VMs on a Host Occasionally Respond Slowly to User Operations Because of a Slow Storage Link Rate

Publication Date:  2015-03-05 Views:  298 Downloads:  0
Issue Description
VMs on the host at a site occasionally respond slowly to user operations, but the VMs on other hosts work properly. 
Alarm Information
No alarm is generated for any of the FusionCompute V100R002 versions.
Alarms indicating dramatic network port speed changes are generated for any of the FusionCompute V100R003 versions.
Handling Process
Disconnect and then connect, or replace the network cables. 
Root Cause
After a storage device is connected to a host, the multipathing software on the host sends I/O data of the host to the storage links of the storage device based on a specified algorithm and then writes the data into the device. This mechanism helps to improve I/O performance and implement link redundancy.

Both Huawei multipathing software and common Linux multipathing software send I/O data based on the link polling algorithm. If one of the links is interrupted, the software excludes the link when sending the I/O data. However, if a link is faulty but not interrupted (low link rate), the software still sends I/O data to this link. Because the I/O data is sent to the storage link at a low rate, the host I/O processing is adversely affected.

Then, you can locate the storage links with low rates by checking ports on the access switch connecting the host and the storage device and those on the access switch connecting to the aggregation switch.

Suggestions
1. During routine maintenance, you must ensure performance of the network between the host and the storage. Otherwise, VM performance is adversely affected.
2. In V100R003 and later versions, FusionCompute monitors switches and reports alarms when the network port speed dramatically changes. During routine maintenance, you must pay attention to these alarms and handle them in a timely manner.

END