On 4/1/2014 2:09 PM, Barrett Lewis wrote: > I have a dedicated, consumer hw, media/file server with a 6 drive > raid6 of 2tb drives, all plugged directly into the sata ports on my > motherboard, an asrock z77. There are 6 models of the Asrock z77. All but one contain a PCH heatsink designed to look cool rather than properly cool the chip. The Asrock z77 Extreme 11 has a fan so is an exception, and also has an onboard 8 port LSI SAS controller (9211-8i), so I assume you do not have the Extreme 11. > A while back I had a problem which seemed > like cascade failure of drives, but Stan Hoeppner and Phil Turmel > helped me to figure out it was a PSU having gone bad and delivering > dirty power. > > After replacing the PSU things worked fine, or so I thought. At some > point I noticed I have quite a bit of trouble making it through a > resync without the machine locking up. When I realized it wasn't tied > to a resync in particular but any extended heavy I/O, I lowered the > sync_speed_max to 10,000, I was able to get through a repair (no > mismatches found!). With consumer PC hardware random lockups occurring only under heavy disk IO are most often the result of thermal buildup in the PCH (Northbridge) chip. This can occur when all the drives are connected to its SATA ports as in your case, but it can also occur when using one or more SAS/SATA HBAs if the PCIe slots are connected through the PCH. The odds are very good that your lockups are a result of the poor PCH heatsink design on the Asrock boards exacerbated by insufficient case airflow across the heatsink. What case is this z77 board in? Be specific please so I can pull up the schematic. Regardless of case the solution is straightforward and inexpensive: install a low profile solid copper active cooler, such as this one: http://www.frozencpu.com/products/6717/vid-102/Enzotech_SLF-1_Forged_Copper_Northbridge_Southbridge_Low-Profile_Heatsink.html?tl=g40c16s501 The SLF-1 has 53-59mm hole spacing. Asrock doesn't provide such information in their manual and after 30 minutes I can't find forum posts or other sources presenting this info. Measure your PCH heatsink mounting hole spacing before ordering. If it's less than 53mm center-to-center you need the SLF-30, and if it's more than 59mm you need the SLF-40. If you think your case airflow over the PCH is actually greater than zero you can go with the CNB-R1 passive unit which has 3 mounting rings to fit all hole spacings. But with it you lose two expansion slots. Here's the product lineup: http://www.enzotechnology.com/air_cooling.htm There are other brands. Enzo products are solid copper and compact, with these 3 fan models fitting under your PCIe cards. You lose no PCI slots as with nearly all other chipset coolers. I recommend them because they are high quality and work well, which is why they are also more expensive than most others. That being the case, ~$35 including shipping is a small sum to part with to eliminate the lockups. > I'm guessing that the motherboard has some problem (perhaps > originating from the bad PSU?), and I want to switch to a dedicated > HBA card to make this more modular. The one glaring problem is the woefully inadequate PCH heatsink. Replacing it as suggested will very likely eliminate the lockups, for about 1/8th the cost of a discrete LSI HBA. And if it doesn't you will still have increased the lifespan of the PCH chip by at least a couple of years due to lowering operating temperature by 10-15°C or more. > Stan had suggested the LSI SATA/SAS 9211-8i in many threads in the > archives. If I use this card as my HBA, is there any particular > motherboard which would be better suited than others? Wait and cross this bridge later. If it turns out this board has other problems that we can't identify and fix, there's a micro-ATX Intel server board with 6 SATA-2 ports on the PCH, socket LGA 1155, dual Intel GbE ports, integrated video, etc for ~$160 at Newegg. Your CPU, RAM, and drives will drop right in, and you won't have to spend another $200 on the LSI. It'll save you ~$150 overall compared to a consumer board+LSI. Cheers, Stan -- To unsubscribe from this list: send the line "unsubscribe linux-raid" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
