us. As the data flowed to the isolated NIC on the hardened node, the construction of the data analysis and management interface began. The result has been named the Blackhole IDS. The key features are packet reception and digestion of all illegitimate traffic originating from the Internet. This data is classified by target port, originating IPv4 address, timestamp and payload. Summary logs are generated hourly and historical results are kept indefinately unless manually cleared by the system administrator. Nightly scans are also run on the internal networks and the appearance of new, previously unknown nodes, is logged and reported. Unlike Project Whiterose, whose live data may be viewed by visitors to dewindt.us, the Blackhole IDS is not publicly facing, but anyone interested in the project can contact deWindt Research for screenshots and more information about the configuration.

Project SERMUX -- 2020-2023

With the primary server racks at dewindt.us being located on the first floor of the building and the office space and lab on the second, it became clear that remote console access to the switches and routers on the first floor was needed. It had become cumbersome to attach console cables to all network hardware in the main racks and connect it to servers with free serial ports that could use minicom for access. The first options that came to mind were using reverse telnet on a Cisco router using multiport serial blades and octopus cables, installing a DEC terminal server and using LAT to connect remotely or installing a Cyclades Cyclom card in a server and use it as a dedicated serial server. After pondering the costs of these potential solutions, and replacement parts given the age of the discontinued Cyclades cards and accessories it was determined that some sort of solution based on a Pi Zero would be the most logical choice. After much research, a Pi Zero W was installed in one of the main racks with multiple Gearmo 4 Port USB to RS-232 FTDI adapters connected. The project was then named SERMUX (SERial MUltipleXor) and coding of a complex operating environment called smsh (SERMUX Shell) which mimics many aspects of Cisco IOS and f5 TMOS tmsh. This solution has been in production since late 2020 and the source for the project is open source and available upon request.

Project Metcalfe -- 2020-Present

As 2020 began to draw to a close, more and more Enterprise grade networking hardware was becoming readily available on the open market at a fraction of the normal resale cost. In this buyer-friendly market, deWindt Research purchased a quantity of Layer 2 and Layer 3 Ethernet devices and constructed a small testing lab named after the co-inventor of Ethernet, Robert Metcalfe. The goal was to build a completely isolated network consisting of multiple Layer 2 switches for edge use and at least one Layer 3 device so that the lab was accessible from the production network while remaining isolated by a firewall to prevent any unwanted impacts on other systems. The goal was accomplished and the infrastructure is now complete and in use. For the internal Layer 2 switches, Cisco Catalyst 2950 and 2960 units were chosen for their features and capabilities while the edge, Layer 3, device chosen was a Juniper Netscreen 204. The Netscreen integrates completely with the production Cisco routers and grants the ability to run testing scenarios from any office location outside of the lab as well as remotely via tunneling should it prove useful. With the facility fully operational, deWindt Research is going to expand its performance testing from server and storage hardware to complete end-to-end performance testing using the Metcalfe lab hardware. With technologies such a PVST, Ethernetchannel, Port Trunking and 802.1Q become easier and more economical to deploy in SOHO environments, this research should prove useful for a variety of reasons.

Project BlueIron -- 2022-Present

As the planned expansion of both the production and research server clusters drew to a close in July 2022, the retired servers were free to be repurposed for a project that was completely non-mission-critical or be sent to surplus. As the machines were all completely functional IBM's of virtually identical configurations, it was decided to turn them into the first parallel computing cluster ever constructed at deWindt Research. The machines are now running for experimentation purposes only with a single node acting as the coordinator and the other nodes being assigned the role of worker nodes. The system is up and running with SLURM, ganglia and other custom clustering tools written in-house on CentOS 7.9. This OS was chosen as the machines themselves are all 32-bit, the reason they were removed from production and research, and an RPM based system was the most straightforward to use for the first attempt at building a parallel computing cluster. More a proof of concept than anything, BlueIron is currently running test routines to merely prove all of the software works as designed. Once ample time has been found to test and explore all of the intricacies and specifics of parallel computing a true cluster may be considered as a future expansion project if a use can be found. The system is named BlueIron for the fairly obvious reason that it is comprised of solely IBM (a.k.a. Big Blue) physical (a.k.a. iron) nodes. To save on power and cooling, BlueIron will only have all cluster nodes booted when real testing and experimentation is being done and not 24x7x365 like production.

As 2024 began, it was decided that it was time to deal with the fact that the BlueIron Cluster was deployed using the now-deprecated CentOS Operating System. Some experimentation was performed using RedHat Enterprise Linux VMs and it was determined that since all other systems are Debian based, there was no longer an advantage to continue to utilize an RPM based system solely to follow existing documentation. The decision was made to migrate to Debian 12 x86 as that allowed a vastly updated kernel and layered packages to be deployed. As of 2025 the migration to Debian is complete and the BlueIron network is now isolated from production behind a Juniper Netscreen 204. All nodes are configued for Wake On LAN allowing all or some of the members to be booted for any given project.

Project Rewind -- 2024-Present

In early 2024 the decision was made to try and get back to the origins of dewindt.us by way of acquiring a Sun Sparc system to experiment with using a modern Linux distribution. After much research, a Sun Microsystems Sparc Ultra 5 system was located, still in the original box, for sale at a reasonable price. The system ws brought online using the unstable Sid branch of the Debian distribution and was integrated into the production cluster. After additional research over the following months, it was determined that the core programmers for Debian Sid sparc64 were in many cases the same programmers working on Debian Sid for PowerPC. At this point a back burner project became to keep an eye out for a reasonable deal on some Apple G3 or G4 systems to dedicate to working with Debian ppc. In March 2025 a matched pair of G4 Sawtooth towers was acquired, upgraded and put into service. With a matched pair for the ppc build, a second Sparc Ultra 5 was purchased to match the exising system yielding a matched pair of systems for the sparc64 build as well. Using these systems, dewindt.us can once again contribute directly to the Linux community in our own small way. We will continue to leave these systems up and running as long as the Debian projects for their architectures remain active.



Copyright © 2024 - 2025 dewindt.us, All rights reserved.