HPE Storage Users Group

A Storage Administrator Community




Post new topic Reply to topic  [ 3 posts ] 
Author Message
 Post subject: Performance Consideration for the Copy CPG
PostPosted: Sat Mar 21, 2020 6:26 pm 

Joined: Wed Jan 16, 2019 4:26 pm
Posts: 35
Location: Florida
So I've read the HPE snapshot doc along with the best practices guide and one thing still isn't clear: the impact of the snapshot cpg settings on write performance.

If I'm reading correctly, the copy-on-write (COW) mechanism has to first copy the original block from the user cpg to the copy cpg before new incoming data is written. It seems to me that heavy incoming writes could cause enough "I/O multiplication" to become a drag on performance if the disks underlying the copy cpg can't keep up.

Is this a problem in actual practice? It seems once the original block is copied out, future updates to that block wouldn't cause more COW activity until another snapshot has been taken so maybe it's not much of an issue.

I ask because I'm trying to design a new cpg/AO strategy and I'm wondering if I should do something crazy like set up a separate AO for the copy cpgs to get some of those blocks into SSD.


Top
 Profile  
Reply with quote  
 Post subject: Re: Performance Consideration for the Copy CPG
PostPosted: Sun Mar 22, 2020 1:39 pm 
Site Admin
User avatar

Joined: Tue Aug 18, 2009 10:35 pm
Posts: 1239
Location: Dallas, Texas
The universal answer is "that depends".

Worst case scenario for illustration purposes, if your User CPG was R1 SSD (fastest), and your Copy CPG is Nearline Raid6 10+2 (slowest), and your benchmark is 100% sequential writes, from the starting block to the last block of the VV for 100% change rate. Yes you would measure a performance delta.

In real world practice, it depends on the specific workload, but generica averages you looking at 80% reads, and less than 10% change rate for the written data. Keep in mind that a new write to empty thin provisioned (as opposed to a rewrite of a provisioned block) space does not have a COW cost. For the overall average 3PAR system that is not oversaturated with workload, the penalty of COW goes unnoticed. Of course, if any array is already overloaded/saturated with workload, then adding more work of anytype (COW, Replication, compression, etc) will have a larger impact.

What is your current average latency of your AO volumes, vs what is acceptable to your application? Example: If you are sitting at a 4ms average but can tolerate 10ms, you have some headroom. Again "it depends" on how much SSD vs spinners you have, how AO is set.

_________________
Richard Siemers
The views and opinions expressed are my own and do not necessarily reflect those of my employer.


Top
 Profile  
Reply with quote  
 Post subject: Re: Performance Consideration for the Copy CPG
PostPosted: Mon Mar 23, 2020 8:50 am 

Joined: Mon Sep 21, 2015 2:11 pm
Posts: 1017
Location: Europe
And as long as your system isn't hammered, you have the magical thing called write cache :)

If you read the virtual copy whitepaper (https://h20195.www2.hpe.com/v2/GetPDF.a ... 486ENW.pdf) it states pretty early in the text:
Quote:
HPE 3PAR Virtual Copy implements an
efficient variant of copy-on-write (COW) and redirect-on-write (ROW) methodologies. For COW, the HPE 3PAR OS uses a delayed copy-onwrite (DCOW) process that eliminates any performance impact to host I/O. DCOW is used for snapshots of fully provisioned and thinly
provisioned volumes. DCOW relegates the reading of the original data, updating of the base volume with the new data, and the copying of
the original data to a background process after the write update has been acknowledged by the host.


So what I'm reading is that it may impact the overall maximum "performance capacity" of the 3PAR as is has to do some background IOs but the host performance (as in latency) shouldn't be affected. So if you're running your array at 90% of it's performance capacity it might be a really bad idea to start doing snapshots but if it is just lurking along at 25% you shouldn't notice any difference.

_________________
The views and opinions expressed are my own and do not necessarily reflect those of my current or previous employers.


Top
 Profile  
Reply with quote  
Display posts from previous:  Sort by  
Post new topic Reply to topic  [ 3 posts ] 


Who is online

Users browsing this forum: Google [Bot] and 33 guests


You cannot post new topics in this forum
You cannot reply to topics in this forum
You cannot edit your posts in this forum
You cannot delete your posts in this forum
You cannot post attachments in this forum

Search for:
Jump to:  
Powered by phpBB © 2000, 2002, 2005, 2007 phpBB Group | DVGFX2 by: Matt