What are the Delay Centers from different performance monitoring tools?
Applies to
- ONTAP 9
- Active IQ Unified Manager (AIQUM)
- OnCommand Unified Manager (OCUM)
- AIQ PAS
Answer
- ONTAP relies on QoS to break down the latency and identify the components that are incontention, each Delay Center represents the delay in its corresponding component.
- So far, various performance monitoring tools have used different terms for each Delay Center, this KB article intends to clarify the differences and help correlate the various Delay Centers from different tools.
Delay Centers |
CLI |
AIQUM/OCUM |
AIQ PAS |
Perfstat |
Harvest |
CPU Delay in Protocol Layer |
Network |
Network Processing |
CPU N-blade |
CPU_network or CPU_protocol |
Network |
CPU Delay in Network Layer |
Network |
Network Processing |
CPU N-blade |
CPU_network |
Frontend |
Delay in Cluster Interconnect |
Cluster |
Cluster Interconnect |
Cluster Interconnect |
DELAY_CENTER_CLUSTER_INTERCONNECT |
Cluster |
External Delay |
Network |
Network |
Network |
DELAY_CENTER_NETWORK |
Network |
CPU Delay in WAFL Layer |
Data |
Data Processing |
CPU D-blade |
CPU_wafl_exempt |
Backend |
Disk Delay |
Disk |
Aggregate Operations |
Disk IO |
DELAY_CENTER_DISK_IO |
Disk |
CP Delay |
Data |
Data Processing |
WAFL Susp CP |
DELAY_CENTER_WAFL_SUSP_CP |
N/A |
Delay from other suspensions |
Data |
Data Processing |
WAFL Susp Other |
DELAY_CENTER_WAFL_SUSP_OTHER |
Suspend |
NVLOG TRANSFER Delay |
NVRAM |
Data Processing |
NVLOG_Transfer |
DELAY_CENTER_NVLOG_TRANSFER |
NVLog Mirroring |
QoS Max Delay |
QoS Max |
QoS Limit Max |
QoS Limit |
DELAY_CENTER_QOS_LIMIT |
Throttle |
QoS Min Delay |
QoS Min |
QoS Limit Min |
QoS Min |
DELAY_CENTER_QOS_MIN_THROUGHPUT |
N/A |
Cloud Delay |
Cloud |
Cloud Latency |
Cloud IO |
DELAY_CENTER_CLOUD_IO |
Cloud |
FlexCache Delay |
FlexCache |
N/A |
FlexCache RAL + FlexCache Spinhi |
DELAY_CENTER_FLEXCACHE_RAL + DELAY_CENTER_FLEXCACHE_SPINHI |
N/A |
Sync SnapMirror Delay |
SM Sync |
Sync SnapMirror |
Sync Repl |
DELAY_CENTER_SYNC_REPL |
N/A |
Volume Activation Delay |
VA |
Volume Activation |
COP |
DELAY_CENTER_COP |
N/A |
Notes:
- The command from CLI to break down the latency is:
Cluster::> qos statistics volume latency show
- The long delay from a particular Delay Center is a good indicator of that component being in contention, further performance analysis should be focused on that component.
- The ones in Bold are the Delay Centers where the delays from multiple components are combined together, in this case, multiple components need to be taken into consideration:
- Network & Data from CLI
- Network Processing & Data Processing from AIQUM
- CPU N-blade from AIQ PAS
- Delay accounting might be slightly different for different verticals, especially in the Network/Protocol related Delay Centers, refer to the table below.
- ASA makes a difference in Cluster Interconnect, refer to the table below
Delay Center | Protocol | Perfstat | CLI | AIQUM | AIQ PAS |
---|---|---|---|---|---|
CPU Delay in Protocol Layer | NAS | CPU_network | Network | Network Processing | CPU N-blade |
CPU Delay in Network Layer | NAS | CPU_network | Network | Network Processing | CPU N-blade |
Delay in Cluster Interconnect | NAS | DELAY_CENTER_CLUSTER_INTERCONNECT | Cluster | Cluster Interconnect | Cluster Interconnect |
External Delay (R2T, XFER_RDY) | NAS | N/A | N/A | N/A | N/A |
CPU Delay in Protocol Layer | iSCSI | CPU_protocol | Network | Network Processing | CPU N-blade |
CPU Delay in Network Layer | iSCSI | CPU_network | Network | Network Processing | CPU N-blade |
Delay in Cluster Interconnect | iSCSI | N/A | N/A | N/A | N/A |
iSCSI ASA | DELAY_CENTER_CLUSTER_INTERCONNECT | Cluster | Cluster Interconnect | Cluster Interconnect | |
External Delay (R2T, XFER_RDY) | iSCSI | DELAY_CENTER_NETWORK | Network | Network | Network |
CPU Delay in Protocol Layer | FCP | CPU_protocol | Network | Network Processing | CPU N-blade |
CPU Delay in Network Layer | FCP | N/A | N/A | N/A | N/A |
CPU Delay in Network Layer | FCP ASA | CPU_network | Network | Network Processing | CPU N-blade |
Delay in Cluster Interconnect | FCP | N/A | N/A | N/A | N/A |
FCP ASA | DELAY_CENTER_CLUSTER_INTERCONNECT | Cluster | Cluster Interconnect | Cluster Interconnect | |
External Delay (R2T, XFER_RDY) | FCP | DELAY_CENTER_NETWORK | Network | Network | Network |
Notes:
- Perfstat is the only exception where CPU_protocol and CPU_network are distinguished.
- In other tools, they are combined together in the Network CPU related Delay Centers, such as Network (CLI), Network Processing (AIQUM) and CPU N-blade (AIQ PAS).
- For NAS, all the protocol related overhead occurs in nwk_exmept, just as the network related overead. So it only has CPU_network.
- ASA is an exception for SAN.
- For non-ASA, SAN doesn't have indirect access when the primary path is utilized, so no delay from Cluster Interconnect.
- For ASA, indirect access is always expected, so delay from Cluster Interconnect should be expected.
- For ASA FCP, CPU Delay in Network layer should be expected as in the backend, the indirect access will utilize nwk_exempt CPU domain.
Additional Information
Delay Centers | Definition |
CPU Delay in Protocol Layer | Delay in handling the protocol requests, such as NFS, CIFS, FCP and iSCSI |
CPU Delay in Network Layer | Delay in network handling, such as converting a TCP frame to a protocol request, or vice versa |
Delay in Cluster Interconnect | Delay in communication between different nodes when there is indirect traffic |
External Delay | Delay outside of ONTAP, such as iSCSI Ready to Transfer (R2T) or FCP Transfer Ready (XFER_RDY) |
CPU Delay in WAFL Layer | Delay in WAFL CPU |
Disk Delay | Delay from fetching the required data from Disk layer |
CP Delay | Delay from WAFL Suspensions on CPs, usually B/b CPs |
Delay from other suspensions | Delay from WAFL Suspensions on the reasons other than CP related |
NVLOG TRANSFER Delay | Delay from nvlogging |
QoS Max Delay | Delay from QoS Limit |
QoS Min Delay | Delay from QoS Floor, the volumes with QoS Floor will not see this delay, but they might starve the other volumes without QoS Floor |
Cloud Delay | Delay between ONTAP and the cloud tier on which user data is stored |
FlexCache Delay | Delay between FlexCache volume and Origin volume when ONTAP has to lookup or retrieve the data from the Origin, or write the data to the Origin. |
Sync SnapMirror Delay | Delay from Sync SnapMirror replication |
Volume Activation Delay | Delay from WAFL schduling when a node has more than 1,000 FlexVols |
The links below have more detailed explanation on each Delay Centers: