Each access node of an EMC Centera has a certain number of available
communication threads. The optimum performance level is reached when each EMC Centera access node processes approximately 20 simultaneous connections. You can exceed 20 connections for each access node. However, performance degradation typically results.
When an application communicates with an EMC Centera cluster, it uses one or more of these threads to pass data to and from the EMC Centera cluster. With DiskXtender, these threads are used to read data from and write data to the EMC Centera cluster.
It is important to understand the communication process between DiskXtender and EMC Centera so that you do not exceed the maximum number of available
communication threads. The following topics provide additional details and guidance on configuring options that enable you to control the lines of communication and maximize performance.
How DiskXtender uses the communication threads
By default, DiskXtender can either read data from or write data to each piece of virtual media during any given time period. In other words, if DiskXtender is writing File G to a piece of media and a user requests File A from that same piece of media, then the user must wait until File G is written before File A is retrieved and
displayed. The writing of File G and the reading of File A both would require a single thread, each at different times, as illustrated in Figure 6 on page 50.
Figure 4 Single thread per piece of virtual media
In an active environment where file migration is necessary at the same time that file fetches are likely to occur, you can configure DiskXtender to both read from and write to a single piece of media at the same time. This functionality is available through the Enable simultaneous moves and fetches feature, which appears on the Options tab of the Service Properties dialog box. When simultaneous moves and fetches are enabled, two EMC Centera threads may be occupied by DiskXtender communication with a single piece of media—one for moves and one for fetches, as illustrated in Figure 7 on page 51.
GEN-000583
Step 2: After File G is migrated, then File A, which resides on Virtual Media 1, can be fetched Step 1: File G is migrated to Virtual Media 1 on the EMC Centera cluster
Virtual Media 1
Virtual Media 2
Virtual Media 1
Virtual Media 2 File G
Extended drive
File A Extended drive
EMC Centera cluster
EMC Centera cluster
Figure 5 Simultaneous moves and fetches
With simultaneous moves and fetches enabled, DiskXtender will use only two threads for a piece of media if it is both moving and fetching files. It will not open two threads to a piece of media for moving files, nor will it open two threads for fetching files.
If there are multiple pieces of virtual media in a media group—and even multiple media groups with multiple pieces of media in each—then a single DiskXtender installation can occupy numerous communication threads with an EMC Centera cluster at any one time. DiskXtender can initiate and maintain as many as 256 simultaneous connections to EMC Centera, regardless of the number of configured EMC Centera media services.
DiskXtender will not reserve a thread to a piece of media if it does not need to.
Communication threads to EMC Centera are initiated and maintained only if a file is requested from media or if there are files that need to be migrated and the file migration schedule is active.
GEN-000585
Virtual Media 1
Virtual Media 2
Virtual Media 3
Virtual Media 4
EMC Centera cluster File A
File G
File G is migrated to Virtual Media 1 at the same time that File A is fetched from Virtual Media 1
Extended drive
DiskXtender writes files randomly to any piece of active media in a media group.
This enables DiskXtender to continue writing files to media at the same time that it is fulfilling fetch requests for files on other pieces of media, as illustrated in Figure 8 on page 52.
Figure 6 Random migration to EMC Centera
You can configure DiskXtender to write files to media sequentially rather than randomly by using the Media fill method feature on the Options tab for each media group. However, sequential fill is strongly discouraged for EMC Centera media groups because it can lead to poor performance. “Media group considerations for EMC Centera” on page 71 provides additional information.
Calculating the optimal number of available threads
The equation for determining the optimal number of threads available for DiskXtender file migration and file fetches is as follows:
(N x 20) - A - R = Available threads where:
◆ N is the number of access nodes in the EMC Centera cluster.
◆ 20 is the optimal number of threads per node.
◆ A is the number of threads used by other applications, including other DiskXtender installations.
◆ R is the number of threads used for EMC Centera replication, if replication is enabled.
For example, if you are planning to write to an EMC Centera cluster with four access nodes, the cluster is dedicated to the DiskXtender installation, and replication is enabled and uses two threads, then the optimal number of threads available for file activity is 78, or:
(4 access nodes x 20) - 0 - 2 = 78
The optimal number of available threads for DiskXtender can then be used when determining the number of pieces of active media that should be maintained in each media group. By controlling the number of active media, you are effectively
controlling the number of threads that can be used for file migration.
DiskXtender can automatically create EMC Centera media to maintain a specified count of active media in a media group. Active media are media that are still available for file writes. For EMC Centera media, this means that the media has not yet reached the maximum of 256 GB or 100,000 files.
For example, if the DiskXtender installation has two extended drives, each extended drive has two media folders, and each media folder has one EMC Centera media group, then there are a total of four media groups targeting the EMC Centera cluster:
2 drives x 2 media folders x 1 media group (each) = 4 media groups If there are 78 available threads for the installation and you anticipate that file migration and fetch activity will be evenly distributed among the media groups, then you can divide the number of threads by the number of media groups:
78 available threads / 4 media groups = 19.5 pieces of active media Since it is not possible to have a half of a piece of media, you may want to round up to 20 pieces of active media for each media group.
Note: In an active environment where simultaneous moves and fetches are enabled and are likely to occur, divide the number of active media by two. Remember that with simultaneous moves and fetches, two threads are used by each piece of media.
Fine-tuning the equation to optimize performance
The equation for determining the optimal number of available threads—and
ultimately the number of pieces of active media for each media group—does not take into consideration the volume of file migration and file fetch activity to and from the EMC Centera cluster at different times of the day. As a result, you may need to adjust the DiskXtender or EMC Centera configuration to remain within the recommended number of available threads per access node and to maximize system performance.
Consider the following tasks to fine-tune the usage of available threads:
◆ Increase the number of access nodes in the EMC Centera cluster to increase the number of available threads. When more threads are available, determine whether to increase the number of active media in each media group:
• If file migration is dominating the communication threads, then increase the number of active media in each media group to take advantage of the additional threads for file migration.
• If file fetches from full media (not active media) are dominating the communication threads, then you may want to leave the same number of active media in each media group so that the additional threads can be used for file fetches.
◆ Schedule file migration so that it does not occupy the communication threads when they are needed for file fetches. By default, the file migration schedule is active at night, so that files are migrated when users are typically not fetching as many files. You may need to customize this schedule for your environment.
◆ Reduce the number of fetch requests from media so that they do not occupy the communication threads when they are needed for file migration. To reduce the number of requests for files on the EMC Centera cluster, leave as much active data on the extended drive as possible. Use DiskXtender purge rules to purge only file data that is no longer likely to be requested. Chapter 12, “Purging Files,”
provides details on configuring file purging.