CUIT Hosting Service

Certain services, especially those in emerging technologies, are not provided as central services by CUIT but are required by a given department. In these cases, CUIT takes into account the centralized nature of the service (i.e. whether it will provide service across the University or by one department) and provides partial support by hosting the service on dedicated hosts bought by that department.

How It Works

First the client needs to write an informal proposal, stating the reasons they cannot use currently provided central services, and the centralized nature of the proposed service. This proposal will be considered by CUIT Management. They will take into account the centralized nature of the service, the ability of CUIT to support it (given our current level of knowledge of the technologies involved, and staff availability), and similar factors.

If the proposal is accepted, appropriate hardware will be determined by CUIT and the client, based on hardware currently supported by CUIT, the expected load on the server (or servers) and what level of redundancy is appropriate. CUIT will then buy the hardware (or allocate from existing inventory) and charge the client for the cost. In rare cases the client may buy the hardware and hand it over to CUIT.

CUIT staff will then install and administer the host much as they do their central servers, giving access as needed to the client. In almost all cases, root access will not be given, but appropriate access using file protections and sudo will be made available. CUIT staff will work with vendors to install specialized software, and will provide basic tuning of the system (e.g. determining if additional memory is needed).

Service Levels

CUIT provides many production services. Some of them are vital and need to be up at all times. Some of them are less sensitive to the occasional outage. Depending on the nature of each service, we expend more or less resources towards keeping it up at all times. These resources include things like redundant hosts and RAID arrays, high-availability Network File Servers, backups, monitoring, and staff time including off-hours intervention.

In order to reduce the amount of after-hours work needed, and make it possible for CUIT to properly support our servers, any service that must be up even during off-hours must be replicated across two or more servers. If the service does not merit buying two servers, CUIT staff will be unable to guarantee off-hours uptime.

In some cases, with a service in continual development, we recommend that two more hosts be purchased for development and testing. This allows new versions of the service to be developed and tested without interrupting production service. These hosts may also be transfered to production when demand for the service grows unexpectedly.

Keeping servers up-to-date with needed patches often requires rebooting the servers. To keep all servers patched in a timely fashion, CUIT generally schedules these reboots during evening and early morning hours, according to when the service has reduced usage. CUIT staff will work with the client to schedule the reboots, and will generally announce the downtime 24-48 hours in advance. In most cases, replicated services will experience no downtime. Non-replicated services will be down during the reboot, but this generally takes less than 15 minutes. Given the non-critical nature of non-replicated services, CUIT staff are unlikely to schedule reboots outside an extended daytime window of M-F 8am-7pm.

CUIT Systems staff is not expert at everything, especially those things which are not provided on our central servers. We will ensure that a server is up, and that the basic services are working. For special software installed on behalf of a client, we will make a best effort attempt to get it working. In most cases, this involves basic analysis via tools like ps, restarting daemons via /etc/init. scripts, and rebooting the host. We have found that these tools are fairly effective at getting an existing service working again. The most notable exception is some form of database corruption, where repairs have to be done before the service will work again. Actions in cases like this will generally involve waiting until working hours, when staff are better able to consult peers and vendors concerning the possible solutions.

Monitoring and Problem Reports

All servers run by CUIT are monitored by SyMon, an automated system which checks basic server health as well as specialized services. CUIT staff will work with the client to determine automated ways of checking the Hosted Service. They will also work to determine a monitoring schedule and notification procedures. Notification will generally involve contacting the client through an appropriate mailing list or alias, and as appropriate paging of Systems group staff.

When the client or their users determine that a service is unavailable, there are procedures for notifying the Systems group of the problem. In urgent cases, the client should call the CUIT Computing Support Center at 212-854-1919 and report the problem. They should indicate that the service is Hosted by CUIT and that they have been working with the UNIX Systems group. They should make sure to leave as much contact information as possible (email and phone) to enable the Systems group to contact them for more information.

In less urgent cases, and as followup to Support Center calls, the client should send email to unix@columbia.edu. If they have been working with one or two particular members of the group, they may choose to send mail to those members, but should make sure to CC unix@columbia.edu.

These procedures (calling the Support Center and mailing unix@columbia.edu) enable the Systems group to respond to the problem in the most efficient fashion, without requiring the client to send multiple emails and make numerous phone calls trying to reach someone who can help.

Additional Charges

CUIT will purchase the hosts with a three year warranty, so there will be no maintenance charges. If additional hardware is needed to get or keep the service running (e.g. extra memory or disks needed, original hardware undersized) the client will be asked to pay for it. In some cases, the client will be asked to pay annually for bulk disk space on our Network Applicances filers.

Renewal and Upgrades

Over time, the purchased hardware will become older and the service may become more or less used. This will result in the need to replace the hardware with newer or different capacity hardware. Hardware generally needs to be replaced every three years; more often if demand for the service has increased. When this takes place, CUIT will reconsider whether the service is something they should centrally support. If this is the case, CUIT will provide the replacement service. If not, CUIT will again work with the client to determine the appropriate hardware, and the client will be asked to purchase the hardware as before. Unless special requirements prevent it, the old hardware will be kept by CUIT to use for other services or for its trade-in value.

How to Request This Service

To request this service, please contact Melissa Metz, Manager of Unix and Email Systems, melissa@columbia.edu, 854-3538.

« Back to Specialized Computer Services Home