You can apply write pacing at the group level, or at the device level for individual RDF
device pairs that have TimeFinder/Snap or TimeFinder/Clone sessions off the R2
device.
Group-level pacing
SRDF/A group-level pacing paces host writes to match the SRDF/A session’s link
transfer rate. When host I/O rates spike, or slowdowns make transmit or apply cycle
times longer, group-level pacing extends the host write I/O response time to match
slower SRDF/A service rates.
When DSE is activated for an SRDF/A session, host-issued write I/Os are paced so
their rate does not exceed the rate at which DSE can offload the SRDF/A session’s
cycle data to the DSE Storage Resource Pool.
Group-level pacing behavior varies depending on whether the maximum pacing delay
is specified or not specified:
l
If the maximum write pacing delay is not specified, SRDF adds up to 50
milliseconds to the host write I/O response time to match the speed of either the
SRDF links or the apply operation on the R2 side, whichever is slower.
l
If the maximum write pacing delay is specified, SRDF adds up to the user-specified
maximum write pacing delay to keep the SRDF/A session running.
Group-level pacing balances the incoming host I/O rates with the SRDF link
bandwidth and throughput capabilities when:
l
The host I/O rate exceeds the SRDF link throughput.
l
Some SRDF links that belong to the SRDF/A group are lost.
l
Reduced throughput on the SRDF links.
l
The write-pending level on an R2 device in an active SRDF/A session reaches the
device write-pending limit.
l
The apply cycle time on the R2 side is longer than 30 seconds and the R1 capture
cycle time (or in MSC, the capture cycle target).
Group-level pacing can be activated by configurations or activities that result in slow
R2 operations, such as:
l
Slow R2 physical drives resulting in longer apply cycle times.
l
FAST operations that lead to imbalanced SRDF/A operations.
l
Director sparing operations that slow restore operations.
l
I/O to the R2 array that slows restore operations.
On arrays running Enginuity 5876, if the space in the DSE pool runs low, DSE drops
and group-level SRDF/A write pacing falls back to pacing host writes to match the
SRDF/A session’s link transfer rate.
Device-level (TimeFinder) pacing
HYPERMAX OS
SRDF/A device-level write pacing is not supported or required for asynchronous R2
devices in TimeFinder or TimeFinder SnapVX sessions if either array in the
configuration is running HYPERMAX OS, including:
l
R1 HYPERMAX OS - R2 HYPERMAX OS
Remote replication solutions
134 Product Guide VMAX 100K, VMAX 200K, VMAX 400K with HYPERMAX OS
To Next Page
Loading...
