|
PerrineClostermann posted:So I installed FreeNAS on my old Core 2 Duo machine, with 6gb RAM. my drives are plugged into SATA2 ports. The pool is configured in RAIDZ2. I'm getting 30MB/s transfer speeds, would that be pretty typical for this kind of setup? mayodreams posted:What NIC are you using? lovely onboard Reltek devices don't handle a lot of throughput well. I bought Intel NICs for all my devices for this reason. Easiest way to figure out what's causing the bottleneck is to start benchmarking (the FreeNAS handbook has a section on various tools you can use for this). This will also give you a baseline for future reference if something goes wrong. EDIT: Quoted the original question because of new page. BlankSystemDaemon fucked around with this message at 14:56 on Feb 24, 2016 |
#
?
Feb 24, 2016 14:42
|
|
|
|
| # ? May 15, 2024 23:03 |
|
|
What's the recommended raid configuration for a Synology NAS? I've got a Synology ds416j with three 4tb drives in RAID 5 currently, but I've thought about picking up another and converting to RAID 6. There is also the option of using Synology Hybrid Raid (SHR). What is generally considered the best option?
|
|
#
?
Feb 24, 2016 15:22
|
|
|
From what I can tell, they're Marvell controllers
|
|
#
?
Feb 24, 2016 15:32
|
|
|
PerrineClostermann posted:So I installed FreeNAS on my old Core 2 Duo machine, with 6gb RAM. my drives are plugged into SATA2 ports. The pool is configured in RAIDZ2. I'm getting 30MB/s transfer speeds, would that be pretty typical for this kind of setup? Is that on read or write? For writing, that may just be how fast you're able to calculate parity. If it's on reads though, then it's probably a network controller issue.
|
|
#
?
Feb 24, 2016 15:52
|
|
|
blk96gt posted:What's the recommended raid configuration for a Synology NAS? I've got a Synology ds416j with three 4tb drives in RAID 5 currently, but I've thought about picking up another and converting to RAID 6. There is also the option of using Synology Hybrid Raid (SHR). What is generally considered the best option? SHR is raid 5 or 6 internally (depending on amount of parity drives). You should be using SHR
|
|
#
?
Feb 24, 2016 17:01
|
|
|
Skandranon posted:Is that on read or write? For writing, that may just be how fast you're able to calculate parity. If it's on reads though, then it's probably a network controller issue. This is on write. Could be, it is an old e6750, running at its stock speed of 2.66GHz. But then again, FreeNAS is reporting plenty of spare CPU power. Looks like 20% CPU usage. Individual disks are at 8-10MB/s, according to FreeNAS...
|
|
#
?
Feb 24, 2016 17:17
|
|
|
PerrineClostermann posted:This is on write. Could be, it is an old e6750, running at its stock speed of 2.66GHz. But then again, FreeNAS is reporting plenty of spare CPU power. Looks like 20% CPU usage. Individual disks are at 8-10MB/s, according to FreeNAS... It may only be able to use a single core for parity calculations, and you are railing that core.
|
|
#
?
Feb 24, 2016 17:42
|
|
|
Skandranon posted:It may only be able to use a single core for parity calculations, and you are railing that core. If that were true, wouldn't he be at 40%+ CPU usage? It's a dual core.
|
|
#
?
Feb 24, 2016 17:47
|
|
|
Don Lapre posted:SHR is raid 5 or 6 internally (depending on amount of parity drives). You should be using SHR Looks like I'll have to wipe everything on the drives and start from scratch if I want to swap to SHR. Not that big of a deal since I haven't dropped a ton of stuff on there yet.
|
|
#
?
Feb 24, 2016 18:14
|
|
|
Twerk from Home posted:If that were true, wouldn't he be at 40%+ CPU usage? It's a dual core. Maybe? CPU might be starved for data.
|
|
#
?
Feb 24, 2016 18:23
|
|
|
blk96gt posted:Looks like I'll have to wipe everything on the drives and start from scratch if I want to swap to SHR. Not that big of a deal since I haven't dropped a ton of stuff on there yet. Yea, do it now. Then if you move to a larger synology you can just migrate your discs over and keep expanding.
|
|
#
?
Feb 24, 2016 20:15
|
|
|
Don Lapre posted:Yea, do it now. Then if you move to a larger synology you can just migrate your discs over and keep expanding. I am assuming qnap has a similar setup, right? Like would I be ok to start off in a 4 bay or larger with 2 drives and just add them as per my whims and discretionary spending? Strongly considering a TS-x53A
|
|
#
?
Feb 24, 2016 20:27
|
|
|
No idea on qnap. With a synology you can move to new units once you need more bays and all your stuff stays the same, you can even migrate a legit synology to a home built xpenology.
|
|
#
?
Feb 24, 2016 20:57
|
|
|
Skandranon posted:Maybe? CPU might be starved for data. Then the CPU wouldn't be the bottleneck. Even Marvell controllers should do better than 30mb/s, and my router is pretty decent (Asus NT66 or whatever, the comically expensive one).
|
|
#
?
Feb 24, 2016 21:26
|
|
|
PerrineClostermann posted:Then the CPU wouldn't be the bottleneck. Even Marvell controllers should do better than 30mb/s, and my router is pretty decent (Asus NT66 or whatever, the comically expensive one). The CPU isn't, but some part of the parity calculation probably is. Again, if you are able to get significantly more during read operations, then the problem is something to do with how it's calculating/writing parity.
|
|
#
?
Feb 24, 2016 21:48
|
|
|
Having the hardest time getting some numbers could anyone give me an approximate of how much they paid for a Netapp FAS2520 about 4tb
|
|
#
?
Feb 26, 2016 17:01
|
|
|
Don Lapre posted:No idea on qnap. With a synology you can move to new units once you need more bays and all your stuff stays the same, you can even migrate a legit synology to a home built xpenology. What if I only have 1 drive and I put it in JBOD mode, can I go SHR when I put another one in without having to format?
|
|
#
?
Feb 26, 2016 19:26
|
|
|
socialsecurity posted:Having the hardest time getting some numbers could anyone give me an approximate of how much they paid for a Netapp FAS2520 about 4tb You want this thread - http://forums.somethingawful.com/showthread.php?threadid=2943669
|
|
#
?
Feb 26, 2016 19:28
|
|
|
Skandranon posted:The CPU isn't, but some part of the parity calculation probably is. Again, if you are able to get significantly more during read operations, then the problem is something to do with how it's calculating/writing parity. ...This is weird. Now I'm getting speeds over 60MB/s. Nothing's really changed. I'm also getting read speeds close to 100MB/s or thereabouts, my last transfer was yesterday.
|
|
#
?
Feb 26, 2016 19:30
|
|
|
PerrineClostermann posted:...This is weird. Now I'm getting speeds over 60MB/s. Nothing's really changed. My six drive Raid-Z2 actually WRITES faster over Samba than it reads, before the NIC hosed up and would only work at 10BaseT. It was near 100 MB/s on writes from my Win7 desktop, and capped at 75 MB/s on reads. I never did figure that out.
|
|
#
?
Feb 26, 2016 19:57
|
|
|
uhhhhahhhhohahhh posted:What if I only have 1 drive and I put it in JBOD mode, can I go SHR when I put another one in without having to format? Technically you'll still need to reformat the drive your data is currently on, however there is a workaround: 1. Instead of adding the new drive to your existing JBOD volume create a new SHR volume using the new drive. 2. Move all of your data from the JBOD volume to the SHR volume 3. Delete the JBOD volume 4. Add the old drive to the new SHR volume
|
|
#
?
Feb 26, 2016 19:58
|
|
|
fletcher posted:Wow, sounds like a good home for some hard drives. Crazy that they all died! Were all of the drives purchased around the same time? I bought a bunch of Samsung HD154UI 1.5TB drives back in the day and ended up with three or four that were within about 30 serial numbers of each other (a couple nearly sequential). While most of them held up for a few years (I did have at least one die prematurely). The ones that were all closest in serial number all died the exact same way, with head crashing leading to grinding the actual disk. Must've been a barely-defective run of them.
|
|
#
?
Feb 26, 2016 20:11
|
|
|
My primary file server at home is an 8-drive raid-5 of at least 6 year old Seagate Barracuda ST31000340NS on an Adaptec 5805Z, the famous ones that if you didn't update the firmware past the SN04 revision they would randomly poo poo the array. I like to live dangerously but not too dangerously, I have an identical server that boots weekly and mirrors everything over. I really should get new drives, but I still have a stack of 4 that I can use as replacements. 
|
|
#
?
Feb 26, 2016 20:31
|
|
|
RAID5 is scary 
|
|
#
?
Feb 26, 2016 21:21
|
|
|
IOwnCalculus posted:Were all of the drives purchased around the same time? I bought a bunch of Samsung HD154UI 1.5TB drives back in the day and ended up with three or four that were within about 30 serial numbers of each other (a couple nearly sequential). While most of them held up for a few years (I did have at least one die prematurely). The ones that were all closest in serial number all died the exact same way, with head crashing leading to grinding the actual disk. Must've been a barely-defective run of them. I still have 3 of those HD154UI running. Nothing really important on them thankfully.
|
|
#
?
Feb 27, 2016 00:49
|
|
|
I had two left that I pulled and let sit on a shelf for a few months, both of them kicked the bucket once I fired them back up. Can't complain, they had a poo poo load of hours.
|
|
#
?
Feb 27, 2016 01:49
|
|
|
The "problem" with RAID-Z and read performance are the full stripe writes, every filesystem block is a whole RAID-Z stripe. So when reading a single block, it has to touch all the data disks to assemble the block. With linear reads, ZFS' prefetcher notices and starts reading ahead, taking rotational and seeking latencies of the disks out of the picture. With random IO however, you're just hosed and your IO is subject to said latencies. If you want maximum read performance (for cold data) in ZFS, you have to go with mirrors/RAID-10.
|
|
#
?
Feb 27, 2016 23:35
|
|
|
So by doing some testing, it seems my box has...interesting read write speeds: Read: 503376780 bytes transferred in 1.483086 secs (339411702 bytes/sec) [300MB/s] Write: 503376780 bytes transferred in 1.102334 secs (456646423 bytes/sec) [450MB/s] ...Being able to write faster than I read is...strange, to say the least. The speeds were tested using DD to copy video files to and from itself, RAM, etc. Anyway, this is in contrast to my speeds on CIFS, where I write at...90MB/s. Is this really just CIFS' overhead? Is there a better way for my windows machine to access my NAS?
|
|
#
?
Mar 3, 2016 21:17
|
|
|
90MB/s is 720Mbps - e.g. nearly gigabit network speeds if you ignore overheads. You NIC might be lovely but if NFS isn't noticeably faster then that's probably the quickest you can go.
|
|
#
?
Mar 3, 2016 21:33
|
|
|
PerrineClostermann posted:So by doing some testing, it seems my box has...interesting read write speeds: SMB is network so if you use ZFS it will use Sync'ed writes/ For laughs you can turn them off: zfs set sync=disabled datasetname. SMB should be faster my Windows 8 machine pulls 111 MB/Sec easily.
|
|
#
?
Mar 4, 2016 07:39
|
|
|
Does the Fractal 304 often go on sale to $60? http://www.newegg.com/Product/Product.aspx?Item=N82E16811352027 I was checking out the stuff I'd need to convert my lil microATX htpc into a FreeNAS box [6 bay case, PCIe SATA, and.. the drives], and that seemed like a good enough deal to jump on right now.
|
|
#
?
Mar 4, 2016 18:35
|
|
|
Looks like it is pretty regular that that price - http://camelcamelcamel.com/Fractal-Design-Mini-ITX-Computer-FD-CA-NODE-304-BL/product/B009LHF4FO?context=browse
|
|
#
?
Mar 4, 2016 18:51
|
|
|
Mr Shiny Pants posted:SMB should be faster my Windows 8 machine pulls 111 MB/Sec easily.
|
|
#
?
Mar 5, 2016 02:24
|
|
|
Combat Pretzel posted:Same here. 110MB/s even while switched over the lovely switch inside my DSL modem. Well what the hell then. Cat 5e, gigabit ports, intel NIC/marvell NIC, asus nt66 router.
|
|
#
?
Mar 5, 2016 05:53
|
|
|
Could try checking the network throughput using iperf. It's a testing tool that can be run as server on one side and as client on the other, which then just hammer the connection to test TCP/IP transmission speeds. It's available in the repos on *nix and there's a Windows build on their site. Should top out at 970Mbit/s, at least it does here between two Intel NICs. --edit: Just to see if things are in order or not. Using the RealTek on my old desktop mainboard for instance resulted in 470MBit/s maximum between it and an Intel, even when connected directly with a crossover wire. Combat Pretzel fucked around with this message at 17:54 on Mar 5, 2016 |
|
#
?
Mar 5, 2016 17:45
|
|
|
Not quite relevant for this performance scenario but one of the things I noticed was that encryption or compression latency is a huge limiting factor in certain protocols and environments. I turned on rsync -avz and got maybe 30 MBps transfers from within ESXi VM to VM (they're supposed to be 10 GbE connections with 9k MTU/MSS using vmxnet3 vNICs). I took off the z and throughput jumped to 300 MBps+
|
|
#
?
Mar 5, 2016 18:35
|
|
|
Am I remembering correctly that someone said not to buy HDDs from Newegg here due to an unusually high rate of failure? I'm going to pick up a 4TB HGST drive and it's currently $30 cheaper st Newegg. If $30 is the difference between a reliable and an unreliable drive, I'll buy it from Amazon.
|
|
#
?
Mar 5, 2016 23:04
|
|
|
Newegg packs them fine
|
|
#
?
Mar 6, 2016 01:29
|
|
|
If I recall it was Amazon who sometimes packed hdds poorly if they weren't already in a box.
|
|
#
?
Mar 6, 2016 01:54
|
|
|
|
| # ? May 15, 2024 23:03 |
|
|
Anecdotal, but there's a whole bunch of Reddit posts with pictures from Newegg showing really poor packing around drives, including one guy that did 3 RMAs in a row on the same order, and got drives floating loose in a box every time. That said, Newegg's always done right by me in terms of drives, but I wouldn't be surprised if YMMV based on what distribution facility your stuff ships from.
|
|
#
?
Mar 6, 2016 03:07
|
|
