SCSI vs Direct IO

FAQ, getting help, user experience about Primo Ramdisk
Post Reply
User avatar
RAMbo
Posts: 65
Joined: Wed May 11, 2011 7:50 am

SCSI vs Direct IO

Post by RAMbo » Mon Mar 25, 2019 9:46 am

http://www.romexsoftware.com/en-us/prim ... -disk.html
http://www.romexsoftware.com/en-us/prim ... -disk.html
At disk creation I can select "SCSCI" or "Direct-IO.
Direct IO is faster, so obviously that's what has my preference :-)

Question, how can I know "Direct-IO" works for the software that's going to use that disk?
I'll put a database on the RAMdisk. Does that usually work on a direct IO disk?


I guess, it's just "Try and see" but in general, does a database benefit from Direct IO? The speed increase on 4k acces is large.

ImageImage

User avatar
support
Posts: 2364
Joined: Sun Dec 21, 2008 2:42 am

Re: SCSI vs Direct IO

Post by support » Thu Mar 28, 2019 4:01 am

I have to say that it's just "try and see" stuff as you said...Generally, yes, a database will benefit from Direct-IO ramdisks.
Primo Ramdisk | PrimoCache
Romex Software Support

User avatar
Jaga
Posts: 375
Joined: Sat Jan 25, 2014 1:11 am

Re: SCSI vs Direct IO

Post by Jaga » Thu Mar 28, 2019 5:11 am

Just a comment on your test results - CrystalDiskMark is up to version 6. To get more useful numbers an upgrade might help. I typically only use old versions to compare different hardware with older benches.

User avatar
RAMbo
Posts: 65
Joined: Wed May 11, 2011 7:50 am

Re: SCSI vs Direct IO

Post by RAMbo » Thu Mar 28, 2019 5:14 am

Those are not my test results.
I just linked to pictures of an old thread by Romex on io.

User avatar
Jaga
Posts: 375
Joined: Sat Jan 25, 2014 1:11 am

Re: SCSI vs Direct IO

Post by Jaga » Thu Mar 28, 2019 9:10 pm

Ahh, that's why. :)

As a sidenote - I've noticed some odd speed results in recent testing with RAMdisk. Some of the results with CDM are higher than a L1 on a NVMe, and some are slower than the L1. For certain things it appears the RAMdisk is better, and some it is slower than a Primocache cache task.

Post Reply