5

I have a LSI Megaraid 9271-8i using 6x samsung 850 pro SSDs and I can't modify the "disk cache policy" it's greyed out in the properties for any virtual disks.

  1. Why is this, the 850 pros are listed in the compatibility matrix?

  2. Is this having a significant impact on performance (am I missing out)?

  • 6
    Why do you want to NVRAM cache SSDs? You'll just increase operational latency, and that's pretty much it... – BaronSamedi1958 Dec 24 '16 at 17:50
  • That's what I was assuming. I just wanted to know if there was any reason to enable/disable it. Why do the SSDs have the cache in the first place if it "only increases latency"? – user273074 Dec 24 '16 at 17:52
  • Also aren't we talking about DRAM? I thought the cache was a buffer to stage writes so they can be reported as written even before they are actually stored on the drive, thus increasing speed? – user273074 Dec 24 '16 at 18:27
  • 5
    On-device write buffer backed up by built-in capacitors is a non-controllable thing, while cache on RAID controller makes a lot of sense for spinning disks but not for flash memory. – BaronSamedi1958 Dec 24 '16 at 18:34
  • So, there are controls for the cache onboard the raid controller. I can adjust them and set the SSDs to not use it as I saw no performance benefit when I did. The "disk cache policy" was a separate setting independent of that one. It seams to imply its for the cache onboard the actual disk. Not the raid controller. Is this the case or am I mistaken? – user273074 Dec 24 '16 at 19:09
1

This is working as intended.

In traditional spinning rust, hard disk drives, there is cache on the controller to help improve the performance since read/write is slow, particularly so on 5400 and 7200 RPM drives. When using a card like LSI's they have their own caching (usually much larger than the HDDs do) and when they write to the drive, they instruct the drive to bypass their local cache and write straight to disk. This is preferable where you can't have data corruption due to a power loss. When power is lost, the drives lose their cache so anything that hasn't been written to disk is lost.

The raid card can be supplemented with a battery so that it can survive power loss scenarios and ensure that data gets written to disk. SSDs may or may not have onboard cache. The Samsung 850 Pro does not, so the interface is telling you it can't disable it. Samsung has ways to use your system ram as cache, but it's not physical memory on the drive itself. You still run the risk of data loss or corruption if your LSI MegaRaid does not have battery backed cache and you're using the raid card's caching. Often times, in enterprise scenarios, there are enterprise class SSDs which actually have a small bank of capacitors to ensure the drive finishes all pending read/write operations before it goes dark which is great when coupled with battery backed cache and/or a proper UPS solution.

With cache on the raid controller, if it's not battery backed and you're using all SSDs, it's best to just bypass the cache as it's one less place for data to get lost/corruption if there is a power blip and, as you said, there is no real performance benefit when using SSDs. Even if you were using traditional drives and you're sensitive to data corruption issues (think databases), then you'd best bypass caching systems that are not battery backed. Cache is much more important for the traditional HDDs to improve latency for writes and frequent reads of data from the disks.

| improve this answer | |

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.