Case study: HA cluster on W2K3 for Teleradiology Practice

[Antw1]

Because of the central role a PACS holds in the architecture of a DICOM network, designing a High Availability solution for PacsOne has been on our wish list for a long time.


We've been deploying PacsOne in various production environments for the past couple of years, and in nearly all cases, the PACS is installed as a central node between the modalities and review stations.
If the PACS becomes unavailable, modalities are unable to output studies to the network, and studies are unavailable for diagnostic viewing.

In one of our recent deployments, a radiology practice receives 20-40 MRI studies daily from 15 remote Imaging Centers over multiple internet VPN connections.
The PacsOne server is used to centrally collect remote studies where they are later reviewed online by 3-4 Radiologists using Neologica's RemotEye web DICOM viewer.

Considering the case load, it was evident that any delay in receiving or publishing studies would very quickly build-up.
Since reading these studies is the core business activity, it also appeared that the PACS server should absolutely be considered mission-critical.


A 2-nodes active-passive High Availibity cluster on Windows 2003 Enterprise Server seemed to be a better option over completely redundant systems (i.e. 2 identically configured hardware and software replicating over the network) because of the ability of setting up a shared storage and the centralized monitoring and administration of cluster resources.

In essence, active-passive clustering provides a hot-spare node, without requiring to equip each system with multiple large capacity hard-drives.

W2K3 HA clusters are installed on short list of hardware platforms and share a SCSI-atached storage subsystem.
Our system is build on 2x HP Proliant DL 360 and 1x HP StorageWorks MSA 500.
Because DICOM data is voluminous, a large part of the system cost is directly linked to the amount of hard-drive storage purchased; so being able to share a single storage device is very cost-effective.
Also, the amount of network traffic involved in duplicating DICOM data from one fully-redundant system to its partner can become substantial in busy environments.

In addition, W2K3 HA clusters offer a centralized administration console used to monitor the health of each node and trigger the failover of the cluster activity from one node to the other.
Although the failover is automated only for hardware failure or system shutdown, HA clustering provides a very easy way to switch the cluster network addressing from one node to the other and in the same operation stop/start defined services.


Using the Windows Cluster Management Console, PacsOne was defined as a cluster resource so it would only run on the active node.
This is necessary because of the possibilty for the PacsOne service running on the passive node to pick up routing jobs in the database, while not having access to the images held on shared storage device.

Both PacsOne installation use the same AE titles, port and IP address, the latter two being configured as managed cluster resources.
The web interface runs under IIS 6.0, and although it technically could be stopped and started along with other cluster resources depending on the node status, we didn't feel it was required since the cluster is only accessed through its cluster IP or DNS name for both HTTP and DICOM traffic.

This system has now been in production for 4 months and is showing its strength when performing Windows Updates installations.
If the update requires a node to shutdown, the other takes over automatically.


I didn't go into technical details pertaining to the cluster implementation, but feel free to ask questions by replying to this post.

-Antoine Renault, MCSA, North Salem Imaging

[pacsone]

Thanks very much, Antoine! This should be very helpful to our users who are also looking for adding HA features to PacsOne Server.

Once again, many thanks for sharing your experience with the PacsOne Server users' community and please keep them coming!

[dkumar]

Are you still deploying PACSONE for teleradiology groups? I am in the process of staring a new teleradiology group and would like to know more about your deployed solutions

[pacsone]

Please contact Antoine directly (click on the <email> button at the end of the 1st message) about this topic, as he is the original author of this topic.