Spam Filtering For Small/Medium Business?

or_is_it writes "The company I work for has been growing dramatically and I've been charged with the task of being the gatekeeper for our GFI Spam filters. This involves manually inspecting the subject line/to/from for all caught messages in each filter rule folder. For a company of about 50 people, in one day the number of spam messages can exceed 2,000. Neglect it for a day and you end up with quite a task on your hands. I've made the rules lax enough so important messages can go through, along with a few stray spams, for which I get bitched at. Tighten the rules up and then maybe an important time-sensitive email never gets to its intended recipient, and I get bitched at. Manually reading through all those subject lines is supposed to prevent that, but I'm only human and genuine messages can easily get overlooked. How do larger organizations deal with the spam issue? I can't imagine having one centralized person manually inspecting everyone's junk-mail header is the optimal solution. Purchasing a different commercial mail filter product is a possibility, but I'd like to hear some anecdotal evidence before jumping ship."

Client-based?

To be honest, for somewhere of that size I'd be tempted to use some sort of client-based filtering (along the lines of spambayes [http://spambayes.sourceforge.net/]) which would put the power and responsibility in the hands of your users.

Combined effort is necessary

I have a setup where I use a configuration of Sendmail as first line protection and I use several sources for spam filtering.

dnsbl/enhdnsbl is enabled for zen.spamhaus.org, bl.spamcop.net, combined.njabl.org, list.dsbl.org, dnsbl-1.uceprotect.net, dnsbl-2.uceprotect.net, dnsbl-3.uceprotect.net and sbl-xbl.spamhaus.org. With all these enabled there are very few spam messages falling through.

Adding to this I am using Mozilla Thunderbird which has a very good intelligent junk mail filter. The only disadvantage is that the junk mail filter has to learn what's junk or not.

The use of dnsbl/enhdnsbl also does bounce back to the sender with a reasonable message for the cases where a message is denied so the sender shall be informed about any messages that are denied. Of course - it isn't fool-proof, but it works for me.

Re:Combined effort is necessary

The use of dnsbl/enhdnsbl also does bounce back to the sender with a reasonable message for the cases where a message is denied so the sender shall be informed about any messages that are denied. Of course - it isn't fool-proof, but it works for me.

Do you generate a bounce, or do you reject with a 500 error and a proper message at spam time? You should not generate a bounce to remote mail. Ever. This is the cause of e-mail backscatter and is a significant problem. Always reject at SMTP time with a 500 error.

Re:Combined effort is necessary

Wow. You need to review your config!

From experience: you only need Spamhaus Zen and SpamCop for connection checking.
If you parse DATA before you accept it, you should incorporate URIBL.COM it's very good, and helps catch Yahoo and Gmail spam (which will get past Spamhaus and Spamcop all the time) because it scans bodies for naughty links

dsbl.org is REDUNDANT -- incorporated in Spamhaus Xen.
Spamhaus SBL-XBL -- incorporated in Spamhaus Xen.
NJABL.org is dead and a mirror of the CBL, I believe (-- incorporated in Spamhaus Xen also)

Never send bounce notices for spam. What notices leave your server are likely going to forged From: addresses....

Re:Client-based?

Pardon me, but I just don't see the "size". I personally (and professionally) receive well over 3'000 spam e-mails each and every day. I take about three to five minutes to run through them. For 6'000 in two days, I take four to seven minutes.

I do it without a spam filter of any kind. I have only two technique.

First, simple rule-based filters throw clients and friends into their own folders by from: line alone. That covers everyone I know in advance.

The second set of rules simply looks for my full name, my company name, my e-mail signature, my telephone number, or my mailing address. These into the "it's damn likely a legitimate e-mail" folder. This folder gets about 2 spam e-mails per week.

The remaining I simply run through, in outlook express of all clients. Sorting wins the day. The greatest trick? Sort by the to: field. It doesn't take long to see that 75 messages went to [e-mail], [e-mail], or some other horribly malformed address to that doesn't exist. Sorting by subject does similar things -- like give you "70% off . . ." which get selected and deleted in a block of one hundred at a time.

Your spam has very simple patterns to look for. Sort by them, click the first, shift-click the last, and hit delete.

Last year, I was contracted by Viagra's H.R. department to do some quick work, I made it through unscathed.

Re:Client-based?

I've seen Postini-filtered mailboxes. Don't bother.

Only solution that I know works is my own: Postfix with amavisd-new, spamassassin, clamav, postgrey, along with FuzzyOCR on smaller installs, though setting that up on a separate system to filter through might cover a large organization. Don't forget to include things like Spamhaus' Zen list, any of the *.countries.dk.net blocklists to filter out any geographical areas from which you don't expect legitimate mail, and also helo filtering--if the connecting mail server can't say helo/ehlo with something that resolves in DNS, it can just bugger right off.

Tell your boss that expecting not to lose email with spam filters in place is unreasonable, and that tasking one human to eyeball all the rejects is a serious misapplication of time and money.

Best of all, you should educate your boss to realize that email is not a reliable messaging system. There are far too many points of failure that could cause a message to be lost, most of them being outside of your own or your company's control. There exist many better ways to send time-sensitive material, like fax, overnight mail, and telephone calls. If a severe amount of money is to be lost because an email didn't make it on time or made it not at all, then the message should have been sent over a more reliable medium in addition to being emailed.

Only the severely clueless would rely on a system like the one you have set up. You have to allow for a certain failure rate in any system. That's a basic principle of quality control methods that have been in use for decades.

Barracuda SPAM filter

I purchased a Barracuda for my organization of about 120 employees, and it has been fantastic. I fine tuned a few options on the config and it has blocked about 200,000 emails in the almost two months i have deployed it. There are very few false positives, and very few that get through its filters. I actually get calls of gratitude from the end users about how happy they were not receiving any more SPAM messages. The hardest part was informing them the user base on the difference between the mailing lists they were on and SPAM. Barracudas support has been good as well.

Re:

Ya, i rolled a baracuda out in a similar environment back in 04, and the users couldn't stop singing the praises compared to the filtering our mx offered + my manual filtering. I strongly recommend baracuda for this size roll-out.

Re:Barracuda SPAM filter

Gah they are so expensive. And to keep them up to date is ridiculously expensive. I prefer free with ASSP.

Additionally I have a serious problem with the backscatter they cause. They should reject mail at SMTP time and not bounce them.

But Barracuda support is very very good. Very responsive and timely and overall a good people orgaization which can make the difference for wanting to deal with them.

Re:Barracuda SPAM filter

There are multiple, very serious problems with Barracuda appliances. I've already commented on their propensity to generate backscatter elsewhere in this thread. They're also poorly supported, have systemic security issues, may have privacy implications (since Barracuda personnel have unauditable access to your mail stream), are expensive, use community resources such as DNSBLs in ways contrary to those resources' policies, and do not use current best practices in spam control. (This last is unsurprising given that Barracuda personnel do not participate in the discussions and consensus-building which generates those BCPs.)

Consider as well that the Barracuda appliances consist of (a) an open-source operating system (b) an open-source MTA (c) an open-source web server (d) an open-source spam scanner (e) an open-source virus scanner (f) other pieces of open-source software and (g) use community-mintained DNSBLs and RHSBLs. This is all held together with proprietary (closed-source) code, mostly for the purpose of providing a poorly-designed GUI interface. Any competent email system administrator should be able to create their own near-equivalent in an afternoon; it's not difficult. Such homebrewed creations have repeatedly been shown to vastly outperform Barracudas on multiple metrics, including cost, scalability, customization, security, and perhaps most importantly -- adaptability to new spammer techniques. (Barracuda is years behind the times and falling further back.)

It's very tempting to "just buy an appliance" and consider the problem solved, but it doesn't work. There's no substitute for expertise -- and given that much of that expertise is available for free, for the asking, on lists such as spam-l and spamtools and so on, it's difficult to understand why anyone would choose not to avail themselves of it.

email != IM

> maybe an important time-sensitive email never gets to its intended recipient

When will users learn...
Email is not instant messaging - with bad greylisting / random connection reset / busy server, you can get >=2 hours delay. And it's normal.

Re:email != IM

Your assessment of the current state of email is correct. But, blaming users for using it to fill a need when there is no realistic alternative is silly.

email is ubiquitous and easy. 99.5% of the time, it's nearly instantaneous. Should I really have to get an IM account on google, yahoo, aim, microsoft, etc.... so I can deal with time-critical messages? And, for that matter, should everybody else?

Re:email != IM

Businesses shouldn't be using those for internal communications anyway. Set up a jabber or irc server internally for that.

Force keywords in the subject line

I've had to send emails to recipients within the Australian Defence Force (specifically, the Army), and every email sent from a civilian must include a keyword within the subject line. The keyword is to do with whether or not the information is classified or unclassified. Sure, getting all the clients to send all their emails with [companyname] in the subject line is a little annoying, and may not be possible depending on your circumstances, but the chances of spam having that keyword within it is virtually impossible.

Set up an automated filter whereby anything that doesn't have the keyword in the subject gets dumped into a spam box to be sorted later. If the senders do the right thing, it assures their emails will be directed to the correct person.

This is just one example of active spam filtering as opposed to the passive spam filtering used in IT today.

Power to the people :)

Whatever solution you get, the simple answer is:

1) Set up the system to put junk mails in a folder the user can see

2) Train the end user to check their junk mails

3) Show the user how to set the spam triggers high or low and what the implications are

If user says they're too busy/important, advise them that due to your workload, their email box will be added to the "manually checked list" which gets done once per week. Point out the impact of losing a time-critical email wrongly flagged.

Most times they do it themselves. For those who are dead set on having someone else do it, hire a temp or arrange for an office junior to do it.

If you're in IT, you have better & more important things to do than check for real mail in a junk mail box...

Nothing's perfect...

As you've found, an automated system can be tuned, but you'll always have false positives/negatives.

I like the way spamassassin [apache.org] works - it can provide a rating for each message, which provides a mechanism for users to set the bar to their own preference, instead of having a single setting for the entire organization.

I'm not talking about using individual configurations for spamassassin, it's not realistic to expect most users to be able to deal with all the gory detail of spam filters.

Rather, spamassassin can set a header to indicate its confidence that a message is spam:

X-Spam-Level: ****

It adds an asterisk for each "point" of spam score. Users should be able to create an email filter which picks off suspected spam and puts it into a separate folder based on a header like that. Maybe drop all 10+ messages centrally, and let users tweak a local filter to their liking, depending on whether they prefer false positives or negatives.

I use spamassassin as an example only because that's what I use. There are no doubt others which can provide something similar which users could filter on.

Postini

Postini's anti-spam service does wonders. We use it for about 200 accounts and people love it. It works, rarely gets things wrong and is simple. IT (me) loves it because spam is no longer my problem. For a fee that would be less than my effort and aggravation is worth, they take care of it. We are currently investigating expanding use to compliance filtering and archiving as well.

For the record, Google purchased Postini in the not to distant past.

Frontbridge Spamshark

How do larger organizations deal with the spam issue?

I used to work for a mining company you've heard of. Our department had responsibility for managing the email vendor, who used Spamshark to filter spam coming into the organisation. From my limited knowledge of the setup, Spamshark does basic blacklisting etc. but also does selective blacklisting on specific IPs when an email is flagged by a user. So Alice flags a message as spam, Spamshark figures out the message id, grabs the IP address it came from (it knows because it previously handled the email), and then blacklists that IP for a certain amount of time. Now this internal blacklist is then shared to all the other customers who use Spamshark, so they are now protected too; resulting in a 5 nines hit rate on spam.

Like I said we just handled vendor relations, and the above description might not be totally accurate, but this is what I gathered when we dealt with them. I also remember getting about 10 complaints of spam a month for an organisation with 10's of thousands of email addresses - so it was very effective.

OpenBSD spamd

I've had excellent results with this particular product. Spamd uses blacklisting, greylisting, and tarpitting. It really is delightfully evil and still makes me smile because it includes a fake smtp daemon which sets the tcp rcv window to 1. This is a kick in the nuts to the spammer. I've used it with resounding success at a client who was recieving 2000 spam emails a day. Prior to implementing spamd, we were using just a Barracuda. When I combined spamd and the Barracuda, spamd caught about 1975 of the spam messages and the barracuda took over from there. No false positives and we've been running for three months. This link details how to set it up, http://www.linux.com/feature/61103 [linux.com].

ESVA all day long

I've been running this for quite some time with fantastic results. It's a VMWare appliance.

Inside, there is greylisting and MailScanner. Within MailScanner, there is SpamAssassin, some RBL, ClamAV and all sorts of things.

For my organization, I find that in addition to everything else "stock" I can safely filter out all countries but the U.S. since we don't do business outside of our state, let alone our country... so it's safe to assume that anything from outside the US will be spam.

It is extremely effective. I have helped to get the VM set up in environments with multiple domains and it works very well too.

One problem with it is that it is rapidly aging. The user community has made some effort to get the VM up to date in some ways, but the 2.0 version as far as anyone can tell is still in discussion and planning. The project creator and leader is a one-man-show and he seems to have a life outside of this project for some reason. The user community is frantic to get something to replace the aging 1.7.1.5 machine we all use as the reference point for our installs.

This is largely a known-solved problem

The place to ask this question isn't here, it's on the "spam-l" mailing list, which arguably has the highest concentration of the world's most experienced anti-spam researchers and developers. Simple techniques for tackling this have been repeatedly covered there over a period of many years, and their behavior is well-understood and predictable, making them viable choices for production systems. So I would suggest that you subscribe to that list (via [e-mail]) and repeat your question there, along with some indication of your MTA environment.

Meanwhile, here is some general guidance. First, do not waste your money on commercial products -- they're expensive, poorly-maintained, and in many cases (e.g. Barracuda) actually make the spam problem worse via backscatter. (There are now several thousand Barracudas on a communally-maintained blacklist, making it obvious to everyone working in this field that Barracuda is completely incompetent.) Second, do invest your money and time in open-source solutions: it is easy for anyone who possesses baseline competence in mail to craft their own, superior spam handling system using postfix or sendmail or another open-source MTA, DNSBLs, RHSBLs, judicious configuration, and other tools such as rbldnsd, mimedefang, SpamAssassin, ClamAV, and so on. Third, a little googling will reveal near-cookbook procedures for combining these pieces of software together into a useful system; which cookbook procedure is appropriate for you depends on your environment -- which brings me to the fourth point, which is that you need to perform log analysis in order to understand your particular mix of spam/not-spam. Everyone's is different, which is why one-size-fits-all solutions usually fail. Only after you have some clue about the size and shape of your problem will you be able to determine which approach(es) are likely to minimize both false negatives (FN) and false positives (FP).

As an aside, one set of highly effective anti-spam tactics involves enforcing RFC requirements that have been in place for many years: for example, all mail servers must have rDNS; that rDNS must resolve to a host which in turn resolves back to the IP; the domain of the host must exist; the host must HELO as a valid FQDN or bracketed-quad IP; the envelope-sender's domain must exist; the host must not HELO as you; the host must wait for the SMTP greeting before HELO'ing; the host must handle a multi-line SMTP greeting; the MX records for the host must point to valid IP space; and so on. Enforcement of these requirements yields differing rates of spam control (which is again why log analysis is crucial) but has the very valuable property that it can be done at low computational and bandwidth cost. Substantial experience with these suggests that enabling them and augmenting them with a few DNSBLs (especially the Spamhaus Zen zone) is enough to deal with the overwhelming majority of the spam problem at most sites, reducing what's left to a much smaller issue to be dealt with.

that's actually a good solution

I can't imagine having one centralized person manually inspecting everyone's junk-mail header is the optimal solution

Actually, that strikes me as a good solution; it's certainly better than having other employees dealing with spam as part of their daily routine and losing 30 minutes/day for everybody in the company. And by centralizing it, you have the ability to pick the tools to make your work more efficient, as opposed to having 50 employees each fiddle with their own spam filters.

3 Steps

This is just a simple guide compiled from my experience:

1. Do what you can on the server. I like to use SpamAssassin to add spam scores to beginning of subject lines, so they sort by score in my inbox (I use "/*_SCORE(0)_*/"). I also automatically delete anything over a score of 11, since the highest I've ever seen a legitimate email score has been "10.something". Realistically, anything above an 8 is the sender's fault and they need to do something about it and anything above an 11 you can safely blame the sender (you won't be the only spam filter deleting their emails).

2. Provide the tools on the client. ThunderBird's "spam marker" is a must, and because it learns from what you mark, you aren't just marking them in vain. Also, to deal with spam in real-time, instead of using the junk folder, I like using the "delete junk!" button from the "Buttons!" add-on. Incoming junk gets marked and marked as read, and after marking the spam the filter missed, I hit "delete junk". Very easy and quick. Pre-configure Thunderbird for everyone.

3. Educate and support. If you have 1 and 2 in place, then make sure everyone knows what you are doing and why you chose to do it. Write a short manual or something. Educate them about their tools. They also need to know NOT to publish their addresses.

The idea is to make spam highly visible, and to make it *quick and easy* to deal with. Knowing you've facilitated these two goals should be enough to impress your employer and earn the respect you deserve from everyone you serve :)

I spent a few days migrating 100,000 emails from Windows Mail, because it was horrible. Thunderbird is a godsend and the add-ons make all the difference. If there is something you dislike or want, chances are someone made an add-on for it.

btw 2000 messages is *not* a lot of spam. It will get far worse with time.

Re:Despite other issues

Why do people keep suggesting gmail as a viable option?

It's really not that good.

Re:SpamAssassin

I cast my vote for SpamAssassin.

When set-up with good rules and RBLs it blocks at least 99% spam with very low false positives (I've never had a false positive).

Send anything tagged as spam to another account such as spam@domain (I do this) then you can manually check for false positives to further reduce the chance of losing legit email. (or if a user complains that an email they expected never arrived)