BUILD HighPoint Rocket 640L 4 Port HBA

[Robert Trevellyan]

In the interest of science, I've moved my main pool to my HighPoint Rocket 640L 4 port HBA. Unless I post back here reporting a problem, you can assume it's still working and draw your own conclusions from that knowledge.

Moderators, feel free to move this post wherever you think it belongs.

Background: I bought this HBA when my FreeNAS was still a toy and I was trying to cram as many old 250GB and 500GB drives as possible into a Vostro 420. Since then I've deployed it in various ways, most recently for a couple of months running a secondary 2-drive mirror pool holding a bunch of DVD movie rips for direct streaming. At no time have I detected any problem with this HBA.

Chipset: in the boot sequence it shows up as <Marvell 88SE9230 AHCI SATA controller> (thanks @mav). The card has no hardware RAID capability, so the question of flashing to IT mode does not arise. All SMART data passes through unscathed.

Workload: my FreeNAS performs the following duties, in roughly descending order of importance:

• backup destination for 4 Macs and 1 PC
• anywhere from 1-4 Ubuntu Server VMs hosted by a VirtualBox jail (1GB RAM per VM)
• BTSync always-on read-only node
• Plex Media Server, maximum of 2 simultaneous clients, media up to 720p, Sonarr & SABnzbd

The only time the pool is really busy is during bi-weekly scrubs.

Hardware list from my signature:
Dell PowerEdge T20 | Pentium G3220 @ 3GHz | 32GB
RAIDZ2: 4x WD20EFRX
APC Back-UPS CS 500
FreeNAS-9.3-STABLE

 

[BigDave]

In the interest of science, I've moved my main pool to my HighPoint Rocket 640L 4 port HBA. Unless I post back here reporting a problem, you can assume it's still working and draw your own conclusions from that knowledge.

Ok, but just so you know, if anything bad happens, the ZFS police are coming to take your Pony (see my signature for reference):p

 

[Robert Trevellyan]

if anything bad happens, the ZFS police are coming to take your Pony

Yes, I've hidden my pony ;)

 

[jgreco]

Yes, I've hidden my pony ;)

We're having a Mother's Day cookout here.

 

[Robert Trevellyan]

Small update ... I realized that if something goes wrong, science would be better served if the pool were split 50/50 between the onboard ports and the 640L ... and the pool would be more likely to survive.

 

[Mirfster]

Small update ... I realized that if something goes wrong, science would be better served if the pool were split 50/50 between the onboard ports and the 640L ... and the pool would be more likely to survive.

True science does not like partial answers. All in for the win or the fail. There can be only one answer. ;)

 

[Robert Trevellyan]

Next scrub is due this coming Sunday. Stay tuned ...

 

[Robert Trevellyan]

Here's something I just noticed today in my dmesg output:

Code:

<snip>
ahci0: <Marvell 88SE9230 AHCI SATA controller> port 0xe050-0xe057,0xe040-0xe043,0xe030-0xe037,0xe020-0xe023,0xe000-0xe01f mem 0xf7c10000-0xf7c107ff irq 16 at device 0.0 on pci4
ahci0: AHCI v1.20 with 8 6Gbps ports, Port Multiplier not supported
ahci0: quirks=0x900<NOBSYRES,ALTSIG>
<snip>
pass2 at ahcich7 bus 0 scbus7 target 0 lun 0
pass2: <Marvell Console 1.01> Removable Processor SCSI-0 device
<snip>
pass2: Serial Number HKDP221516WL
pass2: 150.000MB/s transfers (SATA 1.x, UDMA4, ATAPI 12bytes, PIO 8192bytes)
SMP: AP CPU #1 Launched!
<snip>

Note the lines beginning "pass2". I have no SATA 1.x devices in my box, so on a hunch I Googled the serial number, which returned "About 58 results". Almost all of them include dmesg output from some flavor of BSD and the pattern is always the same. Some flavor of Marvell controller is detected, later followed by a non-existent SATA 1.x device with the same serial number, right before the processor launch.

Does anyone know how to interpret those lines? Is this some kind of phantom device detection associated with the BSD driver for this controller? Could it possibly be relevant to the problems some people have with Marvell controllers? Or is there nothing to see here folks, move along?

 

[Robert Trevellyan]

I see it more clearly now, but I still don't understand:

Code:

camcontrol devlist
<WDC WD20EFRX-68EUZN0 82.00A82>    at scbus0 target 0 lun 0 (pass0,ada0)
<WDC WD20EFRX-68EUZN0 82.00A82>    at scbus2 target 0 lun 0 (pass1,ada1)
<Marvell Console 1.01>             at scbus7 target 0 lun 0 (pass2)
<WDC WD20EFRX-68EUZN0 82.00A82>    at scbus8 target 0 lun 0 (pass3,ada2)
<WDC WD20EFRX-68EUZN0 82.00A82>    at scbus9 target 0 lun 0 (pass4,ada3)
<Imation Nano Pro PMAP>            at scbus11 target 0 lun 0 (pass5,da0)
<Imation Nano Pro PMAP>            at scbus12 target 0 lun 0 (pass6,da1)

So I guess the controller is claiming to have a SATA 1.x device it calls "Marvell Console", with a serial number that's the same on every system. Over here it was speculated to be the RAID configuration interface, which seems reasonable.

I guess there's no smoking gun here.

 

[Mirfster]

speculated to be the RAID configuration interface

For some reason I find that a bit scary... ;)

 

[Robert Trevellyan]

According to the documentation

The Rocket 640L is coming with another "Console" device which is using for communicated with the "MSU" utility software"[sic]

and there are directions for installing the MSU utility in Windows.

 

[Robert Trevellyan]

I forgot to post ... Sunday morning's scrub completed without any issues.

 

[Mirfster]

Nice

 

[AlainD]

Thanks for letting us know you're test results.
Seems, a nice way to go from 6 sata ports to 10 ports.

 

[nightshade00013]

Ok, but just so you know, if anything bad happens, the ZFS police are coming to take your Pony (see my signature for reference):p

So that means with 7 drives in a raidZ3 I get a pony and a cow or is that a pony and a half.

 

[Robert Trevellyan]

raidZ3

Just a pony, but it's a really reliable pony.

 

[nightshade00013]

Just a pony, but it's a really reliable pony.

I guess that works then I really like reliable.

 

[Robert Trevellyan]

So, it finally happened, but not the way I had imagined.

Background
I've been experimenting with Ubuntu Server + webmin as an alternative to FreeNAS. Any further discussion of that belongs in Off Topic. The relevant part is that I had a pool made of a single mirrored pair of old Seagate 500GB disks in a PowerEdge T110, and I decided to transfer it to a pair of 2TB disks in my PowerEdge T20. The HighPoint Rocket 640L is installed in the T20.

Setup
In the T20, I connected the Ubuntu boot device (20GB Avant SSD) and the 500GB Seagate disks to the 640L. The pair of 2TB WD Reds were connected to the internal ports. Then I issued a zfs replace to replace one of the Seagates with a WD Red. Then I had the bright idea of starting the 2nd resilver at the same time, so I kicked off another zfs replace. Pretty soon, zpool status was reporting a resilver rate around 100Mbps. Felt pretty smug...

Oh Nos
I wanted to check ... something, so I clicked a link in the webmin interface, but it didn't respond. Then I noticed the resilver was stuck at around 62GB done, even though the percentage complete kept rising. Then I saw the console was spewing I/O errors for the boot device. I tried to log in at the console, but got booted out. Eventually, the resilver reported 100% complete, but zpool status showed a faulted pool with 2 devices being replaced (this part is a bit hazy). I issued poweroff, but it failed. I tried reboot, but it hung. After a while, I did a forced poweroff.

What Now?
I moved the Seagates to the remaining internal ports, leaving only the boot device on the 640L. The system booted, but zpool import said the pool was corrupted, still replacing, etc. With nothing to lose, I issued the import and ... zfs started the resilver from 0%. Less than an hour later it was done, everything hunky-dory, with the pool sitting pretty on the WD Reds.

What Happened?
My interpretation is that the 640L barfed on the heavy workload. I've never had a problem with it before, but I never asked it to deliver so much data so quickly before. If I'd been doing something other than resilvering a mirror to new disks (almost 100% read load on the 640L), the outcome could have been ugly. I'm also pretty impressed with the way the ext4 filesystem on the boot device came through unscathed.

Of course, it's possible that Ubuntu did something wrong, and the 640L is innocent, but I doubt it.

 

[SweetAndLow]

So, it finally happened, but not the way I had imagined.

Background
I've been experimenting with Ubuntu Server + webmin as an alternative to FreeNAS. Any further discussion of that belongs in Off Topic. The relevant part is that I had a pool made of a single mirrored pair of old Seagate 500GB disks in a PowerEdge T110, and I decided to transfer it to a pair of 2TB disks in my PowerEdge T20. The HighPoint Rocket 640L is installed in the T20.

Setup
In the T20, I connected the Ubuntu boot device (20GB Avant SSD) and the 500GB Seagate disks to the 640L. The pair of 2TB WD Reds were connected to the internal ports. Then I issued a zfs replace to replace one of the Seagates with a WD Red. Then I had the bright idea of starting the 2nd resilver at the same time, so I kicked off another zfs replace. Pretty soon, zpool status was reporting a resilver rate around 100Mbps. Felt pretty smug...

Oh Nos
I wanted to check ... something, so I clicked a link in the webmin interface, but it didn't respond. Then I noticed the resilver was stuck at around 62GB done, even though the percentage complete kept rising. Then I saw the console was spewing I/O errors for the boot device. I tried to log in at the console, but got booted out. Eventually, the resilver reported 100% complete, but zpool status showed a faulted pool with 2 devices being replaced (this part is a bit hazy). I issued poweroff, but it failed. I tried reboot, but it hung. After a while, I did a forced poweroff.

What Now?
I moved the Seagates to the remaining internal ports, leaving only the boot device on the 640L. The system booted, but zpool import said the pool was corrupted, still replacing, etc. With nothing to lose, I issued the import and ... zfs started the resilver from 0%. Less than an hour later it was done, everything hunky-dory, with the pool sitting pretty on the WD Reds.

What Happened?
My interpretation is that the 640L barfed on the heavy workload. I've never had a problem with it before, but I never asked it to deliver so much data so quickly before. If I'd been doing something other than resilvering a mirror to new disks (almost 100% read load on the 640L), the outcome could have been ugly. I'm also pretty impressed with the way the ext4 filesystem on the boot device came through unscathed.

Of course, it's possible that Ubuntu did something wrong, and the 640L is innocent, but I doubt it.

Good post and thanks for following up with the story. This kind of information is great so people can make good decisions.

Sent from my Nexus 5X using Tapatalk