Windows Server Summit 2024
Mar 26 2024 08:00 AM - Mar 28 2024 04:30 PM (PDT)
Microsoft Tech Community
LIVE

Parity Space Write Performance

Copper Contributor

Can you explain why write performance of a parity space is so poor, especially when other types of SDS platforms can give acceptable RAID50 write performance with the same (or less) hardware? Are there plans to improve this?

2 Replies
S2D uses dual parity and if you look at any system and compare Mirror to Parity, Parity is always slower for writes and about the same for reads. When you write to a parity volume you have to read-in the parity, make the write change and re-calculate party and write that back. S2D is a distributed system where these writes happen across nodes so there are network transits for the write actions. S2D is the only solution I'm aware of that provides mirror accelerated parity which allows a volume to take writes in the mirror part of the volume (fast writes) and then in the background rotates the data to parity. We have done enhancements to the parity in this falls 1709 release. If you use a scale of 0 100 with mirror performance being 100, Windows Server 2016 performance would be at the lower part of the scale and the 1709 release it will be closer to the 100 side of the scale, However due to the calculations for parity it will never be as good as memory. When comparing parity to mirror on S2D, it's also good to realize that you are transiting servers for the I/Os so network latency/bandwidth also has an impact that you don't see if you are using a stand-alone system where all I/O is going through a local storage bus.

Hi - Robert here,

 

two things that make me happy to hear ;)

1. 1709-Version is W2k16-Version with all the new features that the recently realeased preview got (incl. dedup) means it´s coming next month, right ?

2. Perfromance boost for Parity Tear ....

 

Another Question - I´ve been asking this in the comments of the blog post about "Consumer SSD for S2D - Don´t do it" here https://blogs.technet.microsoft.com/filecab/2016/11/18/dont-do-it-consumer-ssd/ (search for "Is the “insistence on write-through” configurable?") and did not get any reply on the Testlab-Sceario ... Is there anything you can tell me about it ?

 

thx in advance

Robert