Standard Deployment Scenarios¶
General Deployment Considerations¶
It is important to appreciate the use-case or use-cases for which you wish to deploy Kolab, and map that to the appropriate deployment scenario.
After all, Kolab Groupware is Made To Measure, and supremely flexible. It is best deployed after you articulate what you seek to get out of it – especially for businesses.
Larger numbers of users make it more important to ensure service availability.
Larger numbers of users make load-balancing (and high-availability through load-balancing) more attractive, more efficient and more cost-effective.
Contrary to popular belief, larger numbers of users make capacity planning more unpredictable, up to next to near impossible. Here, you require the ability to scale up and down as turns out to be needed, and not be locked into a certain scale with limited options to break out.
Should running an extra couple of virtual machine be relatively cheap, then you are likely looking for Multiple Servers for Each Service.
The larger the (potential) data footprint, the more important it becomes to consider your storage options.
The more flexibility you require in scaling;
- the more likely you run idle over-capacity, or
- under-performing services because of under-capacity, but
- the easier it is to answer either of the aforementioned problems.
- the more you need means to adapt quickly, preferably automated, by which we mean infrastructure services such as provisioning and configuration management services.
A Couple of Users¶
In terms of capacity, for a family server, SOHO situation or micro- entity, it is probably more than sufficient to run A Single Server Kolab Deployment.
If the data is really important, and you do not have data redundancy built into the single server (through RAID, or regular backups that are easy to restore), such an environment might want to opt for a Kolab Deployment on Redundant Servers.
Dozens of Users¶
When a deployment seeks to serve groupware to dozens of users, A Single Server Kolab Deployment suffices unless the average user’s usage pattern is extremely high.
A Free Software ISV with employees working remotely, from home, for example, very much relies on electronic communications. Operating in the Free Software community, it’s communication patterns are ever increasing, with its employees subscribing to upstream communities’ mailing lists, and the organization itself possibly providing services to its community. These would be considered keyboard-bound users.
While servicing only a couple of dozen users, the number of messages exchanged easily exceeds thousands per day.
For such a deployment, you may consider a deployment scenario as depicted in Multiple Servers with Multiple Services Each.
In contrast, if your dozens of users are actually plumbers, carpenters, firemen, airline pilots or otherwise regularly distracted from the keyboard and monitor, A Single Server Kolab Deployment may suffice – although your users may in this case be synchronizing their mobile devices, significantly increasing the usage pattern outside of their presence at the keyboard.
Scaling up from a single server deployment to a deployment with multiple servers is relatively straight-forward, but, depending on what service you choose to migrate off of the single host, possibly involves data migration, some down-time, and configuration changes. As such, such migrations require sufficient preparation and planning.
Suffice it to say that also, should you find the deployment type you choose initially underperforms, user acceptance of the solution may very well be negatively impacted – sometimes beyond repair.
Somewhere other than here, document the process of scaling up from one single server on to multiple servers.
Hundreds of Users¶
Providing Kolab Groupware to hundreds of users is an environment of some scale. The starting point is likely Multiple Servers with Multiple Services Each, however;
- You may already have centralized authentication and authorization,
- You probably already have an existing infrastructure, possibly including a perimeter network.
To illustrate why this is important:
The Kolab web interfaces do not require a dedicated web server – if you have one already, then you may want to consider installing the Kolab web interfaces on that, if not simply for the fact that public IP space is limited.
Other considerations come into play deploying Kolab Groupware, and those are included in the following sections:
Redundancy – with regards to data – is a matter to be considered separate from high-availability.
There are separate, distinct replication levels and scenarios one can consider, and storage devices types to be taken in to account.
Generally, the largest volume of data is in IMAP spools. These are also very I/O intensive – much more so than CPU- or memory-intensive.
Redundancy in storage has it’s own deployment guide section.
Nothing overloads a helpdesk more than hundreds of users calling in at very much the same point in time, because a service is unavailable (and might, as a domino effect, render other services unavailable).
The larger your userbase, the more important it is to ensure services remain available – even during planned service windows.
A typical approach is to provide “two of each”.
To use the term load-balancing is to describe the act of providing enough instances of each service to supply enough capacity to deal with the demand on said service. This is separate from the time-window of such demand, however – the time-window demands scaling, scaling demands load-balancing.
Load-balacing is where Kolab Groupware shines, since any of the service components can be split up in such many roles as well. Therefore, each quantitively meaningful difference in demand for a given service in a particular role can be scaled up and down as is needed.
With a quota of 1GB, a total data footprint of 100GB - 900GB is still manageable, but should your users (be allowed to) have larger mailboxes and/or use the File Storage features in Kolab, you are more likely speaking to the tune of several terabytes (if not right from the start, you’ll likely get there over time).
This is yet another area were scalability comes into play. One could start with a single Cyrus IMAP server, like so:
You will want to make sure your users’ desktop applications, and the
rest of the Kolab software uses a DNS entry to connect to IMAP (for
imap.example.org), so that it is easier for you to change
what it is they actually end up connecting to.
With several terabytes of data, when you get there, the desired scenario might look like:
This is a simple change that can be prepared ahead of time, and implemented during a service window, if and/or when it is needed.
You have options with regards to the target topology of the Cyrus IMAP Murder. Please refer to Cyrus IMAP Murder Topologies.
A Thousand Users¶
The magical boundary of a thousand users depicts each individual user’s usage pattern becomes unpredictable, as for one the number of mobile devices they synchronize are not necessarily under control any longer.
Generally, the same facets apply as they do for Hundreds of Users, just with higher load, more storage, more stringent requirements, likely resulting in the need for Multiple Servers for Each Service, at least partly – some services may still be combined.
Despite a number of users that is likely larger than the majority of Kolab installations, it is still well within the boundaries of normal operations, and should not require any specialist attention.
Yet, environments of this size and over will want to perform a Proof- of-Concept environment to familiarize themselves with the inner workings of Kolab Groupware, assess the viability of Kolab particulars for large numbers of users stuck with particular work-flows and their system administrator’s ability to effectively maintain the infrastructure with this new technology (monitoring, alerting, trending, configuration management, reporting, etc.).
Several Thousands of Users¶
The larger the enterprise (or: the larger the number of users), the more significant capacity planning becomes in relation to deploying Kolab.
We have mentioned before that provided a larger number of users, capacity planning becomes a more volatile subject and can be less accurately determined. However, with larger numbers of users, laws of averages come in to play, and the ability to scale with demand can be facilitated – a margin of error becomes manageable.
Tens of Thousands of Users¶
The differences between “tens” and “hundreds” of thousands of users are negligible with regards to the general deployment scenario.
A clear distinction is often having spread various areas of responsibily for the infrastructure across multiple teams or departments, each of them eligible to resist change making their jobs more involved, and/or not all of them as familiar or comfortable with the introduction of new technology in their respective stacks.
Hundreds of Thousands of Users¶
When an enterprise with 350.000 employees plans for a Kolab deployment, it is unlikely all of the users will be migrated over the course of a single service window, and despite what other groupware vendors might tell you, it is near to impossible to accurately plan for the capacity required.
It is also important to appreciate any existing infrastructure and network topology, and for Kolab to integrate into that environment.
Ranging from small and medium-sized business, large enterprise and service provider deployments, the number of users involved ranges from 50 to anywhere in the hundreds of thousands or millions.
Organizations with Multiple Domain Namespaces¶
When the people throughout an organization use different email domains, but need to maintain the ability to share groupware data with everyone else in the organization, it is important to appreciate the effects of using the primary recipient email address (mail) of the user as the authorization ID (result_attribute).
This might be the case in a holding with multiple subsidiaries. To illustrate, an example holding corporation Holding, Inc., with subsidiaries Foo, Inc. and Bar, Inc.:
This is not considered a case for ISPs or providers of Hosted Kolab, but may be applicable to customers of ISPs and consumers of Hosted Kolab. ISPs and Hosted Kolab providers should refer to dedicated sections:
Let’s suppose that:
- People working for Holding, Inc. use email addresses
- People working for Foo, Inc. use email addresses
- People working for Bar, Inc. use email addresses
All people are contained within a single LDAP tree (for, perhaps, Human Resources is a department within the holding, providing services to the holding as well as its subsidiaries, like it is providing Kolab Groupware services in this example):
In a default Kolab Groupware installation, the following mailboxes might be created:
Because Cyrus IMAP uses the email domain as an authorization realm, and no cross-realm authorization is allowed, in this scenario, John, Joe and Jane cannot share mailboxes - though John and Jim can.
This may be the desired effect, and if it is, you can skip reading the rest of this topic.
If it is not the intended effect however, and you seek to allow all people to share groupware data with all other users, you must consider the following:
Should all people be given a primary recipient email address of
@holding.inc, in an attempt to make all users end up in the same authorization realm, then they are implicitly allowed to send using that email address.
This in itself may not be desirable.
To enable users to share groupware data while their primary recipient
email addresses make them end up in different authorization realms, you
should set the
result_attribute setting in kolab.conf(5) to the
name of an attribute that does not contain a realm identifier (i.e.
something without an ‘@’ in it), such as the
uid attribute, which by
default does not include a domain name space. This would create the
following mailboxes (if the surname is used for the uid attribute):
You may also consider setting
imapd.conf(5), although this implies only the null realm is ever
going to be used.
For larger deployments, we also recommend reading about Cyrus IMAP Murder Topologies.