FAS 2240-4 bottleneck reading large files

[bgp]

Hi

Hardware:

FAS 2240-4 NetApp Release 8.2.4P6 7-Mode 24*1TB-SATA-Disks, FС 8G

Tell me where is the bottleneck for reading.
Netapp disk system reads a maximum of 200 mb / s

I copy a large zip file of 90 gigabytes, a compressed backup copy of the database.

Copying from LUN on fiber channel 4gbit and further on the local network 10 gigabits to a RAID 0 server

I would like to have a speed of ~ 300 mb / s

The disk system cannot handle writing, it shows sysstat well, but I don’t see any bottlenecks with reading

NVRAM bottleneck ???

[SpindleNinja]

What are the specs of the server that's connected at 10Gb?  is it a comparable generation to the 2240?   

Be sure that you're comparing apples to apples as best you can and not apples to oranges.  

[bgp]

ibm system x3650 m2 7947 and  ibm system x3650 m3

[SpindleNinja]

What drives do you have in there?   

[bgp]

Netapp - drives SATA 1GB x 24 ,  there are no drives in IBM servers ESXI is loaded with usb flash,  the NAS drives server has 15,000 rpm Raid0

[SpindleNinja]

Just to clarify the "Server NAS" is the IBM model or those where your ESX hosts?   What's the "Server NAS" hardware.  

But sounds like you're comparing SATA to SAS and RAID DP to RAID 0? 

[bgp]

The NAS Synology RS820+  server for tests was built on it RAID0, the local network 10G was checked all well.

I don't compare RAID-DP with RAID0, my task is to copy the backup copy from RAID DP to another server as quickly as possible.

I understand that SATA disks are slow to write and I understand this well and see the sysstat utility, but I do not see the sysstat read delay counters, the CP ty column does not show this.

fixed the schema

[SpindleNinja]

The CP TY is for write caching only.   more info here:  

https://kb.netapp.com/Advice_and_Troubleshooting/Data_Storage_Software/ONTAP_OS/FAQ%3A_Consistency_Point

Read caching comes from SSDs or Flashcache etc (i don't think the 2240 has).     According to your output,  you're hitting cache 100% for reads (system ram).   But it doesn't look like it's maxing disk or cpu util.  

Is there any dedup or compression  turned on the volumes?  

Also,  look at what else is in the chain.  FC switch etc...  How are their counters looking?

[bgp]

FAS 2240 cannot use Flash cache, only Flash Pool.

dedup and compression off

The entire chain was tested without Netapp speed   ~ 350 - 400 mb / s

I made a large ram disk (100gigabyte) in a virtual machine and copy speed ~ 400 mb / s along this chain

We still have a shelf 4243 SAS disks 15000 rpm 300 GB x 24, they work on reading 300 MB / s, which is also not very fast

[SpindleNinja]

so bottom line,  you're  wondering why you can't hit 100% utilization on reads on the 2240?  

[bgp]

Yes.

[SpindleNinja · Accepted solution]

Think of ONTAP as a dam,   It's designed to take as much water on as it can within its design (writes) ,  however, it can control how much it will release the water (read).   And controlled release, is safer and more stable than just letting it all go at once as that could destroy the damn.  

[bgp]

I thought about this, an overloaded system will have a bad response, and other users will not be able to work comfortably.
You're right.
We will stop on this, we will not over-utilize the system by squeezing all the juices out of it.

Thank.

[Admin edit for language]

[paul_stejskal]

So a couple things here: 1) is your FAS 2240 sized for 300-400 MB/s from the account team? 2) 3.5" 7200 RPM SATA drives have a poor profile of performance, so you're hitting a potential latency bottleneck on the disks themselves just because of the rotational speed and size, and SATA interface. The statit output shows decent response (22 chains * .2 ms is about 4.4ms per op).

It's also possible the bottleneck is further upstream. Have you looked at the response times from vSphere?

[bgp]

ESXI results:
1. Reading is a good result
2. On writing it is difficult for him

attached screenshots of esxi results

[paul_stejskal]

Ah you're hitting a limit on the node it appears. High water mark CPs every second. That would do it.

[bgp]

Yes