Fix it now
Event ID 129 from Storport is “Reset to device was issued”; Event ID 153 is “The IO operation at logical block address was retried”. Neither is a failing disk. Both mean something between Windows and the media answered too slowly, and Microsoft’s published remedy starts with drivers, firmware and MPIO settings.
Set-MSDSMGlobalDefaultLoadBalancePolicy -Policy RR
Set-MPIOSetting -NotificationState Enabled
Set-MPIOSetting -NotificationPeriod 30
Set-MPIOSetting -NewPDORemovePeriod 20
Set-MPIOSetting -CustomPathRecovery Enabled
Set-MPIOSetting -NewPathRecoveryInterval 10
reg add HKLM\SYSTEM\CurrentControlSet\Services\mpio\Parameters /v PathVerifyEnabled /t REG_DWORD /d 1 /f
- Before you change anything, update all storage, HBA and multipath drivers and firmware. That is step one in Microsoft’s own remedy, and it is the step people skip.
- Read which port driver is resetting. The source of the 129 names the adapter, so you can tell a SAS controller from an iSCSI initiator from a virtual adapter.
- Compare path counts against a healthy node with
mpclaim -s -d, then read the current settings back withGet-MPIOSetting. - On a cluster, increase the disk resource timeouts so the node tolerates a slow failover instead of failing the resource.
- If the node is a virtual machine and event 153 keeps recurring, treat driver currency as the fix: Microsoft records recurring 153 in virtual environments as a Storport driver issue.
The values above are Microsoft’s baseline, not your array vendor’s. Where the vendor publishes different figures for your model, theirs win – but a server with no considered settings at all should start here.
If the events stop and latency stays inside the range your array vendor considers normal, you can stop here. The next section explains what a reset actually is and why replacing a disk usually does nothing.
Why it happens
Every I/O Windows issues carries a timer. When a request has been outstanding longer than the driver is prepared to wait, the port driver escalates rather than failing the request outright, and event 129 is that escalation reaching the adapter: a reset was issued to the named device, and the outstanding requests are re-issued behind it.
Event 153 is what the layer above sees. The disk driver notices an operation failed, retries it, and when the retry succeeds it records the logical block address and the disk involved. On a healthy system you see neither. On a system with a slow array you see a trickle of 153, then bursts of 129 whenever the array is under pressure, and eventually a volume dismounted while nobody was watching.
Microsoft lists 129, 153 and 140 together as evidence of storage or network bottlenecks, and files them alongside events 51, 55, 98, 1069, 1205, 5120, 5121 and 5124 as the disk, file system and cluster errors worth correlating. That grouping is the useful part: these events are read as a set, in one time window, not individually.
Drivers and firmware are out of step
You have this one if The events began after an update, or the node has never been updated while its peers were.
- Record the current HBA firmware, HBA driver, array firmware and switch firmware.
- Move to a combination the storage vendor supports for your Windows Server version, as a set rather than one component at a time.
- Apply outstanding Windows updates and hotfixes; Microsoft names this as a general step for every one of these events.
- Monitor for a full week afterwards, because these faults are load dependent and a quiet weekend proves nothing.
A path has failed and everything is queueing on what is left
You have this one if Resets started at a specific moment with no change in workload, and the path count on this node is lower than on its peers.
- Compare
mpclaim -s -dagainst a healthy node and count paths per LUN. - Turn on path verification:
reg add HKLM\SYSTEM\CurrentControlSet\Services\mpio\Parameters /v PathVerifyEnabled /t REG_DWORD /d 1 /f - Check
PathRecoveryIntervalandPDORemovePeriodunder the same key; Microsoft’s defaults are 30 and 20 seconds. - Restore the missing path and confirm the events stop before changing anything else.
The array cannot keep up with the load it is given
You have this one if Latency counters climb into hundreds of milliseconds before each reset, and the events follow a schedule.
- Correlate timestamps against backup, replication and antivirus schedules.
- Stagger the jobs so only one heavy reader runs at a time.
- Ask the storage team for the array’s own latency and queue figures for the same window, so the conversation is not one-sided.
- If the workload is genuinely beyond the array, move the busiest volumes rather than tuning timeouts to hide it.
The disk class timeout is set for a faster world than this one
You have this one if Volumes dismount or cluster disks fail over during recoveries the storage completes correctly a few seconds later.
- Read
TimeOutValueunderHKLM\SYSTEM\CurrentControlSet\Services\disk. - For iSCSI-attached storage Microsoft recommends 179 or larger.
- Do not exceed 179. Past 180 seconds the system logs event ID 17 and restarts the machine.
- Change it on one node, confirm the behaviour, then roll it across the cluster.
Raising a timeout stops the resets by making Windows wait longer. That is right when the array is legitimately slow for short periods and wrong when it is failing.
Full reference
Microsoft’s published MPIO baseline
| Setting | Value | Where |
|---|---|---|
| Load balance policy | Set-MSDSMGlobalDefaultLoadBalancePolicy -Policy RR |
Microsoft DSM, global |
| Notification state | Set-MPIOSetting -NotificationState Enabled |
MPIO |
| Notification period | Set-MPIOSetting -NotificationPeriod 30 |
MPIO |
| PDO remove period | Set-MPIOSetting -NewPDORemovePeriod 20 |
MPIO; default 20 seconds |
| Custom path recovery | Set-MPIOSetting -CustomPathRecovery Enabled |
MPIO |
| Path recovery interval | Set-MPIOSetting -NewPathRecoveryInterval 10 |
MPIO; the registry default is 30 seconds |
| Path verification | PathVerifyEnabled = 1 |
HKLM\SYSTEM\CurrentControlSet\Services\mpio\Parameters |
Read the current position back with Get-MPIOSetting before and after, and change one thing at a time on one node. A cluster where every node was tuned at once, from a runbook nobody kept, is the hardest version of this problem to unpick.
The events that travel with these two
| Event | What it is |
|---|---|
| 129 | Storport: a reset was issued to the named device, and outstanding requests were re-issued |
| 153 | Storport: the I/O operation at the named logical block address for the named disk was retried |
| 140 | Storport: the device specified does not exist |
| 11 | A controller error, listed by Microsoft alongside 129 and 153 as path, driver or firmware trouble |
| 157 | The disk has been surprise removed – the point at which retries stopped working |
| 55 and 98 | NTFS reporting damage, and the instruction that the volume needs an offline full chkdsk |
Known behaviour worth ruling out first
Microsoft records one specific pattern as a known issue rather than a configuration fault: recurring event 153 inside VMware or other virtual environments reflects a Storport driver issue, and the response is to make sure the drivers are current and then go to the vendor. If the node is virtual and 153 is the dominant event, do that before spending a week on array latency.
The second thing to rule out is a slow failover rather than a slow array. Microsoft describes delays exceeding thirty seconds during failover, with I/O errors, MPIO reporting a degraded state, and events 46, 129, 140 and 153 arriving together. The remedy there is the driver and firmware refresh plus the MPIO baseline above, and on clusters raising disk resource timeouts so the node tolerates the delay.
Measuring instead of guessing
- Add the PhysicalDisk counters Avg. Disk sec/Read and Avg. Disk sec/Write for the affected disk and watch them through the failure window, not through an average day.
- Record the path count per LUN on a good node so you have something to compare against when it goes wrong at three in the morning.
- Keep the firmware and driver versions in the change record. “It started after the update” is only useful if you can say which update.
- Check the System log for event 157 and for NTFS events in the same window. Resets that end in a surprise removal or in file system damage are a different severity of problem from resets that recover.
Changing the disk class timeout means editing the registry on a machine whose storage is already unreliable. Export the key first, change one node, and confirm the value survives a reboot before rolling it out. Setting it above 179 seconds moves you into the automatic-restart behaviour Microsoft documents, which is worse than the fault you are treating.
When a licence is the actual fix
Nothing here is fixed by a licence, and any suggestion otherwise deserves suspicion: this is a storage timing problem on hardware you already own. There is one situation where a purchase is part of the answer. If the reason you cannot apply the vendor’s supported driver and firmware combination is that the node runs a Windows Server release the vendor no longer certifies, or one outside Microsoft’s support window, you will not tune your way out of it. Moving that workload onto a current release puts it back on a supported storage stack. Arco supplies Windows Server 2025 Standard, which covers two virtual machines plus one Hyper-V host per licence, and can work out the core count and the CAL position before you commit.
Every code this article covers
| Code | What it points at | Source |
|---|---|---|
Event ID 129 |
Storport: a reset was issued to the named device after an I/O exceeded its timeout, and the outstanding requests were re-issued | Microsoft Learn |
Event ID 153 |
Storport: the I/O operation at the named logical block address for the named disk was retried | Microsoft Learn |
Event ID 9 |
Listed by Microsoft among the iSCSI events as the target failing to respond in time to a request | Microsoft Learn |
Confirm the fix worked
- PhysicalDisk latency through a full backup window stays inside the range your array vendor considers normal.
mpclaim -s -dshows the expected path count on every node.Get-MPIOSettingreports the values you applied, after a reboot.- A week of production passes with no new event 129 or 153.
- No volume was dismounted in that period: no event 157 and no new NTFS events on those volumes.
Questions people ask about this
Is my disk about to die?
On direct-attached storage, possibly, and the drive’s own health and predictive failure flags are worth reading. On SAN or iSCSI storage this is far more often latency or a lost path, and replacing hardware that is not faulty changes nothing.
Can I just raise the timeout and move on?
Sometimes, and Microsoft publishes a longer value for iSCSI for exactly that reason. But a longer timeout only buys time for storage that is intermittently slow. If the array is failing, you have delayed the moment you find out – and past 179 seconds you have traded a slow LUN for an automatic restart.
Why do the events arrive in bursts?
Because a reset re-issues everything that was in flight. If the underlying condition has not cleared, those re-issued requests time out too, and the escalation repeats until either the storage answers or the volume is dismounted.
The events only happen on one virtual machine. Does that narrow it?
Yes. Microsoft records recurring event 153 in virtual environments as a Storport driver issue, so check driver currency there before looking at the array at all.
Does a newer Windows Server release fix this?
Not by itself. It brings a current storage stack, current in-box drivers and access to vendor drivers that are still being certified, which removes one class of cause. It does not make a saturated array faster.
