April 4, 201016 yr I've been lurking on the forums for a couple weeks now, reading old posts and planning a potential build. I have come down to two options, and am somewhat torn between them. - option 1 involves buying a Supermicro X7SBE motherboard for unRAID since it is one sweet board with plenty of upgrade slots, onboard video, dual Intel nics, and IPMI. My main computer would get a boost up to an e8400 Core 2 Duo. - option 2 means buying a new i5 cpu, mobo, and DDR3 ram for my main rig, and sticking unRAID with the leftover bits. (The unRAID server would then be ASUS P5KPL-CM, 2x2GB DDR2 ram, e5400 proc.) Specifically, I'll be forced to run two AOC-SAT2-MV8 boards in the two PCI slots, whereas Supermicro intended that board to be used in PCI-X slots, and the newer units are PCIe. If I fully populate the two SAT2 cards, I'd be running 16 drives over a standard PCI bus. Is that really a wise idea, or would I be better to go with Option 1, and put my desire for a brand new i5 desktop on hold? Typical use would be one 1080p video stream to the home theater. Thanks for the feedback.
April 5, 201016 yr would I be better to go with Option 1, and put my desire for a brand new i5 desktop on hold? Typical use would be one 1080p video stream to the home theater. Personally, I wouldn't. Your "leftover bits" are more than what's needed for a very decent unRAID server. If that's your typical use, the server won't ever break a sweat.
April 5, 201016 yr Do you have a need for 16 drives now? If not how many do you need today. Chances are your left over bits will do fine for an unRAID server in the meantime. If I saw correctly it has a 16x PCIe slot. In that slot you can fit in a 4 port SATA card or a 8 port Supermicro SASLP card. Either of these could be moved over to an X7SBE later on.
April 5, 201016 yr Author I would have an immediate need for 9 or 10 SATA slots, and 12 before the end of the year is not too far-fetched. I've been reading about some possible incompatibilities with x16 PCIe slots and non-video cards. I suppose I could try the SASLP card and see if it worked - that would mean one SASLP and one SAT2 which should perform ok. As you point out purko, I'm not going to be stressing the server all that hard. I just did a new search on benchmarks, and read an interesting post by aiden. My interpretation is that using unRAID, a typical read is only going to be accessing one drive at a time, and a typical write is only going to be two drives, ie the destination + parity. Is that right? From that perspective, even 16 drives on PCI is likely ok as long as I keep my parity drive on the mobo. Hmm.
April 5, 201016 yr My interpretation is that using unRAID, a typical read is only going to be accessing one drive at a time, and a typical write is only going to be two drives, ie the destination + parity. Is that right? From that perspective, even 16 drives on PCI is likely ok as long as I keep my parity drive on the mobo. Hmm. Yep!
April 5, 201016 yr yeah shouldnt be a problem jus tyour parity check will take about 5000 hours or more.
April 5, 201016 yr yeah shouldnt be a problem jus tyour parity check will take about 5000 hours or more. seriously, you've got no fricken idea what you are talking about.
April 5, 201016 yr I just did a new search on benchmarks, and read an interesting post by aiden. My interpretation is that using unRAID, a typical read is only going to be accessing one drive at a time, and a typical write is only going to be two drives, ie the destination + parity. Is that right? From that perspective, even 16 drives on PCI is likely ok as long as I keep my parity drive on the mobo. Hmm. Also, there is a PCI x1 slot, you could get a 2 card controller for that. (Then test it in the x16 slot). Your parity check speeds will take some time. Put the largest drives on the fastest ports. Make sure you invest in a UPS.
April 5, 201016 yr yeah shouldnt be a problem jus tyour parity check will take about 5000 hours or more. seriously, you've got no fricken idea what you are talking about. leta assume 15 drives (lets put parity somewhere else) sharing 133MB/sec ('standard pci'), thats 8MB/sec, and your drive is 2000000MB. at ideal speeds it would br 4100 hours. if there is something wrong with how ive worked this out, enlighten me.
April 5, 201016 yr yeah shouldnt be a problem jus tyour parity check will take about 5000 hours or more. seriously, you've got no fricken idea what you are talking about. leta assume 15 drives (lets put parity somewhere else) sharing 133MB/sec ('standard pci'), thats 8MB/sec, and your drive is 2000000MB. at ideal speeds it would br 4100 hours. if there is something wrong with how ive worked this out, enlighten me. I don't think it would be that slow. I also do not believe the math works quite that way. I think each drive would be read as fast as it could be read. The delay comes from only being able to read one drive at a time and then waiting for the PCI bus to settle. I would love to hear from others who have this card in PCI slots with many drives.
April 5, 201016 yr yeah shouldnt be a problem jus tyour parity check will take about 5000 hours or more. seriously, you've got no fricken idea what you are talking about. leta assume 15 drives (lets put parity somewhere else) sharing 133MB/sec ('standard pci'), thats 8MB/sec, and your drive is 2000000MB. at ideal speeds it would br 4100 hours. if there is something wrong with how ive worked this out, enlighten me. There is. 4100 is the number of minutes. Number of hours assuming 8MB/s throughput is 69~70 theoretically.
April 6, 201016 yr I was using this card along with a Promise SATA300 TX4 card, both in PCI slots. I had four drives connected to onboard SATA controllers with all eight of the AOC-SAT2-MV8 slots occupied. The remaining two drives out of a total of fourteen were connected to the Promise controller. Parity checks typically ran on the order of about 36-40 hours, with the largest drives being 1.5TB (six, including parity). The rest of the drives were all 750GB. I have since switched out both the AOC-SAT2-MV8 and the Promise controller for a Supermicro 8-port PCI-E controller and an Adaptec 4port PCI-E controller. I'm down to thirteen drives total with a 2TB parity drive. The twelve data drives are all 1.5TB. Parity checks now run about 6-1/2 to 7 hours.
April 6, 201016 yr I was using this card along with a Promise SATA300 TX4 card, both in PCI slots. I had four drives connected to onboard SATA controllers with all eight of the AOC-SAT2-MV8 slots occupied. The remaining two drives out of a total of fourteen were connected to the Promise controller. Parity checks typically ran on the order of about 36-40 hours, with the largest drives being 1.5TB (six, including parity). The rest of the drives were all 750GB. How did you have this set up? Like this? Onboard 4x 1.5TB (including parity?) Supermicro AOC-SAT2-MV8 2x 1.5TB 6x 750GB Promise SATA 300 TX4 2x 750GB
April 7, 201016 yr Author Thanks for the good info here. PCI should be fine for my needs afterall, and I'm not concerned if parity checks take a couple days, as long as I can still access the files in the meantime. Strangely, the last thing I'll really miss is IPMI, but the X7SBE is a little pricey just for that!
April 8, 201016 yr Thanks for the good info here. PCI should be fine for my needs afterall, and I'm not concerned if parity checks take a couple days, as long as I can still access the files in the meantime. Strangely, the last thing I'll really miss is IPMI, but the X7SBE is a little pricey just for that! Just a comment, for the above configuration (tentatively how captain_video set-up his server), doing the math, parity check is supposed to only take ~18 hours while the actual parity check took 36~40 hours - pretty much double. Assuming 15 2TB drives on a couple of AOC-SAT2-MV8 cards, you're probably looking at one week for a parity check. If you're using a number of drives, I highly suggest you move them off the PCI bus and use a PCI Express SATA adapter. I would have an immediate need for 9 or 10 SATA slots, and 12 before the end of the year is not too far-fetched. With that many drives on an ASUS P5KPL-CM motherboard, I'd suggest the following set-up: Onboard 4x Drives (including parity) Supermicro AOC-SASLP-MV8 (PCIe x16) 8x Drives There goes your 12 drives and even with all 2TB drives, you're still only limited by spindle speed. Assuming you're only going with 10 drives initially, I'd put 8 data drives on the AOC-SASLP-MV8 so there's only 2 drives onboard (1x parity, 1x data). One thing I've noticed is the unRAID server is really slow during parity checks - enough to make Blu-ray playback stutter. Not sure if it's just my set-up with that problem but that's what I've experienced.
April 9, 201016 yr its probably your setup. i find my parity check just slows down, when i steal disk speed for something else.
April 9, 201016 yr its probably your setup. i find my parity check just slows down, when i steal disk speed for something else. Perhaps. What unRAID version are you using? I know mine's long overdue for an upgrade but I'd like to build an all hotswap server before I do any maintenance on my current one.
April 9, 201016 yr Consider this. A 2 port mass cool x1 PCIe card is in the $20-$25 range. If you put the parity and a data drive on this card, then the 6 1.5tb drives will have fast access. Divide up the remaining drives between the MV8 and the TX4. As I've mentioned, the TX4 is no slouch. I was getting 80MB/s consistent throughput using one. on a 66Mhz Bus I was able to do two transfers and still get 80MB/s transfer rate of data from the drive with dd.
April 9, 201016 yr Just a comment, for the above configuration (tentatively how captain_video set-up his server), doing the math, parity check is supposed to only take ~18 hours while the actual parity check took 36~40 hours - pretty much double. Assuming 15 2TB drives on a couple of AOC-SAT2-MV8 cards, you're probably looking at one week for a parity check. 1.5 TB /13.3MB/s /60 /60 = 31.3 hrs? 1.5TB biggest disk on PCI 13.3MB/s = 133MB/s /10 HDD on bus /60 mins /60 hours. Our patient hero is suggesting max 12 disks on the PCI bus, with 4 onboard ports being used. 2TB / 11MB/s /60 60 = 50hrs or two days. With ten drives that are "needed now". 6 hdds on PCI. 2TB / 22MB/s /60 /60 = 25 hrs. I think the ICH7 is a 33Mhz PCI bus (Asutek motherboard) so the TX4s wont run at 66Mhz which would drop the parity check time down to about a working day. This assumes there is a 2tb drive on the PCI bus. As weebo said put the big drives and the parity drive on the motherboard sata slots. Pickup a cheap PCI-e x1 card to move another couple of drives onto the PCIe bus. When you get the x1 PCIe card, try it in the x16 slot. If it works get another and move another two drives onto the PCIe x16 slot. That'll give you eight "fast" slots and put the remaining HDDs onto the PCI bus. With only a couple on the PCI you should expect reasonable parity check speeds.
April 9, 201016 yr 1.5 TB /13.3MB/s /60 /60 = 31.3 hrs? 1.5TB biggest disk on PCI 13.3MB/s = 133MB/s /10 HDD on bus /60 mins /60 hours. Our patient hero is suggesting max 12 disks on the PCI bus, with 4 onboard ports being used. You're forgetting something. He has a mix of 750GB drives. Once those are done, you're left with just 2 drives remaining on the PCI bus making parity check go faster. 750GB / 13.3MB/s / 3600hr/s + 750GB / 66.5MB/s / 3600hr/s = 18.8hrs Let's assume that by the end of the 750GB he begins to get disk bound (I remember seeing some high capacity green drives that have minimum read speeds of 50MB/s) 750GB / 13.3MB/s / 3600hr/s + 750GB / 50MB/s / 3600hr/s = 19.8hrs Albeit, considering my own parity checks, I have no idea why his is taking much longer than expected (unless, of course, he's reading/writing to the drives at the same time). When I only had 6 drives all on the ICH8R, parity check was only limited by disk speed. Now, that I have 9 drives, I'm getting ~65MB/s but I blame that on either the JMicron or Silicon Image controller. Maybe both?
April 9, 201016 yr 1.5 TB /13.3MB/s /60 /60 = 31.3 hrs? 1.5TB biggest disk on PCI 13.3MB/s = 133MB/s /10 HDD on bus /60 mins /60 hours. Our patient hero is suggesting max 12 disks on the PCI bus, with 4 onboard ports being used. You're forgetting something. He has a mix of 750GB drives. Once those are done, you're left with just 2 drives remaining on the PCI bus making parity check go faster. 750GB / 13.3MB/s / 3600hr/s + 750GB / 66.5MB/s / 3600hr/s = 18.8hrs Let's assume that by the end of the 750GB he begins to get disk bound (I remember seeing some high capacity green drives that have minimum read speeds of 50MB/s) 750GB / 13.3MB/s / 3600hr/s + 750GB / 50MB/s / 3600hr/s = 19.8hrs Albeit, considering my own parity checks, I have no idea why his is taking much longer than expected (unless, of course, he's reading/writing to the drives at the same time). When I only had 6 drives all on the ICH8R, parity check was only limited by disk speed. Now, that I have 9 drives, I'm getting ~65MB/s but I blame that on either the JMicron or Silicon Image controller. Maybe both? which sil? ive found my jmb383 just as good as my nforce500 or marvell controllers (1000MB/sec cache and 120MB read).
April 9, 201016 yr which sil? ive found my jmb383 just as good as my nforce500 or marvell controllers (1000MB/sec cache and 120MB read). Whatever's on the ABIT AB9 Pro.
April 10, 201016 yr 750GB / 13.3MB/s / 3600hr/s + 750GB / 66.5MB/s / 3600hr/s = 18.8hrs How about this... 750GB / 13.3MB/s / 3600hr/s *2 + 750GB / 66.5MB/s / 3600hr/s = @34hrs *2 is for two pci cards. pci can burst to 133 MB/s. This assumes one device(one read requst getting 4 bytes of data), with two devices this could easily halve. bus request, data xfer, bus release etc...
April 10, 201016 yr How about this... 750GB / 13.3MB/s / 3600hr/s *2 + 750GB / 66.5MB/s / 3600hr/s = @34hrs *2 is for two pci cards. pci can burst to 133 MB/s. This assumes one device(one read requst getting 4 bytes of data), with two devices this could easily halve. bus request, data xfer, bus release etc... That's quite possible. If that's the case, using a shared PCI bus for 12 drives really isn't sounding like a good idea. By the way, just did some dd testing on individual drives and both Silicon Image and JMicron controllers posted over 100MB/s reads. I'm thinking the JMicron SATA controller might be on the PCI bus which would mean two drives are sharing 133MB/s throughput - in-line with the 65MB/s parity checks I'm seeing.
Archived
This topic is now archived and is closed to further replies.