Anti-Spam Vendor Analysis November 2003 MiaVia, Inc. Not for Distribution Anti-Spam Vendor Analysis Table of Contents The Players 1 Anti-Spam Solution Deployment Methods 4 Customer Acquisition 5 Partnerships 6 Spam Identification Methods 7 Filtering Accuracy Comparisons 9 Message Processing Speed 12 Other Anti-Spam Solution Features 13 Pricing 14 Appendix 1: Solution Deployment Methods of Anti-Spam Vendors 15 Appendix 2: List of Customers Mentioned by AS Vendor 17 Appendix 3: Spam Identification Methods by Vendor 18 Appendix 4: Spam Filtering Accuracy Test Results 20 Anti-Spam Vendor Analysis, November 2003 The Players Existing anti-spam solution providers include software giants such as Microsoft (embedding anti-spam features in their Exchange server products) and major anti-virus firms who have added new and richer anti-spam functions to complement their anti-virus offerings. Leading anti-spam specialty firms include Brightmail, Postini and MessageLabs. There are nine well-financed newcomers who have received venture capital financing in 2003. Beyond these twenty or so players, there are at least several dozen lower-tier companies, including antispam specialty firms and other software providers who have added some form of anti-spam solution to their product line. These include a few companies that have built proprietary offerings around the SpamAssassin, the highly prevalent open-source solution. Email server companies, such as Sendmail, Mirapoint, Rockliffe, Stalker and others have partnered with various anti-spam vendors and have embedded those offerings as part of their email server suite. Some of the more unconventional anti-spam approaches include trusted/bonded email sender programs, such as those offered by Habeas and Ironport, which provide information to filters rather then serving as filtering systems. Software Giants IBM Corporation (Lotus/Domino) Microsoft Network Associates, Inc. Sophos (ActiveState) Symantec Corporation Trend Micro, Inc. The major email server and anti-virus vendors include some form of anti-spam functions in their products or as add-on products. Generally speaking, the biggest of these offers the weakest solutions. Symantec owns a significant interest in Brightmail and therefore may have less incentive than other powerhouses to invest directly in its own anti-spam technology. Trend Micro has licensed Postini’s engine, while Network Associates and Sophos have both acquired third party solutions (Deersoft and ActiveState, respectively). While Microsoft will provide support for third-party solutions, it can be expected to compete vigorously in the anti-spam arena and is reputed to have filed as many as 40 patents in the field. Network Associates, having acquired the SpamAssassin development team through its purchase of Deersoft, has avowed that it will continue investing in the space, through internal means and possibly further acquisitions. Major Anti-Spam Incumbents Brightmail, Inc. CipherTrust Clearswift MessageLabs Postini Corporation SurfControl plc Tumbleweed Communications Vircom, Inc. Zix Corp. The clear leaders in terms of market presence among the anti-spam specialists include Brightmail, Postini and MessageLabs. Each has reported impressive sales growth and Postini recently raised $10 million is series D venture financing. MiaVia, Inc. Not for Distribution 1 Anti-Spam Vendor Analysis, November 2003 Well-Financed Newcomers The spam problem has attracted numerous new challengers to the incumbent players. Eight new antispam companies have received a combined total of nearly $50 million in investment in 2003, on top of the $10 million raised by Postini this year alone. Anti-Spam Venture Funding Vendor 2003 Funding Prior Funding Investors $4.5 million $1.1 million angel round (2001) Ignitition Partners IPO Israel Seed Partners Newcomers Cloudmark Series A 7/2003 Commtouch, Inc (CTCH) $1.6 million private placement Argos Capital Management A private individual investor Corvigo $ 5.5 million Sequoia Capital Series A FrontBridge Technologies $8.0 million $6.5 million Series B 12/2002 Series C 8/2003 MailFrontier, Inc. Bank of America Venture Partners Sierra Ventures Waitte Ventures $10.0 million $5 million Series B Menlo Ventures New Enterprise Associates (8/2003) MessageGate $5 million Boeing Ventures Series A Polaris Ventures 8/2003 Northwest Venture Associates Newbury Ventures MX Logic Inc. $ 5.5 million Adams Street Partners Series A Grayhawk Venture Partners 10/2003 Proofpoint Inc. Vista Ventures $ 9.0 million Series B $7 million RRE Ventures Series A Benchmark Capital 7/2003 Mohr, Davidow Ventures inVentures Group Stanford University Total, Newcomers $49.1 million $19.6 million Incumbent Financing Postini $10 million Bessemer Venture Partners Series D August Capital 9/2003 Mobius Venture Capital Sun Microsystems Summit Accelerator Partners AltoTech Ventures Pacifica Fund Brightmail NA $35 million Accel Partners Crosslink Capital Technology Crossover Ventures Symantec MiaVia, Inc. Not for Distribution 2 Anti-Spam Vendor Analysis, November 2003 Other Anti-Spam Vendors There are probably many other anti-spam companies in addition to the firms listed below. A few of those listed have licensed an AS solution of one of the above vendors, as noted below. Some of the vendors are anti-spam specialists, while others offer other products. A few have assembled proprietary AS solutions built on SpamAssassin, indicated in the table below by asterisks. Vendor Products/Services Anti-Spam Specialists * Barracuda Networks AV/AS appliance; AS based on SpamAssassin Block All Spam Email hosting & AS service ePrivacy Group Traffic throttling appliance for AS Eridani Star System UK-based AS specialist Mail-Filters.com AS software and service, signature-based Mailshell OEM AS supplier Roaring Penguin Software Inc. Open source and commercial AS Sendio Appliance version of Mailblocks challenge/response Singlefin AV/AS/attachment/content (service & appliance) * SpamCop The Titan Key Client AS based on SpamAssassin Challenge response system Vendors Offering Multiple Products, Including Anti-Spam Aladdin Knowledge Systems Ltd. Public co. Content & software security products * Alien Camel Pty Ltd Anti-virus; anti-spam based on SpamAssassin * Barbedwire Technologies Network security; anti-spam based on SpamAssassin Bluecat Networks Appliances for managing DNS, AV, AS Cobion German Web content filtering & AS Computer Mail Services, Inc. AV/AS & email content security solutions Easylink Services Corporation AV/AS/content filtering; large line of other products * Fortress Systems Ltd. AV/AS software; AS based on SpamAssassin GFI Software Ltd. Security and messaging software Gordano Ltd Messaging software * Inova Brazilian sub. Int’l co. AS based on SpamAssassin IntelliReach Email management solutions Ipswitch, Inc. Messaging, network monitoring, file transfer software Lightspeed Systems, Inc. Network control, AS, porn filtering, file share control * Lyris Email marketing/AS; AS based on SpamAssassin NEMX AV/AS/secure content management for MS Exchange NetIQ Corporation Systems/security management, web analytics Net-Sieve AV/AS/content filtering Nokia, Inc. Diversified mobile communications conglomerate Solid Oak Software Client-based web content filtering Sunbelt Software Diverse line of Windows software tools Sybari Software Inc. AV/AS email security Webwasher AG Web content filtering, Mailshell AS OEM MiaVia, Inc. Not for Distribution 3 Anti-Spam Vendor Analysis, November 2003 Anti-Spam Solution Deployment Methods Anti-spam solutions began as client software applications but have evolved into server-based solutions offering greater control to system administrators and relieving some or all of the burden of spam management from end users. Three server-based approaches are employed: Server software deployed as an email gateway, relay, or MTA add-on at the customer site. An appliance, or server software provided on a pre-configured appliance device, which then may be inserted into the customer’s network as an email gateway or relay in front of one or more mail servers A hosted service that accepts the customers email flow, performs filtering operations at the vendor’s site, then forwards mail to customers according to its spam and virus evaluations and according to customer preferences. Although a few vendors offer more than one of these options, most restrict themselves to one method of delivering the solution. About 70% of the vendors offer server software, 40 percent offer a hosted service and the remaining 30% offer an appliance. This relationship is true for the roughly five dozen companies included in this report and also for a selection of companies considered among the top twenty due to their market presence or funding: Solution Deployment Methods of Major AS Vendors Server Software Appliance Service Brightmail Proofpoint CipherTrust FrontBridge Clearswift Sophos (ActiveState) Corvigo MessageLabs Cloudmark SurfControl Network Associates MX Logic Commtouch Symantec Corporation Tumbleweed Postini Corporation MailFrontier Trend Micro MessageGate Tumbleweed Network Associates Vircom MiaVia, Inc. Not for Distribution 4 Anti-Spam Vendor Analysis, November 2003 Customer Acquisition According to an April 2003 report by Ferris Research 1 global business anti-spam seat deployment will increase from 11 million in mid-2003 to over 500 million seats in 2008 with revenues from anti-spam services topping $1 billion. Corporate anti-spam services will total about $55 million in 2003 and over $850 million in 2008. At the same time, the global ISP anti-spam seat deployment will increase from about 175 million in mid-2003 to nearly 1.2 billion seats in 2008. In terms of revenues, this translates to $66 million in 2003 and over $200 million in 2008. Various industry analysts have estimated that between 50% and 75% of enterprises have adopted some form of anti-spam solution already. Judging from the types of customers listed by some of the anti-spam vendors in this report, there is no sector or industry that is immune from spam as the types of companies who have purchased AS solutions span all industry categories. Some anti-spam vendors appear to be achieving significant market penetration. Appendix 2 provides a list of customers mentioned by AS vendors on their web sites or in their press releases. Additionally, the following vendors claim notable customer acquisition achievements: Postini: "Over 1,700 customers" Sunbelt Software: “Over 1,000 installations in 4 months” …Lists over 700 customers on web site. SurfControl: has more than 20,000 customers worldwide, signed up another 1,400 new customers in the first-quarter - including the DTI and Norwich Union. Brightmail: announced an additional 200 new enterprise customer wins in the third quarter of 2003. MessageLabs: increased FY 2003 revenues nearly 100% over the previous year, with North American market revenues alone increasing nearly 300% and Asia Pacific revenues increasing more than 200%. The company expanded its client base to more than 7,000 businesses worldwide, signing new global and enterprise clients including Andrews & Kurth, Arnold Worldwide, The Bank of New York, Bertelsmann, BP, Diageo, EMI Music, ICI, Random House and StorageTek(TM). MX Logic: "more than 500 customers" 1 InternetNews.com, “Online Portals Won't Defeat Spam,” April 29, 2003, www.internetnews.com/xSP/article.php/2198791 MiaVia, Inc. Not for Distribution 5 Anti-Spam Vendor Analysis, November 2003 Partnerships Distribution and technology partnerships play a significant role in revenue generation for the more established anti-spam vendors. MessageLabs, for example, reported that 60% of its revenues are derived through resellers. As seen in the list below, partnerships with anti-spam vendors are flourishing. Undoubtedly much of the recent infusions of venture capital into the AS sector will be used to forge new partnerships with fledgling anti-spam vendors. Anti-Spam Vendor Partnerships Brightmail Frontbridge Postini Borderware Symantec Ironport CriticalPath Openwave Edox IBM Syntegra Oracle Sun Sprint Sophos Cable & Wireless Symantec AT&T Trend Micro Trend Micro McAfee Cloudmark ZoneLabs Sendmail Mail-Filters.com Sybari FilterLogix Elron Software MessageLabs BT Cable & Wireless Easynet GmbH MCI MiaVia, Inc. Not for Distribution Sophos McAfee Ositis Rockliffe SurfControl Microsoft Check Point Cisco IBM Nokia 6 Anti-Spam Vendor Analysis, November 2003 Spam Identification Methods While anti-spam solutions can be compared on a variety of features, including how spam messages are handled, what reporting and other administrative controls are provided, which operating systems are supported, etc., the foremost feature is generally considered to be the level of filtering accuracy. Accuracy is determined by the spam identification method. While this report will review spam accuracy measurements in a subsequent section, this section summarizes the spam identification methods employed by the various vendors. The information sources used to produce this review consisted primarily of public information, such as vendor web sites, white papers and published articles. The “Cocktail” Approach In general, most vendors rely on multiple spam identification techniques in order to compensate for the inability of any one technique to produce an adequate level of filtering by itself. For the 57 vendors profiled in this report, the average number of spam identification methods used per vendor was approximately six, and only five vendors employ fewer than four methods. The techniques most frequently combined include: Blacklists (51 vendors) Whitelists (46 vendors) Message header attributes indicating spamming behavior (53 vendors) Keyword or key-phrase matching (47 vendors) Statistical model, usually Bayesian, but not always (41 vendors) Some form of digital “fingerprinting” (32 vendors) Less frequently employed spam identification methods include: Comparing embedded links to a spam link library (7 vendors) Machine learning or artificial intelligence aimed at processing natural language, such as Corvigo’s "Intent-based filtering” based on natural language processing, Proofpoint’s “machine learning algorithms” and SurfControl’s “neural network” approach. Specific attached file types (Barracuda Networks, Symantec) Foreign language character sets (Sunbelt Software) Image file content (Aladdin Knowledge Systems) OCR text detection in image files (Cobion) Number of message recipients per message (Clearswift) Programming code detection (Cobion) Statistical analysis of header information (Commtouch) Number of user “spam” votes (MailFrontier) Challenge/response (Block All Spam, Easylink Services, Singlefin, The Titan Key) Seven of the profiled vendors incorporate or base their anti-spam product on SpamAssassin. These vendors include Alien Camel, Barbedwire Technologies, Barracuda Networks, Inova, Lyris, Roaring Penguin Software, SpamCop. The ways in which various spam identification methods are combined into an anti-spam cocktail are not usually clear from reviewing each vendor’s product information and are often kept secret. In some cases an overall score is derived by assigning weights to each different method while in others a series of tests is performed and any spam-signifying result may be taken as conclusive evidence of a message being spam. MiaVia, Inc. Not for Distribution 7 Anti-Spam Vendor Analysis, November 2003 Vendors Using Singular Spam Identification Methods A few vendors have rejected the “cocktail” approach, or at least claim to do so, and have built systems using one or two spam identification methods: Cloudmark uses only a statistical model of spam messages Corvigo uses a machine-learning approach which also can be thought of a as a model of spam based on natural language processing Mail-Filters.com uses a spam signature database in combination with header analysis. Net-Sieve uses Bayesian analysis in combination with heuristics. There do not appear to be any vendors among the group studied in this report that rely completely on a message body fingerprinting approach to spam identification, and none that subject each sample from which fingerprints are derived to a manual message inspection process. Topical Classification Most anti-spam vendors provide a binary classification system that causes each message to be classified as either “spam” or “not spam.” A few vendors attempt to provide a more granular approach so that users can customize the operation of the filter to differing tastes. Postini, for example, grades messages according to broad categories such as “Adult,” “Offers to good to be true,” and “Make Money Fast.” Since all of these categories exemplify content most recipients wish to avoid the categories are probably not terribly useful. Cobion also attempts to judge the categories of messages. Again, the categories are broad and tend to be of types universally disliked, such as pornography, hate, criminal, etc.) so there is probably little value to such a classification scheme. Further, the categories to which messages belong are determined from analyzing passages of text, which can fail whenever messages do not contain text passages (such as image-based messages). MiaVia, Inc. Not for Distribution 8 Anti-Spam Vendor Analysis, November 2003 Filtering Accuracy Comparisons With so many vendors to choose from it is understandably difficult for potential customers to evaluate the different vendors, their methodologies and their claims. Short of trying products that seem like a fit, customers must rely on third-party tests to help them sort out the good from the bad. Third party tests suggest significant accuracy differences among competing solutions and a point to a need for improvement in both identifying spam and avoiding false positive errors. Unfortunately it is quite difficult for third party testers to submit anti-spam solutions to truly rigorous and meaningful tests. The challenge arises from possible biases involving samples messages, how the tested products are configured and how accuracy is measured. The biggest challenge is probably getting a good message sample set: how can a filter tester acquire enough messages that represent both spam and non-spam to simulate actual operating conditions? If too few messages are used as samples and if they are not sufficiently representative then filter accuracy comparisons may be quite biased. With these limitations in mind, this report reviews three anti-spam comparison tests that have been performed during 2003: two tests conducted by PC Magazine and one test for Network World Fusion by the consulting firm Opus One. In each case the results are somewhat suspect, particularly with respect to the false positive rates, because relatively small sets of non-spam messages were used (fewer than 10,000 messages per test). The tests also used somewhat different methodologies and did not include the same products in each test. If the tests are considered indicative, the results indicate extremely wide variation in both positive error rates and spam identification rates. The results also suggest substantial room for improvement with regard to filtering accuracy. This finding may help explain why a recent Infoworld of 902 IT managers revealed that spam software was considered the over-hyped technology of the year.2 Test 1 PC Magazine, 25-Feb-03, “Corporate Antispam Tools…” http://www.pcmag.com/article2/0,4149,849390,00.asp Tested 2,500 per day over several days during February, 2003. false Test 2 Opus One (for Network World Fusion) Buyer's Guide: Anti-spam Test: Spam in the wild http://www.nwfusion.com/reviews/2003/0915spam.html Tested 7,840 spam messages and 3,648 non-spam messages during June, 2003 survey antimost Test 3 PC Magazine, Nov. 11, 2003 “Enterprise Spam Tools…” The highest spam detection rates across http://www.pcmag.com/article2/0,4149,1357946,00.asp all three tests were produced by Tested "4 days worth of messages from a catch-all account" MessageLabs (96%) and Postini (94%), however the false positive rates associated with these results ranged from 4 to 5 messages per thousand, and Postini’s false positive rate was measured at 13-14 messages per thousand in the other two tests. Most of the notable anti-spam vendors’ anti-spam tools were included in at least one of the three tests, but five of the companies that received VC backing in 2003 were not included in any of the tests (Commtouch, Corvigo, MessageGate, MX Logic and Proofpoint). Neither McAfee nor Symantec were included. Only Postini appeared in all three tests, while Brightmail and Frontbridge appeared in both PC Magazine tests. The table below presents a summary of the test results, while details from all three tests are included in Appendix 4. 2 InfoWorld, July 25, 2003 MiaVia, Inc. Not for Distribution 9 Anti-Spam Vendor Analysis, November 2003 Summary of Anti-Spam Accuracy Test Results* Vendor Spam % Comment 2.9% 80.5% Test 2 0% 88.9% Test 3; improved greatly over Test 1 Ciphertrust 0.24% 94.1% Test 1 Clearswift 2.3% 48.5% Test 2 Cloudmark 1.3% 82.0% Test 2 23.4% 83.4% Test 2; over-reliance on keywords 0.7% 77.9% Test 2 20.5% 23.1% Test 2 0.3% 62.1% Test 3; test excluded use of blacklists 56.3% 3.6% Test 2 iHateSpam 2.9% 74.4% Test 3 MailFrontier 0.5% 93.0% Test 3; improvement over Test 2 0.48% 96.0% Test 1 MX Logic 0.5% 77.0% Test 2 Postini 1.4% 84.9% Test 3; Test 2: 0.4% f+ , 94% spam % Singlefin 2.9% 86.2% Test 2 SurfControl 3.3% 76.5% Test 2; worsened from Test 1 Trend Micro 0.8% 60.3% Test 3; improved over Test 2 Tumbleweed 1.2% 81.3% Test 2 ActiveState (Sophos) Brightmail Computer Mail Services Corvigo EasyLink Frontbridge GFI MessageLabs False + % *The most recent test data are shown if a product appeared in more than one test. In the most recent test Postini’s results declined and Frontbridge’s spam detection rate fell. Full results for all three tests are in Appendix 4. From the above data it is clear that some systems performed much better than others in terms of filtering accuracy. The poorest results were noted among companies that are not anti-spam specialists, suggesting a comparative advantage of specialization. Since there are two dimensions to filter accuracy measurements, it is helpful to arrange the data graphically into an x-y chart that shows how each vendor’s solution performed relative to others along axes representing false positive error rates and false negative error rates. The chart below provides this view, after first eliminating the worst performers, since the worst performers’ results are so bad they cause the better performer’s results to be too tightly clustered on a chart to easily discern their differences. Vendor results were excluded from the chart if their false negative rate exceeded 20% or their false positive rate exceeded 2%. The chart also includes test data from all three tests if the results met these thresholds, so in some cases more than one data point is plotted for a single vendor, indicating different results for different tests. MiaVia, Inc. Not for Distribution 10 Anti-Spam Vendor Analysis, November 2003 Of the solutions that performed within the 20% and 2% error thresholds, substantial accuracy differences are still apparent. Brightmail’s zero false positive error result is admirable, but came at a cost of misidentifying 11% of the spam messages. CipherTrust, Postini and MessageLabs appeared to offer the best overall compromise between false positives and false negatives, yet the tradeoff still requires a level of false positives that may be too high for many users (24 – 48 messages per thousand). It should be noted that variability across the different tests could cause significant biases, which may explain the substantially different results observed for Postini among the three tests. The definition of “false positive” can be a significant factor influencing the test results. Some bulk email items can be identified as spam but might not be considered spam by the intended recipient. As a test of this phenomenon, the author of this report signed up to receive a variety of opt-in email messages using a Hotmail address. Hotmail’s email is filtered by Brightmail. The types of subscriptions included general news (such as ABC News, MSNBC News, various IT publications and academic journals). After receiving over 4,000 of these messages the result was that Brightmail marked approximately 25% of these opt-in bulk email messages as spam. This finding, while anecdotal, casts doubt on the accuracy of the false positive measurements reported in the PC Magazine and Network World Fusion tests. MiaVia, Inc. Not for Distribution 11 Anti-Spam Vendor Analysis, November 2003 Message Processing Speed The Network World Fusion test also included measurements of message processing speed. The results showed a significant range of speeds, which might be a decisive factor for some customers. The processing speed may be varied for software-based systems by augmenting the hardware on which it runs, while appliance-based systems are more rigid. Filtering services, such as Postini, are a different matter, with speed determined by the vendor’s infrastructure and general Internet latency. Anti-Spam Message Delivery Speed Trend Micro* 20 Cloudmark* 20 MailFrontier* 20 Tumbleweed 10 Corvigo 7.25 Clearswift 6.7 Postini 6 EasyLink 3.8 MX Logic 3.6 ActiveState 3.2 SurfControl 3 Vircom 2.6 GFI 2.4 Singlefin Computer Mail Services 1.25 0.5 Messages Per Second *The test setup was for 20 messages/second. Products marked with an asterisk theoretically could deliver at higher speeds. MiaVia, Inc. Not for Distribution 12 Anti-Spam Vendor Analysis, November 2003 Other Anti-Spam Solution Features The most important filtering features are undoubtedly filtering accuracy, speed and deployment method. However, a variety of secondary features may influence purchase decisions. These additional features are generally of an administrative nature or provide options for handling messages identified as spam. There also are differences among filtering software vendors as to which operating systems are supported and whether software products are directly integrated with other products, such as mail servers or antivirus systems. A detailed analysis of these aspects is beyond the scope of this report. The Network World Fusion buyers guide (http://www.nwfusion.com/bg/2003/spam/index.jsp) provides rich detail on these subjects. Some of the more distinctive message handling and administrative features that were found in compiling this report include the following: Customization of spam identification for groups and individual end users, enabling users to adjust the threshold used to identify spam. End user quarantine for Exchange: Administrators can choose to quarantine spam on their Exchange server in a spam folder that end users can view from Microsoft Outlook. The Exchange quarantine allows end users to access the quarantine without having to authenticate to a web site or learn a new interface. Emailed quarantine digests that enable end users to retrieve valid messages blocked by spam filters. The digests provide end users with convenient access to their quarantined messages through their email inbox, without administrative intervention or additional tools. Logs and reports on email statistics: system administrators can view logs or generate reports at any time from the desktop using a Web interface. Log file rollover for easy archiving and management of log files. Custom rules editors enable administrators to adapt filtering to handle a particular spam problem that is not addressed by the filter’s default settings or capabilities. Whitelisting important sender domains, for example, can provide a good measure of protection against false positive errors. Error-reporting plug-in for Outlook: With the click of a button, users can submit missed spam (or false positives) to the filter vendor for analysis. Administrators can receive a copy of all junk and false positive submissions, providing insight into how effective and accurate the solution is performing from an end user perspective. Graphical installation lets administrators choose which components to install, sets default settings and guides the administrator through the registration process. The installer also supports uninstallation and upgrading by installing a new version in place over an existing version. Graphical administration console enables administrators to configure logging levels, create and view reports, monitor performance statistics, start/stop services, provide automatic failover, and support remote administration. Fail-over provides the ability of multiple filter clients to fail over to multiple servers and continuance of mail flow in the event of complete filter failure. MiaVia, Inc. Not for Distribution 13 Anti-Spam Vendor Analysis, November 2003 Pricing Pricing observations obtained for this report came from figures supplied by vendors themselves to Network World Fusion for their vendor listings. There appears to be wide variation in pricing, although some general patterns are evident: Anti-spam server software tends to be priced on a per-user basis, between $10 - $30 per user per year, although some of the less prominent vendors offer flat fee software pricing ranging from several hundred dollars up to $8,000 per CPU. NetIQ offers a hybrid price structure with a flat $2,000 fee plus a relatively modest $7.50 per user per year. Pricing for anti-spam appliances also tends to be priced on a per-user basis, either directly or by selling units capable of handling different sized mail volumes and pricing them accordingly. Prices quoted ranged as low as $2,500 and as high as $36,000, with most clustered within a range of $10,000 - $20,000. Anti-spam services tend to be paired with anti-virus as well, and generally are priced in the range of $10 $30 per user per year. MiaVia, Inc. Not for Distribution 14 Anti-Spam Vendor Analysis, November 2003 Appendix 1 Solution Deployment Methods of Anti-Spam Vendors Vendor Aladdin Knowledge Systems Alien Camel Barbedwire Technologies Barracuda Networks Block All Spam Bluecat Networks Brightmail CipherTrust Clearswift Cloudmark Cobion Commtouch Computer Mail Services Corvigo Easylink Services ePrivacy Group Eridani Star System Fortress Systems FrontBridge GFI Software Gordano IBM (Lotus/Domino) Inova IntelliReach Ipswitch Lightspeed Systems Lyris Mail-Filters.com MailFrontier Mailshell MessageGate MessageLabs Microsoft MX Logic NEMX NetIQ Corporation Net-Sieve Network Associates Nokia Postini Corporation Proofpoint Roaring Penguin Software Sendio Singlefin Solid Oak Software Sophos (ActiveState) SpamCop Sunbelt Software SurfControl Sybari Software MiaVia, Inc. Software Appliance Service Not for Distribution 15 Anti-Spam Vendor Analysis, November 2003 Vendor Symantec Corporation The Titan Key Trend Micro Tumbleweed Vircom Webwasher AG Zix Corp. MiaVia, Inc. Software Appliance Service Not for Distribution 16 Anti-Spam Vendor Analysis, November 2003 Appendix 2 List of Customers Mentioned by AS Vendor Brightmail Accel Partners Bechtel Corporation Borland Booz-Allen Hamilton Cisco ConEdison Cypress Semiconductor Deutsche Bank Eastman Chemical eBay John Hancock Latham and Watkins Motorola Lawrence Berkeley Lawrence Livermore Terra Lycos U.S. Department of Housing and Urban Development Williams-Sonoma AT&T Worldnet Bell Canada BellSouth BT Openworld Cincinnati Bell Comcast Critical Path Demon UK (Thus plc) Earthlink Hotmail MSN ViaWest Verizon Cloudmark Johns Hopkins Univ. EMI Documentum Dolby Laboratories Fidelity Restoration Hardware University of Nebraska Nolte Kelly Paper Rainmaker Bertelsmann Frontbridge Alan Matkins Gray Cary Pinkerton Sprint Virgin Bausch & Lomb Kyocera Rand Sunkist Waterpik Brown-Forman Ogilvy Sovereign Bank MiaVia, Inc. Sybase Xilinx GFI Microsoft NASA European Central Bank US Navy Telstra BMW Siemens Volkswagen Fujitsu The Body Shop Triumph Adler Berliner Investment Bank Bayersdorff A.G. First National Bank and Trust British Midlands Peugeot Toyota Caterpillar PerotSystems MailFrontier Wyndham International Pier 1 Imports CDW Public Service of New Mexico Peet’s Coffee & Tea, CDW Viasystems Bellingham Public Schools MailShell Access US Digicon Communications NewSouth Communications PMBx Boston College George Washington University Hood College Lafayette College Ohio State University University of British Columbia University of California, San Francisco University of Groningen University of Wisconsin Federal Trade Commission Ontario Financial Alza Corporation Avigen Inc. AirNet Systems Arrow Trucks Exel Transportation MHF Incorporated Linear Technology Corporation Puffer-Sweiven Inc. Tower Engineering Not for Distribution American Red Cross Big Brothers/Big Sisters National Kidney Foundation Ronald McDonald House United Way MessageGate Tribune Company Postini Accredited Home Lenders AmSouth Bank BRE Properties Inc. Brodeur Worldwide Cable &Wireless West Indies Canon Development Americas, Inc. Capital Partners Management Circuit City Cooley Godward DPR Construction Farm Bureau Financial Services Fenwick & West LLP Foley & Lardner Hormel Foods Corporation Invesco - National Asset JL Audio Lord, Abbett & Co. Merrill Lynch MidMark Corporation Morrison & Foerster LLP Packeteer Peapod, Inc. Pullman & Comley Rand McNally SkyWest Airlines Sterling Capital Mortgage Sutherland Asbill & Brennan LLP Vertis Inc. Virchow Krause Sophos Canadian Space Agency Cingular Wireless CitiStreet City of Richmond Cornell University Davis Vision Inc. (Highmark Blue Cross Blue Shield) Duke University HP Indiana University Macrovision Pulitzer Inc. Stanford University State of Indiana University of California, Berkeley (CCS Department) University of Washington Vignette Corp. WestJet 17 Anti-Spam Vendor Analysis, November 2003 Appendix 3 Spam Identification Methods by Vendor Vendor Aladdin Knowledge Systems Alien Camel Barbedwire Technologies Barracuda Networks Block All Spam Bluecat Networks Brightmail CipherTrust Clearswift Cloudmark Cobion Commtouch Computer Mail Services Corvigo Easylink Services ePrivacy Group Eridani Star System Fortress Systems FrontBridge GFI Software Gordano IBM (Lotus/Domino) Inova IntelliReach Ipswitch Lightspeed Systems Lyris Mail-Filters.com MailFrontier Mailshell MessageGate MessageLabs Microsoft MiaVia, Inc. Blacklist Whitelist Header Keyword URL Bayesian Linguistic Fingerprint Heuristics Rate Limit Other Not for Distribution 18 Anti-Spam Vendor Analysis, November 2003 Vendor MX Logic NEMX NetIQ Corporation Net-Sieve Network Associates Nokia Postini Corporation Proofpoint Roaring Penguin Software Sendio Singlefin Solid Oak Software Sophos (ActiveState) SpamCop Sunbelt Software SurfControl Sybari Software Symantec Corporation The Titan Key Trend Micro Tumbleweed Vircom Webwasher AG Zix Corp. MiaVia, Inc. Blacklist Whitelist Header Keyword URL Bayesian Linguistic Fingerprint Heuristics Not for Distribution Other Rate Limit 19 Anti-Spam Vendor Analysis, November 2003 Appendix 4 Spam Filtering Accuracy Test Results Vendor ActiveState (probable folder only) ActiveState (probably and maybe folders) Brightmail Ciphertrust Clearswift Cloudmark (spam and potential spam folders) Cloudmark (spam folder only) Computer Mail Services Corvigo (junk and bulk folders) Corvigo (junk folder only) EasyLink Frontbridge GFI iHateSpam MailFrontier MessageLabs MX Logic Postini Singlefin SurfControl Trend Micro Tumbleweed (spam hi + spam folder) Tumbleweed (spam Hi folder only) ----------------Test 1-------------False + % False - % Spam % ----------------Test 2-------------False + % False - % 2.9 7.2 0.05 0.24 28.2 5.9 17.9 51.5 14.9 18.0 16.6 15.4 22.1 76.9 4.0 96.0 1.34 9.3 90.7 0.13 19.6 80.4 Spam % 0.0 11.1 88.9 0.3 37.9 62.1 2.9 0.5 25.6 7.0 74.4 93.0 1.4 15.1 84.9 0.3 26.9 73.1 48.5 85.1 82.0 83.4 84.6 77.9 23.1 82.1 0.48 ----------------Test 3-------------False + % False - % 80.5 89.4 71.8 94.1 2.3 1.6 1.3 23.4 16.5 0.7 20.5 0.07 19.5 10.6 Spam % 56.3 96.4 3.6 0.7 10.6 89.4 0.5 0.4 2.9 3.3 0.8 1.2 1.4 23.0 6.0 13.8 23.5 39.7 18.7 27.8 77.0 94.0 86.2 76.5 60.3 81.3 72.2 Test 1: PC Magazine, Feb. 25 2003, “Corporate Antispam Tools…” Test 2: Opus One, Sept. 15 2003 (for Network World Fusion), “Buyer's Guide: Anti-spam Test: Spam in the wild” Test 3: PC Magazine, Nov. 11 2003, “Enterprise Spam Tools…” MiaVia, Inc. Not for Distribution 20