00:26:50AmAnd0A quits [Read error: Connection reset by peer]
00:27:13icedice quits [Client Quit]
00:31:10AmAnd0A joins
00:35:59adia quits [Client Quit]
00:55:36wessel1512 quits [Quit: Ping timeout (120 seconds)]
00:55:59wessel1512 joins
01:20:00benjinsm is now known as benjins
01:24:14wessel1512 quits [Client Quit]
01:24:38wessel1512 joins
01:24:41icedice (icedice) joins
01:46:58AmAnd0A quits [Ping timeout: 252 seconds]
01:47:37AmAnd0A joins
02:21:29<andrew>I'm trying to figure out what width of RAIDZ1 I should use for an SSD pool I'm thinking of building
02:21:49<andrew>I keep seeing this "power of two plus parity" rule being thrown around, but I also see conflicting information
02:22:01<andrew>is there any reason why I *shouldn't* use a 4-wide RAIDZ1?
02:27:31PredatorIWD joins
02:31:07PredatorIWD quits [Client Quit]
02:39:28Rhodezzz quits [Ping timeout: 265 seconds]
02:39:50<TheTechRobo>Whats that org that operates their own DNS and lets you register free domain names that only work if you use their nameservers?
02:41:00<@JAA>OpenNIC? Though there are others which are similar, that's the big one.
02:41:31icedice quits [Client Quit]
02:43:41<TheTechRobo>JAA: Ah yes I believe that's the one. Thanks!
03:13:16icedice (icedice) joins
03:14:22icedice quits [Client Quit]
03:24:34icedice (icedice) joins
04:33:03Danielle quits [Read error: Connection reset by peer]
04:35:16BlueMaxima quits [Read error: Connection reset by peer]
05:23:34michaelblob quits [Read error: Connection reset by peer]
06:22:27nicolas17 quits [Client Quit]
06:22:40nicolas17 joins
06:45:19Zaxoosh quits [Remote host closed the connection]
06:45:37spirit quits [Quit: Leaving]
07:07:50Arcorann (Arcorann) joins
07:17:10icedice2 (icedice) joins
07:20:17icedice quits [Ping timeout: 265 seconds]
07:20:40icedice (icedice) joins
07:22:50icedice2 quits [Ping timeout: 252 seconds]
09:03:13icedice quits [Client Quit]
09:22:54icedice (icedice) joins
10:33:36imer quits [Quit: Oh no]
10:35:56imer (imer) joins
11:46:18ymgve joins
11:56:03BearFortress quits [Client Quit]
12:34:06BearFortress joins
13:10:26AmAnd0A quits [Ping timeout: 252 seconds]
13:11:16AmAnd0A joins
13:17:09icedice quits [Client Quit]
13:26:23HP_Archivist (HP_Archivist) joins
13:38:46icedice (icedice) joins
13:41:05HP_Archivist quits [Client Quit]
13:49:22Chris5010 quits [Ping timeout: 265 seconds]
14:01:34Chris5010 (Chris5010) joins
14:25:35Arcorann quits [Ping timeout: 252 seconds]
14:56:13hitgrr8 joins
16:19:10Zaxoosh joins
17:12:26Chris50103 (Chris5010) joins
17:14:18Chris5010 quits [Ping timeout: 265 seconds]
17:14:18Chris50103 is now known as Chris5010
17:22:05icedice quits [Client Quit]
18:10:16Wingy quits [Client Quit]
18:11:32icedice (icedice) joins
18:13:20gfhh joins
18:55:09<chrismeller>Andrew: i was doing some research on building a NAS a while back and it seems like most people recommend that you just use ZFS and let it do its magic, rather than setting up some specific version of RAID
19:06:32<fireonlive>there's ZFS RAIDZ1/Z2/Z3
19:07:20<fireonlive>(or just straight up ZFS mirror)
19:08:18<chrismeller>Z1 is the default i believe... assuming you have enough disks of course
19:30:03tzt_ is now known as tzt
19:30:07<@JAA>Magic? That seems awkward. The whole point of these is that you can select how much redundancy you want/how many disk failures you want the system to tolerate.
19:35:09<@JAA>I could imagine that a raidzX with 2^n+X disks would be slightly better for the parity calculation performance since it could work on units that are a power of two. But whether that actually matters in practice with modern hardware... Benchmark time.
20:28:26Zaxoosh quits [Remote host closed the connection]
20:45:38driib quits [Ping timeout: 252 seconds]
20:52:23hitgrr8 quits [Client Quit]
20:55:08AmAnd0A quits [Read error: Connection reset by peer]
20:55:27AmAnd0A joins
20:58:57icedice quits [Client Quit]
21:24:58icedice (icedice) joins
21:32:06driib (driib) joins
21:38:19driib quits [Client Quit]
21:38:59driib (driib) joins
21:48:09driib quits [Client Quit]
21:48:54driib (driib) joins
21:53:16<immibis>creating channels on discord breaks because it uses "open"ai to suggest emojis, and "open"ai is down
21:53:24<immibis>This is the future of the corporatized internet
21:54:35driib quits [Client Quit]
21:55:18driib (driib) joins
21:58:36driib quits [Client Quit]
21:59:26driib (driib) joins
22:00:16<icedice>lmao
22:19:05<Ryz>Woooooow
22:19:06<@Fusl>andrew: the general rule of thumb is to `fdisk -l /dev/sdX` on the disk, checking the physical sector size and then applying the proper ashift= value during zpool creation. for a typical hdd and ssd with 4096 bytes sector size, that is 2^12=4096, so ashift=12
22:22:50VerifiedJ quits [Remote host closed the connection]
22:23:24VerifiedJ (VerifiedJ) joins
22:27:30driib quits [Client Quit]
22:28:03driib (driib) joins
22:44:16AmAnd0A quits [Ping timeout: 252 seconds]
22:45:11AmAnd0A joins
22:46:50AmAnd0A quits [Read error: Connection reset by peer]
23:02:40<andrew>Fusl: that part I know, I'm asking about whether the number of disks per vdev in a RAIDZ1/2/3 configuration actually matters
23:03:32<andrew>the problem is that to see how well a specific hardware configuration will perform, I will need to buy the hardware, and before buying the hardware, I'd like to have a decent idea of how well that hardware would perform
23:04:41<@Fusl>it does for recovery/rebuild times. a raidz over 90 disks for example is slower than a raid0 over 6x raidz over 15 of the 90 disks each
23:05:23<@Fusl>and generally, raidz and hdd's don't mix together very well after the zpool becomes a little fragmented
23:05:33<@Fusl>(mostly due to the increased random i/o)
23:05:43<andrew>to provide context: I'm debating which (and how many) SSDs to buy to expand my SSD capacity
23:06:11<andrew>is it a waste of money/IOPS to have RAIDZ1 over four disks instead of three?
23:06:23<@Fusl>nope, that's perfectly fine
23:07:26<andrew>it's that age-old problem of whether to buy more now for less cost per usable GB or buy less now for less cost since I'm probably not going to fill up the storage for a while
23:09:16<@Fusl>imho i'd go with more when using zfs since expanding a raidz isn't easy (i think you'll have to recreate the entire zpool if you want to expand to more disks)
23:09:19<imer>been a while since I looked into this, so hazy on the details, I seem to remember 10 drives per raidzN is the sweet spot of performance/space "wasted" for parity (can still add multiple 10drive vdevs per pool if you want though, so no issues there). not personally tested that with ssds though, I remember numbers checking out on spinning rust from
23:09:19<imer>"yeah, good enough" testing though
23:09:34<andrew>imer: I'm not buying ten SSDs right now :P
23:09:40<imer>well, i dont knoow
23:09:51<imer>I wouldn't be surprised if people were!
23:10:15<andrew>anyways, I'm trying to decide whether to buy used eBay SAS SSDs (like the Samsung PM1643) or some brand new PCIe Gen4 SSDs (for about 50% more cost)
23:10:45<andrew>the brand new drives are consumer grade but their sustained performance still likely exceeds those old SAS drives
23:11:28AmAnd0A joins
23:12:36<andrew>I have concerns about my LSI 9300-16i being a bottleneck if I bought a bunch of SAS SSDs - there's a 7 GB/s limit due to the PCIe 3.0 x8 link, and the SAS controllers allegedly handle "over 1 million IOPS" each, which will easily be saturated by reading from two of the SSDs
23:13:37<andrew>that being said, I'm not sure whether it actually matters that much in the real world, chances are the CPU wouldn't be able to keep up anyways
23:15:28<@JAA>Isn't raidz expansion a thing now? I remember hearing about it like a year or two ago.
23:15:40<andrew>JAA: RAIDZ expansion is still a work in progress, I've been following that PR for a while
23:15:47<@JAA>Ah
23:16:03<@JAA>I guess it was a 'SOOON!!!!' thing then that I'm thinking of.
23:17:25<andrew>I know the thing I'm doing is a bit strange, I'm planning on running a database workload, which needs IOPS, but for cost savings I'm planning on using RAIDZ1 instead of mirrors, and to paper over the IOPS penalty of the RAIDZ I'm considering buying some hecking fast SSDs :P
23:50:42BlueMaxima joins
23:58:52Arcorann (Arcorann) joins