---------------------------------------------- Choosing a corporate Internet Service Provider Martin Cooper <mjc@cooper.org.uk> ---------------------------------------------- The UK Internet industry's rate of growth isn't slowing down, although market forces have been causing and are continuing to cause consolidation, as small service providers are acquired by larger ones - often telecomms suppliers looking to buy in Internetworking expertise to expand into that market. As the link between voice and data networking becomes increasingly blurred by technologies such as Frame Relay ("FR") and Asynchronous Transfer Mode ("ATM"), services such as Real-Audio (tm) and Voice over IP ("VoIP"), and the further liberalisation of the telecomms market leading to ISPs diversifying into providing telecomms services as well as the reverse, choice has been vastly increased. Inevitably this increased choice makes selecting the right supplier to meet your needs that much harder. In this article I intend to address the integrated ISP and Telecomms service provision business by considering a small cross-section of the 67 members of the London Internet Exchange (LINX), the London Docklands based Internet Protocol Exchange Point and Trade Association. I have chosen to examine 9 ISPs, most of whom have been industry players for several years in various guises (UUNET UK, PSInet UK, BTnet, GX Networks, INSnet, VBCnet, Easynet and Netkonect), and a more recent market entrant (Level3). On the way to comparing UK suppliers I intend to consider the business in terms of IP and telecomms network architecture, connectivity and hosting services, value-added services, customer and network support, and Service Level Agreements ("SLAs") / pricing policy. Network architecture -------------------- IP networks can be considered as part of the Open Systems Interconnection ("OSI") model in which they appear at layer-3 (also known as the "Network" layer). Such networks can be carried over a variety of OSI layer-2 (the "Data-Link" layer) networks including IEEE 802.3 (Ethernet), High-level Data Link Control protocol ("HDLC") or Point to Point Protocol ("PPP") over X.21, G.703 or High Speed Serial Interface ("HSSI"), Frame Relay and Asynchronous Transfer Mode, the Synchronous Optical Network ("SONET") and the Synchronous Digital Hierarchy ("SDH"). There are a number of factors to consider in the selection of the underlying layer-2 network upon which the IP network architecture will be built. - What speed do you wish to run trunk links at? - Do you require resiliance? - Do you want to be able to easily split the bandwidth into chunks? - Do you need to define Quality of Service ("QoS")? - Do you need to dynamically assign bandwidth? - Do you need to specify a complex topology with multiple links and complex routing? [ Box 1 - START ] Comparison of types of layer-2 network -------------------------------------- IEEE 802.3 (Ethernet) - Asynchronous but non-deterministic (CSMA/CD protocol) - Broadcast - Specifications exist for speeds of up to 10Mbits/sec (Ethernet), 100Mbits/sec (Fast Ethernet) and 1000Mbits/ sec (Gigabit ethernet) - Usually switched to allow some degree of Quality of Service ("QoS") to be provided - Can be run over copper or fibre cables ("bearers") HDLC/PPP over X.21, G.703 or HSSI - Layered over X.21 and copper cables ("bearers") for relatively low-speed (64 - 768Kbits/sec) use, or G.703 (overlayed with G.704 if channelisation is required) for higher speed (2Mbits/ sec) use. HSSI links require special cables and are generally only used over very short distances, but are capable of speeds of up to 52Mbits/sec - Synchronous - Point to point - Multilink PPP can load-balance over two parallel circuits - Built-in error correction protocols Frame Relay/ATM - Optical Carrier ("OC") / Synchronous Transport Module ("STM") speeds i.e. 155Mbits/sec, 622Mbits/sec, 2.4Gbits/sec - Asynchronous - Point to multipoint - Expected to be run over reliable physical network or with higher layer protocols overlayed, therefore limited error correction - Network is conceptualised as a shared bandwidth frame/cell relay "cloud" with multiple nodes - Permanent and Switched Virtual Circuits ("PVCs" and "SVCs") with Quality of Service ("QoS") parameters e.g. committed bandwidth, delay, burstable bandwidth etc. - Switching of small fixed size frames (125 bytes with FR) or cells (53 bytes with ATM) increases bandwidth overhead - Can handle dynamic routing over multiple physical links with ATM Private Network to Network Interface protocol ("PNNI") SONET/SDH - OC/STM speeds - Synchronous - Point to point - Expected to be run over reliable i.e. optical, network, therefore limited error correction - Specifies breakout speeds for tributary synchronous data-streams (channels) - "Automatic protection and restoration" mechanism for fail-over to redundant fibre if primary fibre fails [ Box 1 - END ] At the IP layer every entity that is connected to the global Internet must have a globally unique IP address. Global IP address space is assigned in chunks ("prefixes") with a specific starting point in the global address space, and of specific length ("prefix length"). IP address space is allocated hierarchically from large chunks ("aggregates") to small ones ("assignments") to best match the Internet routing hierarchy. Routing of IP traffic on the Internet requires the distribution of knowledge about the reachability and location of IP prefixes. On the contemporary Internet this is done using the Border Gateway Protocol ("BGP") version 4. IP networks on the Internet interconnect using BGP4 based on two broad politico-commercial models, "peering" and "transit". Peering is usually a non-settlement (i.e. non-retail) interconnection between two networks for whom exchanging their respective locally originated routes (and therefore traffic) is of approximately equal mutual benefit. Transit is usually a settlement (i.e. retail) interconnection between two networks where one provides the other with non-locally originated routes (and therefore carries traffic that is not destined for its own networks). Most UK corporate ISPs use a combination of peering and transit to achieve full global connectivity and use BGP4 configuration to "tune" this for optimal routing. Good networks are those that have the best (most optimal) routes to any particular global destination, and alternate routes should the best routes be temporarily unavailable. Ideally an ISP of choice will be present at multiple geographically distributed Internet Exchange Points ("IXPs") and have good peering agreements with other ISPs at them. They will also have at least 2 transit providers to get the routes they don't see via their peerings. [ Box 2 - START ] Possible characteristics of optimal routes ------------------------------------------ * Shortest path to destination network * Highest bandwidth intermediate networks * Least congested intermediate networks * Most reliable intermediate networks * Most secure intermediate networks [ Box 2 - END ] Selection of network hardware platform, duplication of key equipment, telecomms infrastructure and power supplies, and diverse routing of physical network links are of high value in developing a high-performance, high-reliability network architecture. Cisco hardware is used by most of the Internet for IP routing equipment, and increasingly its Lightstream and Stratacom ATM hardware is gaining in popularity. Fore Systems has a good reputation for its cutting edge ATM technology as does Cascade for its Frame Relay switches. Nokia is widely respected for its SDH equipment. Connectivity services --------------------- Standard corporate ISP connectivity services fall into 4 main categories: - Routed ISDN - Frame Relay/ATM cloud - Leased line - Direct Special corporate ISP connectivity services include: - Pay per bit - Committed bandwidth with burst capability - Split national/international bandwidth - Regional only transit e.g. UK, EU, or US only - ISDN backup - Dual connect with diverse routing - BGP4 multihoming Telecomms services include: - Dark fibre lease - SDH bandwidth - Frame Relay/ATM cloud access - FR/ATM Virtual Private Networks ("VPNs") Hosting services ---------------- ISP hosting services usually involve one of: - leasing web space on a shared server - leasing a dedicated server - leasing rack space and co-locating your own equipment Value-add services ------------------ * IP and domain name registration IP addresses are initially assigned by ISPs based on your predictions of how many end-systems ("hosts") you will have on your network in 1-2 years. Domain names are usually registered when you first connect and are most often used to identify web services (www.domain) or email users (@domain). * Domain Name Service ("DNS") If you are unable to run your own primary DNS server to authoritatively answer queries from the Internet about your domain name(s), your ISP may be able to run it for you. Most ISPs will provide secondary DNS for your domains as a backup and to load-balance with the primary. * Mail Exchanger ("MX") backup This acts as a protective buffer for your primary MX when it becomes available after a hardware failure or extended loss of Internet connectivity, by load-balancing inbound email with the primary and forwarding it on to the primary in a controlled manner to avoid overloading it. * Smarthosting This avoids the need to run a local mail server to handle queueing and retry for temporary failures to send outbound email to particular destinations. * USENET news News is a series of discussion groups organised into hierarchies and distributed between "peered" news servers by means of a flooding algorithm. News can either be provided as a "streaming feed" using the Network News Transfer Protocol "NNTP") or by giving Network News Reader Protocol ("NNRP") access to your network to allow news reader software to read news directly off the ISPs news server. * Web caching A web cache is a central "proxy" for directing requests to download World Wide Web documents at, which retains local copies to reduce transit bandwidth usage and latency for popular documents. * Multicast/MBONE The MBONE (Multicast Backbone) is a virtual network overlayed on the Internet for carrying real-time multicast services such as streams of audio-visual data. * Network Time The Network Time Protocol ("NTP") is a distributed clock update protocol and can be used to automatically keep the time on your servers correct and synchronised with each other. Customer and network support ---------------------------- Most ISPs have a dedicated "new site" team who co-ordinate new customers' service installations. After this, support is usually provided through a combination of the use of email and the 'phone for configuration changes and fault reporting respectively. 24x7 support based on shift patterns is becoming increasingly common although all ISPs should (at a minimum) provide a pager number for urgent out of hours fault reports. You should also expect all ISPs to be able to do: - Advance warning of scheduled maintenance - Notification of unscheduled maintenance owing to network problems - Customer access-circuit outage notification - Collection of bandwidth statistics on your access-circuit or switch/router port - Replacement of failed or faulty equipment with on-site spares or a 2-4 hour maintenance contract Service Level Agreements ("SLAs") / pricing policy -------------------------------------------------- SLAs and pricing are closely linked and it makes sense not to pay for guarantees you don't need. It is worth considering exactly what level of service you require and negotiating on price if you don't need the standard SLA. SLAs usually cover areas such as network availability, available bandwidth, and average latency. Pricing of Internet services varies a great deal from one provider to another. Different packages cost different amounts to provide and clearly are perceived and valued differently by prospective customers. It is always worth checking what is and isn't "bundled" into the price and negotiating if you don't need particular elements of a package. 64K and 2M lines have become somewhat commoditised so don't expect to be able to negotiate too much here. The UK industry does appear to be quite polarised on price, from almost unbelieveably inexpensive, to astonomically expensive, and curiously the quality does not always match the obvious expectations at either end of the scale. [ Box 3 - START ] ISP leased line prices ---------------------- Leased line prices vary widely acccording to provider geography and access-circuit availability. All price ranges are in UK pounds. Bandwidth Approximate price range --------- ----------------------- 64K #6,000 - #9,600 128K #12,000 - #16,000 256-512K #17,000 - #31,000 2M #32,000 - #48,000 [ Box 3 - END ] Players ------- * UUNET UK - 8 UK Points Of Presence ("POPs") (including Telehouse London) - Peering at LINX and various US and European NAPs/IXPs - Transit from UUNET US - Duplication of all key functions - Cisco and Juniper equipment - 24x7 monitoring and telephone fault reporting - Optional leased and managed router - Can do pay per bit service - Dual connection for resiliance with "Multi-connect" service - Will do BGP4 multihoming - Can do server hosting - Try to be quite standard in their approach to service provision - Usual value-add services - Multicast routes carried internally - Support for real-time services - Good SLAs on network availability, available bandwidth and latency - Most expensive UK corporate ISP! * PSInet UK - 8 UK POPs (including Telehouse and London hosting centre) - Peering at LINX. PSInet US will do free private peering in US - Transit from PSInet US - Increasing ownership/lease of own dark-fibre network, so low cost of ownership - UK backbone network based on Frame Relay - Cisco routers and Cascade Frame Relay equipment - 2-4 hour fix for service affecting faults - 24x7 pager for fault reporting - SNMP customer equipment monitoring - Service can be supplied with or without a router included - Can do hosting at London hosting centre - Will do BGP4 multihoming - Usual value-add services - Good SLAs on network availability, and available bandwidth - Good value for money. Always prepared to negotiate on price * BTnet - 14 UK POPs (including Telehouse) - Peering at LINX - Transit from Cable and Wireless US, GTEI, Digex, UUNET US, Concert (Europe), Level3 - Implement resiliancy wherever possible - Aim to keep backbone network circuit loading below 60% - Cisco router and Cascade Frame Relay equipment used - Spares/support contracts handled internally within BT with strict SLA - 24x7 monitoring and (full) support - Option of purchasing router with connection - Will do BGP4 multihoming - UK only bandwidth service (will also do US or EU only if required) - Usual value-add services - No formal SLAs as yet (i.e. best efforts service), but they are due "Real Soon Now" * GX Networks - POPs in London, Manchester and Cambridge - Peering at LINX, MAE-East (US), MAE-West (US), D-GIX (Stockholm), and AMS-GIX (Amsterdam) - Transit from CERFnet, Verio and Ebone - Indefeasible Right of Use ("IRU") on STM-1 to New York - Transit/hosting swap arrangement with Concentric Network (US) - Access to Fibernet's UK SDH network - Access to Energis' UK Frame Relay network - Data/hosting centre at Harbour Exchange, London Docklands - ATM switches for providing managed ATM services - Aim to keep backbone network circuit loading below 70% - Cisco routing and ATM switching equipment, and Cascade Frame Relay kit - "Cisco Powered Network" accreditation - Spares holding for most equipment. 4 hour support contracts for major problems - Leased and managed router with most services - Can still opt for collected (rather than delivered) service - 24x7 monitoring and (full) support - Will do BGP4 multihoming - Usual value-add services - "Flextra" staged bandwidth upgrade service - "X-Frame 2000" 128K/2M IP over Frame Relay burst service - "Headroom" split national/international bandwidth service - Fairly flexible on providing special services - Fairly good SLAs on network availability, and available bandwidth - Has DTI telco licence * INSnet - UK POPs in London and Manchester (including Telehouse) - Peering at LINX, MAE-East (US), D-GIX (Stockholm), and AMS-GIX (Amsterdam) - Transit from UUNET US and CAIS - SDH capacity on Gemini transatlantic cable - STM-4 SDH ring around London with STM-1 to Wimbledon and ATM overlayed on this for easier bandwidth management - SDH automatic protection and restoration and ATM PNNI routing for resiliance - Cisco IP and ATM equipment. Nokia SDH kit. - Full set of spares at Wimbledon NOC including Power Supply Units ("PSUs") with various support contracts - Aim to keep backbone network circuit loading below 65% - 24x7 pager/mobile for fault reporting. Soon to move to 24x7 manning - 4 hour fix for service affecting faults - Customers get access to bandwidth graphs - Use leased and managed or own router - Will do BGP4 multihoming - Usual value-add services - Happy for customers to run real-time services over their network - Will do special services if it can be done with a reasonable SLA - Like with like pricing policy - Has DTI telco licence * VBCnet - 8 UK POPs (including Telehouse London and Telecity Manchester) - Lines to West Coast USA and Manhattan - Peering at LINX, Manchester Network Access Point ("MaNAP"), MAE-West (US), the Commercial Internet Exchange ("CIX"), and D-GIX (Stockholm) - Transit from NACnet and Abovenet in the US - Key equipment duplicated and distributed across POPs - Some BSD UNIX servers used as routers. Moving towards standardising on Cisco equipment. - Aim to keep backbone network circuit loading under 65-70% - Has on-site equipment spares holding - 24x7 network monitoring, with telephone support for customer fault reporting - Will do BGP4 multihoming - Can do server hosting at most POPs - Usual value-add services - Support for real-time services - Very flexible on special/non-standard services * Easynet - London based (Whitfield Street and Telehouse) - Peering at LINX (no US NAPs yet) - STM-1 to New York. Connectivity to Easynet France (Paris). - Transit from PSInet US, GTEI and Teleglobe in New York. Also Ebone for European transit. - Key equipment duplicated - Some BSD UNIX servers used as routers. Moving towards standardising on Cisco equipment - Aim to keep backbone network circuit loading under 50% - Has on-site equipment spares holding with some 2-4 hour equipment support contracts - 24x7 monitoring and telephone support for fault reporting - Will do BGP4 multihoming - Can do pay per bit service - Can do server hosting - Usual value-add services - Multicast routes carried internally - Fairly flexible on special services - Has DTI telco licence * Netkonect - London based (Telehouse) - Peering at LINX and MAE-East (US) - Transit from UUNET US (via INSnet US) - International circuits run over Gemini transatlantic cable system (diverse routing/SDH automatic protection and restoration) - Combination of Cisco routers, Extreme ethernet switches, with Livingstone and US Robotics equipment used for dial-up - Customers can use their own access-circuits to connect - 24x7 telephone support for fault reporting - Will do BGP4 multihoming - Can do server hosting - Usual value-add services - Economical pricing * Level3 - London based (including Telehouse) - Peering at LINX and various US NAPs - Transit from CERFnet via Level3 US - SDH fibre rings used for resiliance in physical networks - Multiple diverse fibre routes used in POPs - All key routers and servers duplicated - All POPs have spares for all equipment - Customers can use their own access-circuits to connect - 24x7 monitoring and (full) support - "WECSS" system (planned for summer 1999) will allow customers to monitor their own services and trouble tickets - 2 hour Mean Time to Fix ("MTF") across all products - Will do BGP4 multihoming - Usual value-add services (primary DNS extra). - Has a specials request process. Will try to support services if possible, but prefer to have sound technical understanding of all supported products - Good SLAs - Has DTI telco licence Conclusions ----------- Procuring ISP and telecomms services in the UK is quite a complex process, particularly as the old distinctions between the two areas become increasingly blurred. It is likely to be of particular benefit to either try out the service(s) you wish to buy (INSnet provide such a facility) or ask to speak to satisfied reference customers. The market for ISP services in the UK in particular is highly competitive, and it is always worth trying to negotiate on price, particularly where your requirements fall well within the standard Service Level Agreement and you do not require all its provisions to be supported. Thanks to all those from the UK ISP industry for their contributions to my research for this article. Martin Cooper works for the University of Cambridge Computing Service in the Network Systems Group.