The Performance software reports disk statistics By Cluster and By Node. Cluster statistics represent the total of all disk I/O for a cluster. By Node statistics represent each node's contribution to the cluster I/O load. The following example shows disk statistics by cluster:
Tabular Report CLUSTER Page 1 PA Vx.x Tuesday 26-JAN-2006 09:00 to 09:20 ************************ Final Statistics *********************** ************************ Data Analyzed: from 26-JAN-2006 09:00:00.00 to 26-JAN-2006 09:20:00.00 *********************** +------------------------ CLUSTER Disk Statistics -----------------------+ ! ! ! Resp ! ! Paging Swping Rate Time Queue Space ! ! % % (/s) (ms) Length Used % ! ! ------ ------ ------ ------ ------ ------ ! ! DSA0 0.0 0.0 0.0 0 0.0 96 ! ! DSA1 0.0 0.0 0.0 0 0.0 81 ! ! DSA10 0.0 0.0 1.9 38 0.1 61 ! ! DSA111 26.5 0.0 12.4 33 0.4 99 ! ! DSA12 0.0 0.0 0.0 35 0.0 33 ! ! DSA13 0.0 0.0 0.0 27 0.0 85 ! ! DSA14 100.0 0.0 0.0 31 0.0 95 ! ! . . . . . . . ! ! . . . . . . . ! ! . . . . . . . ! ! YQUEM$DFSC7104 0.0 0.0 0.0 0 0.0 ...... ! ! YQUEM$DFSC7105 0.0 0.0 0.0 0 0.0 ...... ! +------------------------------------------------------------------------+ User Command: ADVISE PERF REPORT TABULAR=(BYCLUSTER)/BEG=26-JAN-2006 09:00:00.00 - /END=26-JAN-2006 09:20:00.00/OUT=BYCLU.TAB/SECTION=DISK
The way the Performance software counts a node's contribution when calculating By Node and By Cluster disk statistics is based upon the node's relationship to the disk. There are two types of relationships a node may have to a disk; a disk may be either hosted or served by a node.
If a node is directly connected to a disk by a MASSBUS, UNIBUS or HSC (Hierarchical Storage Controller), the disk is hosted by the node. If a node is not directly connected to a disk and must go through an intermediary node that hosts the disk, the disk is served.
For the purposes of this discussion, the term, direct access, refers to the relationship where a node hosts a disk, and the term, remote access, refers to the relationship where a node serves a disk.
The reason a node's relationship is important when calculating disk statistics is because the I/O of a node with remote access to a disk is processed through a node with direct access to the disk. Therefore, rates for the node with remote access are included in disk statistics for the node with direct access and in the disk statistics for the remote node as well.
The calculation of Total Cluster Disk I/O rates is fairly simple; the disk statistics from the nodes with direct access to the disk are added together. Data from nodes with remote access to the disk is ignored, as this data is already accounted for by nodes with direct access to the disk.
The following table is a summary of how the software calculates By Cluster disk statistics:
|
Disk/Node Relationship |
Calculation of Total Cluster Disk Statistics |
|---|---|
|
One or more nodes with direct access and any number of nodes with remote access |
Add I/O data from node(s) with direct access to disk |
|
No nodes with direct access in the collection specification and one or more nodes with remote access |
Add I/O data from all nodes with remote access only |
Depending upon the number of nodes with direct and remote access to the disk, computing By Node I/O disk statistics is complex and sometimes not possible to calculate for all nodes. This is because it is not always possible to distinguish each node's contribution to the total Cluster I/O rate.
When there is only one node with direct access to a disk and any number of nodes with remote access, the By Node contribution of the node with direct access is calculated as follows: Subtract the I/O statistics of any nodes with remote access from the I/O statistics of the node with direct access. The By Node contribution of nodes with remote access to the disk is the I/O statistic for each remote node.
In the case of more than one node with direct access to a disk and more than one node with remote access, calculation of each node's contribution is not possible because there is no way to distinguish which node with direct access performed the I/O operations for which nodes with remote access.
The following table shows a summary of how the software calculates By Node disk statistics:
|
Disk/Node Relationship |
Calculation of By Node Disk Statistics |
|---|---|
|
One or more nodes with direct access; no nodes with remote access |
Take I/O data from node with direct access to disk. |
|
One node with direct access; one or more nodes with remote access |
Node(s) with remote access, take I/O data from the node with remote access. Node with direct access, subtract all remote access nodes' I/O statistics from the direct access node's I/O statistics. (Information is unavailable if all data is not present, that is, when interval times for all nodes do not align.) |
|
More than one node with direct access; more than one node with remote access |
Unable to report By Node data. |
The following Example Configuration diagram below is an example system configuration. There are two disks, DU1 and DB2, and three nodes, A, B and C. Disk DU1 is hosted by nodes A and B; that is, nodes A and B have direct access to DU1. Nodes A and B are served to disk DB2; that is, nodes A and B have remote access to disk DB2. Node C has remote access to disk DU1 and direct access to DB2.
Detailed examples of how the Performance software would calculate By Cluster and By Node disk statistics using the example configuration are given in the following By Cluster and By Node report format explanations.
|
Copyright © 2008 CA.
All rights reserved.
|
|