andrei Posted December 4, 2016 Share Posted December 4, 2016 I'm currently thinking of getting two 88SE9230-based controllers, namely this one from delock to connect some 4-6TB WD reds (and possibly larger in the future). Aside from the IOMMU issue described here (which doesn't bother me that much if it only affects VT-d), would there be any other reason to avoid cards with a 88SExxxx chipset? I would normally look for other options, but since sata controllers seem to be really hard to find where I live, I don't have much of a choice. This would be my second unraid build, and I would only use it for storing data, a couple of headless linux vms and some docker containers. Quote Link to comment
John_M Posted December 4, 2016 Share Posted December 4, 2016 The Marvell bug only seems to be a problem when IOMMU (Intel VT-d and AMD-Vi) is used. If you're happy to use simple VMs with no pass-through (Intel VT-x and AMD-V) then you should be fine. Some people have mentioned performance issues with certain Marvell chipsets, but by no means everyone. I use a four port 9235-based card in a HP Microserver to control two cache SSDs and four data disks in an external case via eSATA and a port multiplier and I am very happy with the performance (average parity check speed is about 120 MB/s). An alternative safe controller (though only a two port one) is the ASMedia 1060 series. Quote Link to comment
andrei Posted December 4, 2016 Author Share Posted December 4, 2016 Thanks, John! I assume these possible performance issues would be easily detectable by comparing the average parity check speed against my current baseline of just under 120 MB/s and (maybe) outer-cylinder speeds against the ~160 MB/s I currently get with the drives directly connected to the motherboard. Or am I missing something? Quote Link to comment
John_M Posted December 4, 2016 Share Posted December 4, 2016 I think you would be aware of any performance issues if you were experiencing them and I don't think your 9230 is likely to be a problem. IOMMU issues aside, I notice several people who used to use Marvell multi-port SAS controllers (such as the AOC-SASLP-MV8 and the AOC-SAS2LP-MV8) have abandoned them in favour of LSI-based cards. Coinciding with one release of unRAID (6.1 possibly?) several people reported very slow parity checks with the SASLP and SAS2LP. To some extent this can be mitigated against by tweaking tunables. As I said, I'm happy with the performance of my 9235 so I doubt that you will have any problems with your 9230 - the difference, apparently, is that the 9230 supports hardware RAID, while the 9235 doesn't. Quote Link to comment
Recommended Posts
Join the conversation
You can post now and register later. If you have an account, sign in now to post with your account.
Note: Your post will require moderator approval before it will be visible.