Overall Guidelines ------------------ No *NEW* developments under MIPS/Ultrix, but existing services can continue to run on such platforms. New developments should be done under Digital UNIX (OSF). High-Level Network services can be classified into various groups (not defined yet). Services from some groups can happily co-exist together. However, services from wildly different groups should not be mixed on the one platform if at all possible. [ Like the "servers serve, routers route rule". ] Possible categories System friendly: singe server process, UDP, no disk usage e.g. DNS, BOOTP, NTP Average: user typing limited, slowly growing disk usage e.g. multiuser work load System unfriendly: unpredictable number of processes & memory usage, uses up (all) disk when things go wrong, requires monitoring and manual intervention to fix problems e.g. Mail hub Another set of categories Enterprise-wide mission critical services: BOOTP, DNS Support Infrastructure: Mail hubs, Web proxy Content provision: Web servers Occasional: backup tape servers, Experimental: IRC, Email->FAX, video-on-demand, CU-SeeMee, Email-.Pager, Kerberos ... Current Servers and Service Types --------------------------------- Netslaves: Small, very reliable, provides Essential Services. No user accounts, NFS mounts, incoming Mail, or X windows. Severely limit accumulating things to disk -- remote syslogs, etc. All netslaves (nearly) identical to reduce configuration problems, ease recovery from hardware failures. Low CPU and I/O loadings. Reboot time under 2 minutes. Boots and runs without network. Services: BOOTP, DNS, NTP, bootloads, backup Annex security server Netgod: Central network configuration manager. Reliable. No production NFS mounts or X windows. Accounts only for Local Network Administrators. Local Email only. Boots and runs without network. Modest CPU, I?O, memory demands. Services: DNS master, BOOTP master, Netslave configuration master, Primary Annex security server & master, monIP, netcheck Warp: Important function as Web gateway through firewall. Web cache: speeding user requests, saving Internet charges. Lots of Memory, I/O bandwidth and CPU to handle the expected huge numbers of proxy requests. RAID 0+1 for speed and reliability. Few user accounts. Reboot after crash expected to be very slow due to huge number of files in cache. Services: Web proxy and cache Weft: Lots of important intellectual property. Few user accounts, but data disks available R/W from silas. RAID 5 for reliability. Lots of user disks NFS mounted so that their contents can be made available through the Web. Services: http, ftp, gopher, harvest, server stats ... Legacy applications: brother: coolcat gateway, WAIS mogul2: Unix packet sniffer mogul3: MBONE router old-mother: addhost database not ported to netgod yet cisco configuration monitor vaxc, vaxa, silas ... mail hubs, cisco traffic stats nice: X.500 Directory service harbinger: News ================================================================== New Projects ------------ Internet Accounting Project --------------------------- Hardware Requirements: A new, small, dedicated Digital Unix box required for raw packet collection. Modest memory and disk requirements (similar to Netslaves). Some CPU time and lots of disk space required on another larger Digital Unix box for fetching packet counts from collector box, producing summaries, per-department usage figures and for long term data storage and trend analysis. Software Requirements: Free: NNstat (or NetTraMet). Programming Requirements: Estimate 3 months work for a recent graduate. 1st month: Configure the new packet collector box O/S. Build the NNstat or NetTraMet software. Get the raw packet collector running. Get the snapshot collector running. Produce initial statistics. 2nd month: Analyse setup to determine data collection granularity, collection interval granularity, etc. Implement correlation of IP address back to Department / Faculty / CC Host / PC Lab or whatever. Automate weekly / monthly rollover and reporting processes. 3rd month: Customize data collection and reporting, e.g. VRN / Victoria / Australia / O/S breakdown. Reports by higher level protocol: http, FTP, Telnet, NFS ... Analyse for unusual traffic, e.g. http from non-approved Web server. Tune output to match customer requirements. Directory Services Project -------------------------- Hardware Requirements: Requires one server machine 128 MB RAM, several GB disk. Backup / distributed servers are also a good idea. Software Requirements: Requires lots of real-world modeling, "service" (rather than product) development, and procedure integration work. Estimate .5 to 1 man-year Mail Hub Project ---------------- Hardware Requirements: Will require one "server" per "campus" with lots of Disk, I/O bandwidth. Large campuses will require AlphaServer 1000 class machine. Pilot phase: 1 server. Software Requirements: Still undecided as to VMS or Unix based (will require port of Name Router). Estimate 2-3 months work. Some on-going maintenance / management required. Post Office Project ------------------- Basically an extension of the Mail Hub Project Hardware Requirements: More disks, memory etc. Software Requirements: Estimate 1-2 months work. On-going account and disk usage maintenance required. Multiple Web Proxies -------------------- Hardware Requirements: Will require one "server" per "campus" with lots of Disk, I/O bandwidth. Large campuses will require AlphaServer 1000 class machine. Pilot phase: 2-4 servers. Software Requirements: Need to investigate hierachical v. distributed caches Need to investigate different proxy software Estimate 2-4 months work. Experimental Projects --------------------- IRC MBONE CU-SeeMee Audio/Video on demand Kerberos Firewalls / Telnet proxies Authenticated Web / Proxy service Need some separate non-production machines that can be used for investigating these and other projects that arise from time to time. "Pilot" machines are probably not usable either because they will become production / moved to other campuses etc. Workstation Upgrades -------------------- mogul3 has a sub-optimal screen: 1024 * 768 1024 * 800 is required for 4 * VT220 screens 1280 * 1024 would be even better for having more workspace, viewing network management maps etc. mogul3 has a superseded architecture Digital don't make them any more No new Turbochannel options such as Fast Ethernet, MPEG video No-one writes new code for them (see above) *Netscape* isn't available for DECstations !! mogul3 isn't very fast only 25 MHz !! full-screen updates happen very slowly