Is it fine when this value at 9 ms ?
Is it fine when this value at 9 ms ?
Hi Nattakit
Latency below 10 ms should be fine. Our HNAS is typically between 5-15ms without any problems detected.
I received the following from HDS at a time I asked the same question as you.
“Disk Stripe write latency” is the global value specific to a node (not a cluster and not specific to a file system). As you might know when a storage pool is created using multiple LUNs, it creates a stripe. Before we write anything to the backend storage system, the HNAS gathers writes (called, Superflush) to a specific value for each LUN (up to 480KB for 8.x and below releases, up to 4MB for 10.x and above releases). “Disk Stripe write latency” is the latency indicator of multi buffer (Superflush’ed) write requests when it leaves the HNAS to the backend storage system. As the Superflush’ed writes are written using larger I/O sizes (and mostly sequential writes), the response times will also be higher.
On occasions, the HNAS might not be able to gather writes and Superflush efficiency might be poor. In that scenario, the HNAS will be writing at smaller I/O sizes (4KB or 32KB) to the backend storage system. “Disk write latency” is the latency indicator of such writes.
Regards
Sciipiio
Hi Sciipiio,
Thank you for your helpful reply but 9 ms is constant and not reduce. It is should be fine ?
Regards,
Nattakit
Hi,
Anyone knows what disk write and read latency mean ?
Anyone can share me document for HNAS Performance Analyze ?
Write latency - the avg write response time for all System Drives on the NAS head.
Less than 20ms is normal
Stripe Write latency - the avg superflush write response time for all System Drives on the NAS head.Less than 60ms is normal
Superflush is when HNAS coalesce multiple disk writes into a single write operation in an attempt to write a full RAID stripe worth of data to a System Drive so the RAID controller does not need to perform read/modify/write to calculate parity.
Read latency - the avg read response time for all System Drives on the NAS head.
Less than 20ms is normal
The spikes you are seeing could be caused by numerous things. Is there any replication taking place on the HNAS? Frequent snapshot deletions? Spikes of IO from clients?
Hi,
Yes, we have JetMirror 4 jobs run as every 1 minute.
This Storage Pool used for NFS VMware 250 guests with 75% Write 25% Read behavior.
Can you post the CPU/MFB graph also?
What you are seeing is normal with Jetmirror. When it starts or ends a replication, the CPU can temporarily spike to 100%. The spikes in disk latency could be due to Jetmirror sending a burst of requests to the disk or snapshots being deleted. So I would not worry about it.