Transcript Slide 1
SCCS Networking Strategic Plan, May 2005 Increased Reliance • Net now a critical part of infrastructure, its absence prevents much work being done • Requires planning, redundancy (power, paths etc.), management, monitoring, automated trouble detection and diagnosis • Need to work to integrated communication system: – VOIP, cell phone, voicemail/email/IM • Increased movement into un-networked areas (warehouse, bldg 24, ES&H) delays people moves Increasing Diversity • More enclaves: BSD, SSRL, Kavli, LCLS, MCC … • OS’: Linux, Sun, Macs, Windows: each has own configuration interfaces/tricks, duplex foibles etc. • Applications with new network requirements such as jitter, QoS: bulk-transfer, ssh, http plus Web2, apps servers, audio/video delivery, VoIP, Skype, RT control, instant messaging, Network backup, remote backup • More users, more transient population – Less support from local groups, smaller groups, more groups – Greater diversity of support needed (hdw, swr, apps) – More transient and diverse equipment on network • Mobility Impact of Security • Drives increasing use of enclaves – 10 firewalls today, need central firewall management • Need faster approval of applications, projects – Remove approval chokepoints, makes harder to manage, increases aggravation, delays implementation • Makes network/applications less transparent, less easy to use • Blocks ports etc. – Applications fail, apparently network related requires debugging – Debugging tools fail without apparent reason – Needs increased auditing, gluing together of multiple databases/tools etc., • Visitor subnet evolution LAN • Major upgrade started in 1996 to structured wiring and switched network • Upgrade to support Gbit/s to desktop in next 2-3 years • Many switches now reaching end of life already off maintenance, need replacing • Need to clarify funding for non scientific needs: – Projects move in and out of buildings (SLD=>BaBar=>GLAST=>…) – Multiple projects share buildings – Project funding is a poor match to basic requirements • In-house management tools to meet our needs – Commercial tools, expensive to buy, maintain, update, use, only partially met needs – Developers close to moving on – Need a new generation • Mobility, (e.g. wireless, PDAs etc.) • New equipment, integrate VPNs, security? Aging Technology • TCP/IP designed for 64kbps • Bulk-transfer problems on long RTT, fast nets – Requires deeper understanding, new protocols: new experimental TCP stacks, UDP transport, direct data placement, • New apps needs: jitter, QoS • No accounting • Dedicated circuits: – WDM technology – Reservations, policing, accounting, new protocols (MPLS, QoS) • Higher speeds: new cable/fibre to desk, >= 10Gbps • Not designed for today’s security environment Wireless • Originally desirable for a few conference rooms • Increasingly seen as part of required infrastructure, extending to whole site – Enables mobility • Requires training and mastering an extra technology, technology is evolving • No extra support manpower added • Centralized support provides inter-working, uniformity, reduces management • Wireless medium not as robust/reliable as wired, so more problems, troubleshooting, assistance (e.g. conferences, APs shorter lifetime, interference,coverage) • New security requirements: war walking, registration, increased complexity etc. WAN • Increased worldwide collaborations with increasingly diverse communities (not just HEP) – Future less SLAC centric (no HEP expmt@SLACto bring people here for extended periods) • Harder diagnosis of problems, less control, rely on others • Need to mange more contact lists, • More interactions, travel*, remote conferences (phone, video), people networking Monitoring/Management • At higher speeds many of current probing tools fail (timing and NIC offloading) • On dedicated circuits traceroute fails • Ping and probe ports are increasingly blocked • Too many time-series graphs to manually review to allow pro-active problem detection – Need automated problem/anomaly detection • Network too diverse to easily diagnose problems – Need automated diagnosis assistance • Add accounting for resource consumption, SLAs etc. Support • Network Eng. team has shrunk in last few years • Kept very small by requiring uniformity – Single vendor for switches, routers, APs • Focus training & support • Can use vendor specific tools to manage • Implementations of management tools (SNMP, CDP) interwork better for single vendor – Single host per cable • Simplifies automation of knowledge of where things are – Need for problem isolation, user self-help, security tracking – Over-provision network avoid need for QoS, tight management – Avoid locally optimized solutions, but can cause push back • Apparent 24x7 due to dedication of team • Significantly under-strength • Have been doing networking on cheap – Buildings with minimal networking (260, Kavli etc.)