H310 - Slow Performance with multiple disks

[foolishlywise]

Hi guys,

It's the start of reading week so I'm doing some server stuff (moving it into my bedroom, some network stuff too). Usually, I've been way too busy to observe stuff but now I'm starting to put a fair bit of load on the server (running CCTV with redundancy recording).

So, I have a PERC H310 in charge of the HDD's (five of them - D: E: F: G. I'm not sure if this is expected behaviour but:

Lets say I transfer something from D: to F: (eg: 20GB thing). If I then transfer stuff between E: to G: (eg: a 12GB thing) the first transfer between D to F drops to 0 bytes/sec (from 140MB/s) and the transfer between E: and G: goes at 70-80MB/s. When it finishes, the first transfer comes back to life.

Doesn't matter if its a SSD or spinning HDD, same thing always ends up happening. Bear in mind that all drives are on the same card (OS drive isn't).

Bearing in mind my primary rig (X99-WS, 5820K) can do transfers without dropping all day long (with the on-board SATA controller) - but this doesn't make any sense...

 

[Terry Kennedy]

Lets say I transfer something from D: to F: (eg: 20GB thing). If I then transfer stuff between E: to G: (eg: a 12GB thing) the first transfer between D to F drops to 0 bytes/sec (from 140MB/s) and the transfer between E: and G: goes at 70-80MB/s. When it finishes, the first transfer comes back to life.

Doesn't matter if its a SSD or spinning HDD, same thing always ends up happening. Bear in mind that all drives are on the same card (OS drive isn't).

I don't know if this is the case with the H310, but some of the Dell firmware comes with ridiculously small queue depths (like 32). Flashing to generic firmware usually fixes that, but the further a card is from the LSI design, the more likely you are to run into problems. For example, on the H200E the 2 port LEDs are permanently stuck "on" in amber because the LSI firmware doesn't know about them.

If this doesn't help, post some additional info on the number/type of RAID volumes configured, etc.

 

[foolishlywise]

ridiculously small queue depths (like 32)

I thought it could be this, but it'd been flashed with IT firmware back when I got the two H310's about eight odd months ago. Should've remembered as the insulating tape I had to use on the pins got stuck in the PCI slot. I'm no surgeon but got it out in the end!

The card was basically just being used in JBOD mode (hence the IT firmware). So it had around five disks on it all doing their thing individually. In specifics, 2 1TB disks, 1 128GB SSD, 1 4TB disk, 1 3TB disk - no RAID there.

For troubleshooting, I've essentially tried the two H310's in my main rig (I got one spare on the eBay auction), resulting in the most inconsistent write speeds ever. Reflash hasn't done me any favours either. I must note that back in the day (when I first joined STH), I did have huge issues with LSI cards in general being erratic on all systems. No idea if I'm just cursed or something!