Mylo, I pretty much agree with Gil but I don't think most people or orgs have the slightest idea how to evaluate their environments for risks. Plus too many people have the mindset that if they don't know of a way to hack something, no way exists. If this is the direction taken, bring someone else in to do it. Even if you do that it still may not work out well though because of assumptions that are made during the analysis that don't end up being true in implementation. "Oh yeah, of course we look at the logs.... " "Of course we patch right away and watch the security bulletins...". The fewer vectors available to compromise tends to mean the less chance of being compromised. I think max paranoa is the safer path.
IIS on a DC makes me very queasy. Granted it is based on the history of IIS and it is "all fixed" now, but consider... How many exploits do you need against your DCs before it is considered too many? Is a single compromise acceptable? I don't mind losing most one off servers, it hurts but I can survive. If someone walked through a hole on a DC or a cert server your base security for the entire environment, all servers and clients, has been compromised and you can not easily have much faith in those pieces any longer. I can rebuild an IIS server in a couple of hours, how fast can you rebuild from scratch your domain structure? Your Cert structure? Exchange... Well I have all sorts of love for Exchange but right off, if Exchange is running on a GC, you have no fault tolerance or load balancing for directory work, that is the one and only GC that will ever be used. The Exchange provider should be complaining about that all alone. Failover to another GC in another site may suck, but at least it is possible. If someone insists that they can only have one server at a site, at this time my recommendation is that it not be a DC. If you keep your GPOs in check this shouldn't be a serious issue unless you have a crappy network. DCs are a special case and should be treated specially, it isn't just some extra service on a machine. Services I will run on a DC are things like WINS and DNS and quite honestly I don't much like DNS on DCs either. It bothers me to run a service on the machine that the DC is completely dependent on. With WINS, I always deployed LMHOSTS files on the DCs, that way if WINS failed, things still worked. joe -----Original Message----- From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] On Behalf Of Gil Kirkpatrick Sent: Thursday, October 06, 2005 7:07 PM To: [email protected] Subject: RE: [ActiveDir] Server Roles As you mentioned, this topic has been debated frequently on this list. Running other services on a DC raises the hackles on the back of my neck, and I expect that most on the list will have similar reactions. And you've listed most of the reasons why the proposed deployment would be a bad idea. But truthfully, the "right" answer has to be based on a proper risk assessment for your client's environment. I think in the past most people either a) never did a risk assessment, or b) didn't understand the risks with branch office DCs running multiple services. Consequently, most AD professionals now default to "its pure insanity" when asked about this kind of deployment. The answer of course, as with most everything, is "it depends". Because every organization has different perceptions of and sensitivities to different kinds of threats (some organizations have a high tolerance for service failure, but a low tolerance for trade-secret theft, for instance), and because the threat profile is different for each organization (how protected are the remote DCs? How accessible is the network? How effective is patch deployment?) the only way to evaluate the proposed deployment is to do a proper risk analysis in the context of the organizational environment. So if I were faced with this situation, I would recommend a threat assessment and risk analysis project to evaluate the risks associated with this sort of deployment. A good paper is Butler and Fishbeck's Multi-Attribute Risk Assessment http://www.cs.cmu.edu/~Compose/paper_abstracts/butler-fishbeck-02.html, but your favorite CISSP text covers it as well. Because you understand the threats and risks in the proposed deployment, you can make sure that they are properly represented in the analysis, and the customer can weigh the (definite) costs of additional servers against the (potential) costs of a security failure. That all being said, I think that running Exchange, SMS, or IIS on a DC is a Really Bad Idea (tm). My $.25... -gil -----Original Message----- From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] On Behalf Of Mylo Sent: Thursday, October 06, 2005 11:44 AM To: [email protected] Subject: [ActiveDir] Server Roles Hi All, It's a well trodden path (in these forums anyway) that I'm about to discuss but I'd like to get our resident experts 10 cents worth on a rather interesting issue I've run into.. I'm working at a client, reviewing an AD design, where 2 support providers are providing a migration path to an AD2003/Exchange 2003 solution (from NT4/Ex5.5). One of the providers is responsible for AD (desktop/SMS/File and Print) design and the other E-Mail design/deployment. This is a single forest/single domain solution where both have agreed to work in concert, together in the spirit of harmony and SLA's... There's a possibility that proxy tools may be used (e.g. Aelita/Quest type tooling) to 'limit' or delegate AD activities for each party, with these interfaces largely limited to managing AD delegation of OU/user/group/machine objects ... resource management (AV/Backup/SMS/DHCP/DNS/WINS etc) still requires native or 3rd party tooling. The problem lies in the fact that the client (on the advice of the support provider) has opted for consolidating File and print / SMS/ AD roles onto a single server at sites of up to around 200 users. Above this size the solution scales out to multiple servers, but continues to adhere to the principal of dual role, namely placing File and Print together with domain controllers and/or SMS and IIS together with a domain controller. In the legacy solution these roles were separated onto different serves and the file and print locally managed (also meaning that there's an awful lot of crap that will be migrated into AD as a result of combining these roles into one box) ... The combined role approach was given the green light largely for (I believe) cost reasons, but I do have *ahem* a number of concerns with this approach. Security ===== - multiple roles on a single server and no-no's such as placing IIS and SMS on a DC - it tends to look at security from a 'top down' perspective (i.e. it's a single AD provider therefore we're safe)... i don't think this flies simply because of the implications of using 3rd party s/w such as anti-virus and backup on dual-role servers where local admin rights are required, which equates to domain admin rights; providing a rather scary escalation path to being able to doing anything to anybody in the domain. Scenarios where the AD provider outsources to another party (e.g. in smaller countries)....if A (the client) trusts B (the support provider) who trusts C (outsourcee), should A trust C? ... I knew trusts would come in handy one day :-) Stability ===== - Print Services on domain controllers - Migrating clutter off the legacy file and print into AD (10,000's local/global groups) - If there's a mail server on-site with a combined server then e-Mail availability is linked to the whim and stability of file and print services/IIS/SMS etc. - Backup/Restore .. increased chance of human error where day-to-day restore operations associated with File and Print may result in key files being overwritten (relating to DC operations) Availability ======= - Reboots during the day are likely to be more numerous through bulking up roles... affecting the whole office (e.g. AD replication gets stuck, BITS kills IIS etc.) Accountability ========= - Difficult to prove anything was done by anybody at any time. Performance ========= - Means enabling write caching on a DC for the benefit of file and print services (i.e. read-optimised RAID versus write-optimised RAID) Possible solutions ============ 1. Use VS2005 and virtual machines 2. Place File and Print alone on smaller sites with no DC, say up to 25 users and above that use separate DC and File and Print/SMS roles on separate servers . 3. Buy SBS for each smaller site and setup x number of trusts to the central sites :0) 4. Live with it and stop worrying Am I being overly paranoid with this dual/triple role thing or is this really as bad as it looks ? Does anyone actually advocate this as a solution if they were given a greenfields choice? I'd appreciate your candour and feedback... Thanks, Mylo List info : http://www.activedir.org/List.aspx List FAQ : http://www.activedir.org/ListFAQ.aspx List archive: http://www.mail-archive.com/activedir%40mail.activedir.org/ List info : http://www.activedir.org/List.aspx List FAQ : http://www.activedir.org/ListFAQ.aspx List archive: http://www.mail-archive.com/activedir%40mail.activedir.org/ List info : http://www.activedir.org/List.aspx List FAQ : http://www.activedir.org/ListFAQ.aspx List archive: http://www.mail-archive.com/activedir%40mail.activedir.org/
