News for hackers. Stuff that matters.
Re:Linux 5.10 BTRFS performance regression by xaduha on Wednesday December 23, 2020 @08:59AM (#25517668)
You can get 'surplus' HP P410 from Ali or Ebay, they are cheap and work just fine. Always was told that you can't beat hardware RAID and now that they are available people don't seem to be interested. Just looking at it I can say that apart from CPU/GPU chips themselves it's the best piece of electronics in my PC.
Re:Linux 5.10 BTRFS performance regression by BuildTheRobots on Wednesday December 23, 2020 @09:50AM (#25518122)
Where as I was always told to avoid hardware raid if possible, and that's been a mantra for 15+ years.

The reason I was given is twofold. 1) If your RAID card dies it's nice to be able to plug drives into any controller and be able to access the data. 2) "look at how often MDADM fix bugs. Now look at how often your RAID card firmware gets fixed... use MDADM".

Modern filesystems like btrfs & zfs both prefer to deal with the RAID aspects themselves as it gives them far more control. Even if you're a hardware raid afficianado, I wouldn't recommend formatting (eg) ZFS on top of a hardware raid pool.

Re:Linux 5.10 BTRFS performance regression by xaduha on Wednesday December 23, 2020 @10:18AM (#25518409)
> Now look at how often your RAID card firmware gets fixed

It was updated plenty, from 1.58(B) (earliest I can find) to 6.64(B)

https://support.hpe.com/hpsc/swd/public/detail?swItemId=MTX-...

I'm not recommending some abstract hardware RAID, I'm recommending this particular card for personal use, with backup obviously. Although if you just use RAID 0 or 1, then data is perfectly readable outside of RAID with a normal SATA controller.

Re:Linux 5.10 BTRFS performance regression by BuildTheRobots on Wednesday December 23, 2020 @10:53PM (#25524893)
Even for home use I'm not sure I could recommend them. From experience using them in servers, I'd site a handful of problems:

+ Raid 5 requires BBWC + Raid 6 requires licensing + HPE requires a support contract to download most firmware + When used in servers, they report non-hp drives as being in a constant fault state.

Re:Linux 5.10 BTRFS performance regression by xaduha on Thursday December 24, 2020 @05:14AM (#25526341)
> BBWC

why wouldn't you use it if you have it?

> Raid 6 requires licensing

Keys are out there and can be googled, in any case you probably don't need RAID 6 or any extra features for personal use

> HPE requires a support contract to download most firmware

Um, firmware for this card is on the page I linked, free as in beer

> When used in servers, they report non-hp drives as being in a constant fault state

I wouldn't know, haven't seen anything of the sort personally

Re:Linux 5.10 BTRFS performance regression by BuildTheRobots on Thursday December 24, 2020 @05:36PM (#25532149)
I mentioned the BBWC because the first few links I found to buy the card didn't include it.

Nice to see firmware available - in 2014 HP started requiring support contracts or warranties to download firmware updates for the servers which really turned me off them. I didn't realise RAID firmware wasn't included in that (for some reason iLO firmware isn't either, but I think that's probably because it includes OSS). For this reason alone I'd avoid HP on principal.

I'm hoping one day I can find a reasonably cheap RAID card that still supports using hte BBWC in JBOD mode...

Re:Linux 5.10 BTRFS performance regression by cat199 on Wednesday December 23, 2020 @11:31AM (#25519292)
they are different animals .. 'right tool for the right job'

hardware raid has the hw complexity / pickyness you mention, but also is safer in crashes for checksumming raid levels (e.g. 4,5,6) since the stripe checksum happens from controller->disk in hardware rather than os->controller->disk in software, so you're less likely to lose a stripe due to OS issues (all else - like bugs - being equal)

also, IO wise, any mirrored raid is going to require N*mirrors of IO on the system bus, so you're more likely to saturate it

that said, with faster tech (NVMe), individual drives can easily saturate a single card, so its 'worth' paying the multiple IO pentalty multiplexed onto multiple individual lanes since the card is more likely to be the bottleneck than the bus

also, hw raid controllers even in jbod mode might be needed to get enough device fanout, though you're not using the controller for raid in that case

Re:Linux 5.10 BTRFS performance regression by buran77 on Wednesday December 23, 2020 @09:50AM (#25518119)
> Always was told that you can't beat hardware RAID

Recently with super powerful multi-core CPUs being the norm especially in places where RAID is a consideration, hardware RAID makes less and less sense. You just create additional complications like when moving your RAID array between 2 otherwise identical cards but with different firmwares fails. So you end up always buying spares and testing the migration before putting any data, then you freeze the config in place. On the other side you may be stuck forever with a buggy firmware.

Re:Linux 5.10 BTRFS performance regression by speakeron on Wednesday December 23, 2020 @09:33AM (#25517949)
The problem with hardware RAID is that if the hardware fails, you need to replace that specific card with the same type to recover the data. If they're cheap enough and you want to go down that route, then buy two so you've got a backup.

On the server side, the tendency is now towards vertical striping (directly addressing disks on multiple servers through the network with software).

Re:Linux 5.10 BTRFS performance regression by ksec on Wednesday December 23, 2020 @10:17AM (#25518392)
>Always was told that you can't beat hardware RAID

Is that new?

I have always been told to avoid hardware raid when possible.

Re:Linux 5.10 BTRFS performance regression by xaduha on Wednesday December 23, 2020 @10:19AM (#25518423)
There were pseudo hardware RAID controllers aplenty, that's what you were told to avoid.
Re:Linux 5.10 BTRFS performance regression by rincebrain on Wednesday December 23, 2020 @10:31AM (#25518574)
Not necessarily the case - avoiding pseudo-HW RAID controllers can be reasonable advice, and may have been what was said, but some of us were explicitly told to avoid real hardware RAID controllers, period, for the reasons that have been listed upthread. [1]

It's great that you've had good experiences with this particular hardware RAID controller and are recommending it to others, but it's also the case that there are legitimate reasons to avoid all hardware RAID cards in some setups.

[1] - https://news.ycombinator.com/item?id=25518122

Re:Linux 5.10 BTRFS performance regression by xaduha on Wednesday December 23, 2020 @10:36AM (#25518632)
Well, what are your experiences exactly?
Re:Linux 5.10 BTRFS performance regression by vetinari on Wednesday December 23, 2020 @11:45AM (#25519444)
Recently, I've switched P420i to HBA mode -- that's all ZFS running on that Proliant wanted.

Too bad it's not possible to boot from it in HBA mode too.