LSI 9361-8i not a good buy, or am I mistaken?

Notice: Page may contain affiliate links for which we may earn a small commission through services like Amazon Affiliates or Skimlinks.

aarbee

New Member
Feb 9, 2018
17
0
1
44
Hello,
First the story, why I did what.
I have an Qnap 673. Which is perfect for fileserving, and 2 VM's.
Yet it lacked power while using it as a Plex server at home.
So did build an AMD 3700X pc, and added 4x 1 TB 2.5" WD red.
And used Windows 10 Storage Spaces for building a raid system.

Now the problem
Then I did find out that the performance of the Storage Spaces is lower then the Raid 6 I had used in the Qnap.
The Qnap had the same disks in them, and The read performance was high. Write was a bit slower.
I think that the diskspeed was around 100MB per disk. In a Raid 6 it had a readspeed of over 300MB. Write was around 100MB

Though on the AMDcomputer the readspeed is 200MB and the write speed 25-35 MB for small files and 40-70MB for large files.
I considered that low. And I bought an LSI 9361-8i from Ebay. Build it in, and had about the same performance.
Wether I used raid 10 or 6, did not matter.

Does this mean that the Linux version on the Qnap is so much faster then the Windows version?
I had tried running Freenas and other Linux versions on that mainboard, but as it has an onboard 10Gb networkcard, which appeared to be the bottleneck. It was simply not recognized. And of and on other hardware was not recognized.

Back to the issue.
Why is the LSI having the same performance? It is even a tad bit slower than Storage Spaces?

I did read on several places that adding a CacheVault would solve the issue, as the availability of a battery, tells the Cache on the board that it can work fullspeed.
But I cannot imagine a battery speeding up a disk. Can somebody give some insight?

I have been thinking about the CacheCade. But I do not think that it will add anything. Adding SSD's in the Qnap for cache did not help there either, as I am using to few files to have advantage of cache. Maybe I am mistaken.

Some advice would help me.
Thanks ahead,

RobB
 

j_h_o

Active Member
Apr 21, 2015
644
179
43
California, US
The RAID adapter should be significantly faster. Should give you 250-450MB/s read/write. How are you testing the speeds?
Try Download ATTO Disk Benchmark - MajorGeeks and post results, please.

The RAID adapter has 1GB of memory onboard. The battery protects it from power failures, allowing it to use the cache to serve files from that cache directly, without hitting the disks.

CacheVault allows you extend the memory further and it caches more of the array on the SSD.
 
  • Like
Reactions: aarbee

aarbee

New Member
Feb 9, 2018
17
0
1
44
I did test it by copying data from and to an 2 TB NVME disk. And see the copy speed.
I also did some tests with Crystal Diskmark.
This week I did test on Storage Spaces with a mix of WD blue, red and Seagate notebook disks.

I will attach a PDF where in top you see the once noted data from actual copy and below the crystal diskmark this week.
I had the LSI build out of the computer, as I did not see any advantage. Only negatives. More power and a lot of heat. (90oC). I have replaced the cooling paste, and that helped settle the temp down to 60oC.

If I read you correct, it means: No battery, no cache?

And with CacheVault, you mean CacheCade? The first is in fact, from what I read a modern BBU + extra cache. And the latter is software to add a SSD to the cache?

Thanks,

Rob
 

Attachments

j_h_o

Active Member
Apr 21, 2015
644
179
43
California, US
Yes, if there's no battery, the controller does not use onboard cache memory.

You can simulate the effect of the battery by changing the cache policy on the array. Set it to "Always Write Back" (use cache and ignore lack of BBU) instead of "Write Through" (doesn't use cache) or "Write Back" (only does it when BBU is present).

Yes, CacheCade uses the SSD to further cache blocks on the array, extended it beyond the memory that is available on the controller.
 
  • Like
Reactions: aarbee

aarbee

New Member
Feb 9, 2018
17
0
1
44
What is wise?
Put the Card back, and buy the Cachevault or the CacheCade?
Will CacheCade work without the battery?

Thanks again,

RobB
 

j_h_o

Active Member
Apr 21, 2015
644
179
43
California, US
I would check to see if the BBU/setting to Always Write Back improves performance first. Can you set the cache setting to Always Write Back and re-test the speeds?
 
  • Like
Reactions: aarbee

aarbee

New Member
Feb 9, 2018
17
0
1
44
Your disks are probably suspect too. What drive models do you have?
I did miss this post yesterday.
Yes the notebook disks are old and slow.
I will replace the WD blue and Seagate Momentus with 3 other WD red 2.5" And redo the Storage Space test.
And I will make some time to build in the Raidcontroller. But that can take a fortnight. I will get back to this post and let you know.

Thanks,

rob
 

aarbee

New Member
Feb 9, 2018
17
0
1
44
I have replaced the WD blue 500 GB and Seagate Momentus 640 GB with 2 WD red 1 TB 2.5".
And indeed the readspeed went up, pretty much for Raid 5 and Raid 10.
 

Attachments

Last edited:

aarbee

New Member
Feb 9, 2018
17
0
1
44
I would check to see if the BBU/setting to Always Write Back improves performance first. Can you set the cache setting to Always Write Back and re-test the speeds?
@j_h_o
I did test everything.
And really, it works. I have connect the Raidcontroller. And connected 4 2.5" 1TB WD red. And started testing.
Raid controller without Write back cache is mostly faster, especially in Reading.
And it blows a ship out of the water with the cache enabled.
Even copying files to a SSD goes with 200+MBps.
Standard I worked with 64kb blocks but changed it for a test in a Raid 10 setting.
That was even faster. Though I think that 64kb are the better way to go.

Considering what you had said: Go for the BBU or a variant of it.
Thanks for pushing me a bit, to retest it all.

But that controller gets very hot.

RobB
 

Attachments

j_h_o

Active Member
Apr 21, 2015
644
179
43
California, US
Cool. Glad that helped.

Yes, there's really no point in using that controller unless you're using the cache (IMO). So you should either get the BBU (safer) or use it in "always write back" and risk data loss during power outage.

I wouldn't bother with CacheCade for your usecase.

Yeah, the controller expects to be in a case with lots of moving air. You can ziptie a Noctua to the heatsink, perhaps? A NF-A6x25 PWM would probably work well enough, and still be quiet, if you have a 4 pin connector on your board available.
 
Last edited:
  • Like
Reactions: aarbee

aarbee

New Member
Feb 9, 2018
17
0
1
44
@j_h_o
I too think that CacheCade is not the way to go for me.
Thank you for your advice, tips etc.

Now that I am looking in the cache, I also looked in to the card itself.
It appears to be a Dell brande MM445. But it is a 9361 with 1 GB of cache. Meanwhile I learned there is also a 2 GB version.
I always understood that that the Cache was not on the card, but in the CacheVault, but obviously I am wrong.
Yet when I look at the CacheVault, I see the one with a Battery which you can attach to the Card.
But also something to clip on the Raidcard ( CVFM04).
What does it add to the CVM02? Because I saw that you have to add the Battery Pack to the CVFM04.
 

Rowan @ Jetboy

New Member
Oct 30, 2014
11
1
3
53
The CVFM04 is the CacheVault flash module, a daughter card which attaches to the 9361-8i via a connector next to the heatsink. That's where the flash memory is. On the CVFM04, there's a socket to plug in the CVFM02, which is the CacheVault power module; correctly a supercapacitor pack, but generally still called a battery backup unit (BBU). You can't plug the CVFM02 into the 9361-8i without the CVFM04.

Because they're supposed to be used together, these two items tend to come in a pack: LSI00418 / LSICVM02 LSI MegaRAID CacheVault Accessory Kit, which also contains a metal clip for holding the CVFM02 and two different lengths of CVFM04 to CVFM02. What you don't get is anywhere to attach the metal clip; there nowhere to attach it on the 9361-8i You can zip tie it to your case, or, if you have a spare slot, buy a LSI00291 mounting bracket.

Can you use the CVFM04 without the CVFM02? No idea, but I'm not sure that you'd want to.

All three components: CVFM02 connected to the CVFM04, which is mounted on the 9361-8i.

IMG_20200802_125259.jpg

In-situ, with the CVFM02 clipped to a LSI00291 mounting bracket:

IMG_20200802_125323.jpg
 
  • Like
Reactions: aarbee

aarbee

New Member
Feb 9, 2018
17
0
1
44
@Rowan @ Jetboy
Thank you for the explanation. It is clear.

I only wonder, what have I been testing with the writeback cache? Which Cache?
Cache on the card or cache on the disk? Because I read that the particular card has 1 GB of Cache.

Can I count them together (in case there is cache on the vault and on the card).
 

Attachments

Rowan @ Jetboy

New Member
Oct 30, 2014
11
1
3
53
Hopefully someone who knows more about this can confirm it's accurate. My understanding is that:
  • On its own, the 9361-8i has a 1GB DDRIII DRAM cache. If you get a power failure, you're going to lose the data in the cache. *If* you could attach a battery backup unit directly to the card, it would keep your cached data safe for up to 72 hours.
  • The CVFM04 has (presumably 1GB) of NAND flash cache. If you get a power failure, it moves the data in the 9361's cache to its own cache, where the CVFM02 will keep it safe for over three years.
The CVFM04/CVFM02 doesn't give you a bigger cache. It doesn't improve performance. It does add data security. From what I've read, the CVFM04 doesn't add anything without the CVFM02 supercapacitor. In fact it may even disable the 9361's cache unless it's able to secure it. It certainly does when it's conditioning the supercapacitor before first use.

So you're just testing a bare 9361-8i at the moment. You're definitely testing the 9361's cache if you're in write-back mode. Buried in the card settings is the Disk Cache Policy, that allows you to set whether the disk's onboard cache is used. If you haven't set this to 'Disabled', you're probably using the disk cache as well.

You'll likely find that both card and disk caches disabled gives poor performance, which'll get better with disk cache only, even better with card cache only and negligible improvements with both caches enabled. Your mileage may vary. I haven't tested any of this because I want my caches to have a backup power supply. I don't have a UPS, so the disk caches are a liability for me.
 

aarbee

New Member
Feb 9, 2018
17
0
1
44
@Rowan @ Jetboy Thank you for your insight. I had that feeling too.
I am having an UPS. Which the CVFM02 is infact too. Yet only for the Raid-Card.

Honestly I have been playing with the thought of buying WD red SA500 SSD. 4TB. I think I need 3 of them them. So a raid 5 setup.