[ceph-users] Beginner questions
Hello all. Sorry for the beginner questions... I am in the process of setting up a small (3 nodes, 288TB) Ceph cluster to store some research data. It is expected that this cluster will grow significantly in the next year, possibly to multiple petabytes and 10s of nodes. At this time I'm expected a relatively small number of clients, with only one or two actively writing collected data - albeit at a high volume per day. Currently I'm deploying on Debian 9 via ceph-ansible. Before I put this cluster into production I have a couple questions based on my experience to date: Luminous, Mimic, or Nautilus? I need stability for this deployment, so I am sticking with Debian 9 since Debian 10 is fairly new, and I have been hesitant to go with Nautilus. Yet Mimic seems to have had a hard road on Debian but for the efforts at Croit. * Statements on the Releases page are now making more sense to me, but I would like to confirm that Nautilus is the right choice at this time? Bluestore DB size: My nodes currently have 8 x 12TB drives (plus 4 empty bays) and a PCIe NVMe drive. If I understand the suggested calculation correctly, the DB size for a 12 TB Bluestore OSD would be 480GB. If my NVMe isn't big enough to provide this size, should I skip provisioning the DBs on the NVMe, or should I give each OSD 1/12th of what I have available? Also, should I try to shift budget a bit to get more NVMe as soon as I can, and redo the OSDs when sufficient NVMe is available? Thanks. -Dave ___ ceph-users mailing list ceph-users@lists.ceph.com http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com
Re: [ceph-users] Beginner questions
I would definitely go for Nautilus. there are quite some optimizations that went in after mimic. Bluestore DB size usually ends up at either 30 or 60 GB. 30 GB is one of the sweet spots during normal operation. But during compaction, ceph writes the new data before removing the old, hence the 60GB. Next sweetspot is 300/600GB. any size between 60 and 300 will never be unused. DB Usage is also dependent on ceph usage, object storage is known to use a lot more db space than rbd images for example. Op do 16 jan. 2020 om 17:46 schreef Dave Hall : > Hello all. > > Sorry for the beginner questions... > > I am in the process of setting up a small (3 nodes, 288TB) Ceph cluster to > store some research data. It is expected that this cluster will grow > significantly in the next year, possibly to multiple petabytes and 10s of > nodes. At this time I'm expected a relatively small number of clients, > with only one or two actively writing collected data - albeit at a high > volume per day. > > Currently I'm deploying on Debian 9 via ceph-ansible. > > Before I put this cluster into production I have a couple questions based > on my experience to date: > > Luminous, Mimic, or Nautilus? I need stability for this deployment, so I > am sticking with Debian 9 since Debian 10 is fairly new, and I have been > hesitant to go with Nautilus. Yet Mimic seems to have had a hard road on > Debian but for the efforts at Croit. > >- Statements on the Releases page are now making more sense to me, but >I would like to confirm that Nautilus is the right choice at this time? > > Bluestore DB size: My nodes currently have 8 x 12TB drives (plus 4 empty > bays) and a PCIe NVMe drive. If I understand the suggested calculation > correctly, the DB size for a 12 TB Bluestore OSD would be 480GB. If my > NVMe isn't big enough to provide this size, should I skip provisioning the > DBs on the NVMe, or should I give each OSD 1/12th of what I have > available? Also, should I try to shift budget a bit to get more NVMe as > soon as I can, and redo the OSDs when sufficient NVMe is available? > > Thanks. > > -Dave > ___ > ceph-users mailing list > ceph-users@lists.ceph.com > http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com > ___ ceph-users mailing list ceph-users@lists.ceph.com http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com
Re: [ceph-users] Beginner questions
Don't use Mimic, support for it is far worse than Nautilus or Luminous. I think we were the only company who built a product around Mimic, both Redhat and Suse enterprise storage was Luminous and then Nautilus skipping Mimic entirely. We only offered Mimic as a default for a limited time and immediately moved to Nautilus as it became available and Nautilus + Debian 10 has been great for us. Mimic and Debian 9 was... well, hacked together, due to the gcc backport issues. That's not to say that it doesn't work, in fact Mimic (> 13.2.2) and Debian 9 worked perfectly fine for us. Our Debian 10 and Nautilus packages are just so much better and more stable than Debian 9 + Mimic because we don't need to do weird things with Debian. Check the mailing list for old posts around the Mimic release by me to see how we did that build. It's not pretty, but it was the only way to use Ceph >= Mimic on Debian 9. All that mess has been eliminated with Debian 10. Paul -- Paul Emmerich Looking for help with your Ceph cluster? Contact us at https://croit.io croit GmbH Freseniusstr. 31h 81247 München www.croit.io Tel: +49 89 1896585 90 On Thu, Jan 16, 2020 at 6:55 PM Bastiaan Visser wrote: > I would definitely go for Nautilus. there are quite some > optimizations that went in after mimic. > > Bluestore DB size usually ends up at either 30 or 60 GB. > 30 GB is one of the sweet spots during normal operation. But during > compaction, ceph writes the new data before removing the old, hence the > 60GB. > Next sweetspot is 300/600GB. any size between 60 and 300 will never be > unused. > > DB Usage is also dependent on ceph usage, object storage is known to use a > lot more db space than rbd images for example. > > Op do 16 jan. 2020 om 17:46 schreef Dave Hall : > >> Hello all. >> >> Sorry for the beginner questions... >> >> I am in the process of setting up a small (3 nodes, 288TB) Ceph cluster >> to store some research data. It is expected that this cluster will grow >> significantly in the next year, possibly to multiple petabytes and 10s of >> nodes. At this time I'm expected a relatively small number of clients, >> with only one or two actively writing collected data - albeit at a high >> volume per day. >> >> Currently I'm deploying on Debian 9 via ceph-ansible. >> >> Before I put this cluster into production I have a couple questions based >> on my experience to date: >> >> Luminous, Mimic, or Nautilus? I need stability for this deployment, so I >> am sticking with Debian 9 since Debian 10 is fairly new, and I have been >> hesitant to go with Nautilus. Yet Mimic seems to have had a hard road on >> Debian but for the efforts at Croit. >> >>- Statements on the Releases page are now making more sense to me, >>but I would like to confirm that Nautilus is the right choice at this >> time? >> >> Bluestore DB size: My nodes currently have 8 x 12TB drives (plus 4 empty >> bays) and a PCIe NVMe drive. If I understand the suggested calculation >> correctly, the DB size for a 12 TB Bluestore OSD would be 480GB. If my >> NVMe isn't big enough to provide this size, should I skip provisioning the >> DBs on the NVMe, or should I give each OSD 1/12th of what I have >> available? Also, should I try to shift budget a bit to get more NVMe as >> soon as I can, and redo the OSDs when sufficient NVMe is available? >> >> Thanks. >> >> -Dave >> ___ >> ceph-users mailing list >> ceph-users@lists.ceph.com >> http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com >> > ___ > ceph-users mailing list > ceph-users@lists.ceph.com > http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com > ___ ceph-users mailing list ceph-users@lists.ceph.com http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com
Re: [ceph-users] Beginner questions
Dave; I'd like to expand on this answer, briefly... The information in the docs is wrong. There have been many discussions about changing it, but no good alternative has been suggested, thus it hasn't been changed. The 3rd party project that Ceph's BlueStore uses for its database (RocksDB), apparently only uses DB sizes of 3GB, 30GB, and 300GB. As Dave mentions below, when RocksDB executes a compact operation, it creates a new blob of the same target size, and writes the compacted data into it. This doubles the necessary space. In addition, BlueStore places its Write Ahead Log (WAL) into the fastest storage that is available to OSD daemon, i.e. NVMe if available. Since this is done before the first compaction is requested, the WAL can force compaction onto slower storage. Thus, the numbers I've had floating around in my head for our next cluster are: 7GB, 66GB, and 630GB. From all the discussion I've seen around RocksDB, those seem like good, common sense targets. Pick the largest one that works for your setup. All that said... You would really want to pair a 600GB+ NVMe with 12TB drives, otherwise your DB is almost guaranteed to overflow onto the spinning drive, and affect performance. I became aware of most of this after we planned our clusters, so I haven't tried it, YMMV. One final note: more hosts, and more spindles usually translates into better cluster-wide performance. I can't predict what the relatively low client counts you're suggesting would impact that. Thank you, Dominic L. Hilsbos, MBA Director – Information Technology Perform Air International Inc. dhils...@performair.com www.PerformAir.com From: ceph-users [mailto:ceph-users-boun...@lists.ceph.com] On Behalf Of Bastiaan Visser Sent: Thursday, January 16, 2020 10:55 AM To: Dave Hall Cc: ceph-users@lists.ceph.com Subject: Re: [ceph-users] Beginner questions I would definitely go for Nautilus. there are quite some optimizations that went in after mimic. Bluestore DB size usually ends up at either 30 or 60 GB. 30 GB is one of the sweet spots during normal operation. But during compaction, ceph writes the new data before removing the old, hence the 60GB. Next sweetspot is 300/600GB. any size between 60 and 300 will never be unused. DB Usage is also dependent on ceph usage, object storage is known to use a lot more db space than rbd images for example. Op do 16 jan. 2020 om 17:46 schreef Dave Hall : Hello all. Sorry for the beginner questions... I am in the process of setting up a small (3 nodes, 288TB) Ceph cluster to store some research data. It is expected that this cluster will grow significantly in the next year, possibly to multiple petabytes and 10s of nodes. At this time I'm expected a relatively small number of clients, with only one or two actively writing collected data - albeit at a high volume per day. Currently I'm deploying on Debian 9 via ceph-ansible. Before I put this cluster into production I have a couple questions based on my experience to date: Luminous, Mimic, or Nautilus? I need stability for this deployment, so I am sticking with Debian 9 since Debian 10 is fairly new, and I have been hesitant to go with Nautilus. Yet Mimic seems to have had a hard road on Debian but for the efforts at Croit. • Statements on the Releases page are now making more sense to me, but I would like to confirm that Nautilus is the right choice at this time? Bluestore DB size: My nodes currently have 8 x 12TB drives (plus 4 empty bays) and a PCIe NVMe drive. If I understand the suggested calculation correctly, the DB size for a 12 TB Bluestore OSD would be 480GB. If my NVMe isn't big enough to provide this size, should I skip provisioning the DBs on the NVMe, or should I give each OSD 1/12th of what I have available? Also, should I try to shift budget a bit to get more NVMe as soon as I can, and redo the OSDs when sufficient NVMe is available? Thanks. -Dave ___ ceph-users mailing list ceph-users@lists.ceph.com http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com ___ ceph-users mailing list ceph-users@lists.ceph.com http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com
Re: [ceph-users] Beginner questions
Bastiaan, Regarding EC pools: Our concern at 3 nodes is that 2-way replication seems risky - if the two copies don't match, which one is corrupted. However, 3-way replication on a 3 node cluster triples the price per TB. Doing EC pools that are the equivalent of RAID-5 2+1 seems like the right place to start as far as maximizing capacity is concerned, although I do understand the potential time involved in rebuilding a 12 TB drive. Early on I'd be more concerned about a drive failure than about a node failure. Regarding the hardware, our nodes are single socket EPYC 7302 (16 core, 32 thread) with 128GB RAM. From what I recall reading I think the RAM, at least, is a bit higher than recommended. Question: Does a PG (EC or replicated) span multiple drives per node? I haven't got to the point of understanding this part yet, so pardon the totally naive question. I'll probably be conversant on this by Monday. -Dave Dave Hall Binghamton University kdh...@binghamton.edu 607-760-2328 (Cell) 607-777-4641 (Office) On 1/16/2020 4:27 PM, Bastiaan Visser wrote: Dave made a good point WAL + DB might end up a little over 60G, I would probably go with ~70Gig partitions /LV's per OSD in your case. (if the nvme drive is smart enough to spread the writes over all available capacity, mort recent nvme's are). I have not yet seen a WAL larger or even close to than a gigabyte. We don't even think about EC-coded pools on clusters with less than 6 nodes (spindles, full SSD is another story). EC pools neer more processing resources We usually settle with 1 gig per TB of storage on replicated only sluters, but whet EC polls are involved, we add at least 50% to that. Also make sure your processors are up for it. Do not base your calculations on a healthy cluster -> build to fail. How long are you willing to be in a degraded state on node failure. Especially when using many larger spindles. recovery time might be way longer than you think. 12 * 12TB is 144TB storage, on a 4+2 EC pool you might end up with over 200 TB of traffic, on a 10Gig network that's roughly 2 and a half days to recover. IF your processors are not bottleneck due to EC parity calculations and all capacity is available for recovery (which is usually not the case, there is still production traffic that will eat up resources). Op do 16 jan. 2020 om 21:30 schreef <mailto:dhils...@performair.com>>: Dave; I don't like reading inline responses, so... I have zero experience with EC pools, so I won't pretend to give advice in that area. I would think that small NVMe for DB would be better than nothing, but I don't know. Once I got the hang of building clusters, it was relatively easy to wipe a cluster out and rebuild it. Perhaps you could take some time, and benchmark different configurations? Thank you, Dominic L. Hilsbos, MBA Director – Information Technology Perform Air International Inc. dhils...@performair.com www.PerformAir.com <http://www.PerformAir.com> -Original Message- From: Dave Hall [mailto:kdh...@binghamton.edu <mailto:kdh...@binghamton.edu>] Sent: Thursday, January 16, 2020 1:04 PM To: Dominic Hilsbos; ceph-users@lists.ceph.com <mailto:ceph-users@lists.ceph.com> Subject: Re: [External Email] RE: [ceph-users] Beginner questions Dominic, We ended up with a 1.6TB PCIe NVMe in each node. For 8 drives this worked out to a DB size of something like 163GB per OSD. Allowing for expansion to 12 drives brings it down to 124GB. So maybe just put the WALs on NVMe and leave the DBs on the platters? Understood that we will want to move to more nodes rather than more drives per node, but our funding is grant and donation based, so we may end up adding drives in the short term. The long term plan is to get to separate MON/MGR/MDS nodes and 10s of OSD nodes. Due to our current low node count, we are considering erasure-coded PGs rather than replicated in order to maximize usable space. Any guidelines or suggestions on this? Also, sorry for not replying inline. I haven't done this much in a while - I'll figure it out. Thanks. -Dave On 1/16/2020 2:48 PM, dhils...@performair.com <mailto:dhils...@performair.com> wrote: > Dave; > > I'd like to expand on this answer, briefly... > > The information in the docs is wrong. There have been many discussions about changing it, but no good alternative has been suggested, thus it hasn't been changed. > > The 3rd party project that Ceph's BlueStore uses for its database (RocksDB), apparently only uses DB sizes of 3GB, 30GB, and 300GB. As Dave mentions below, when RocksDB executes a compact operation, it c
Re: [ceph-users] Beginner questions
There is no difference in allocation between replication or EC. If failure domain is host, one osd per host ok s used for a PG. So if you use a 2+1 EC profile with a host failure domain, you need 3 hosts for a healthy cluster. The pool will go read-only when you have a failure (host or disk), or are doing maintenance on a node (reboot). On a node failure there will be no rebuilding, since there is no place to find a 3rd osd for a pg, so you'll have to fix/replace the node before any writes will be accepted. So yes, you can do a 2+1 EC pool on 3 nodes, you are paying the price in reliability, flexibility and maybe performance. Only way to really know the latter is benchmarking with your setup. I think you will be fine on the hardware side. Memory recommendations swing around between 512M and 1G per Tb storage.I usually go with 1 gig. But I never use disks larger than 4Tb. On the cpu I always try to have a few more cores than I have osd's in a machine. So 16 is fine in your case. On Fri, Jan 17, 2020, 03:29 Dave Hall wrote: > Bastiaan, > > Regarding EC pools: Our concern at 3 nodes is that 2-way replication > seems risky - if the two copies don't match, which one is corrupted. > However, 3-way replication on a 3 node cluster triples the price per TB. > Doing EC pools that are the equivalent of RAID-5 2+1 seems like the right > place to start as far as maximizing capacity is concerned, although I do > understand the potential time involved in rebuilding a 12 TB drive. Early > on I'd be more concerned about a drive failure than about a node failure. > > Regarding the hardware, our nodes are single socket EPYC 7302 (16 core, 32 > thread) with 128GB RAM. From what I recall reading I think the RAM, at > least, is a bit higher than recommended. > > Question: Does a PG (EC or replicated) span multiple drives per node? I > haven't got to the point of understanding this part yet, so pardon the > totally naive question. I'll probably be conversant on this by Monday. > > -Dave > > Dave Hall > Binghamton universitykdh...@binghamton.edu > 607-760-2328 (Cell) > 607-777-4641 (Office) > > > > On 1/16/2020 4:27 PM, Bastiaan Visser wrote: > > Dave made a good point WAL + DB might end up a little over 60G, I would > probably go with ~70Gig partitions /LV's per OSD in your case. (if the nvme > drive is smart enough to spread the writes over all available capacity, > mort recent nvme's are). I have not yet seen a WAL larger or even close to > than a gigabyte. > > We don't even think about EC-coded pools on clusters with less than 6 > nodes (spindles, full SSD is another story). > EC pools neer more processing resources We usually settle with 1 gig per > TB of storage on replicated only sluters, but whet EC polls are involved, > we add at least 50% to that. Also make sure your processors are up for it. > > Do not base your calculations on a healthy cluster -> build to fail. > How long are you willing to be in a degraded state on node failure. > Especially when using many larger spindles. recovery time might be way > longer than you think. 12 * 12TB is 144TB storage, on a 4+2 EC pool you > might end up with over 200 TB of traffic, on a 10Gig network that's roughly > 2 and a half days to recover. IF your processors are not bottleneck due to > EC parity calculations and all capacity is available for recovery (which is > usually not the case, there is still production traffic that will eat up > resources). > > Op do 16 jan. 2020 om 21:30 schreef : > >> Dave; >> >> I don't like reading inline responses, so... >> >> I have zero experience with EC pools, so I won't pretend to give advice >> in that area. >> >> I would think that small NVMe for DB would be better than nothing, but I >> don't know. >> >> Once I got the hang of building clusters, it was relatively easy to wipe >> a cluster out and rebuild it. Perhaps you could take some time, and >> benchmark different configurations? >> >> Thank you, >> >> Dominic L. Hilsbos, MBA >> Director – Information Technology >> Perform Air International Inc. >> dhils...@performair.com >> www.PerformAir.com >> >> >> -Original Message- >> From: Dave Hall [mailto:kdh...@binghamton.edu] >> Sent: Thursday, January 16, 2020 1:04 PM >> To: Dominic Hilsbos; ceph-users@lists.ceph.com >> Subject: Re: [External Email] RE: [ceph-users] Beginner questions >> >> Dominic, >> >> We ended up with a 1.6TB PCIe NVMe in each node. For 8 drives this >> worked out to a DB size of something like 163GB per OSD. Allowing for >> expansion to 12 drives brings it down to 124
Re: [ceph-users] Beginner questions
I would strongly advise against 2+1 EC pools for production if stability is your main concern. There was a discussion towards the end of last year addressing this in more detail. Short story, if you don't have at least 8-10 nodes (in the short run), EC is not suitable. You cannot maintain a cluster with such EC-pools. Reasoning: k+1 is a no-go in production. You can set min_size to k, but whenever a node is down (maintenance or whatever), new writes are non-redundant. Loosing just one more disk means data loss. This is not a problem with replication x3 and min_size=2. Be aware that maintenance more often than not takes more than a day. Parts may need to be shipped. An upgrade goes wrong and requires lengthy support for fixing. Etc. In addition, admins make mistakes. You need to build your cluster such that it can survive mistakes (shut down wrong host, etc.) in degraded state. Redundancy m=1 means zero tolerance for errors. Often the recommendation therefore is m=3, while m=2 is the bare minimum. Note that EC 1+2 is equal in redundancy as replication x3, but will use more compute (hence, its useless). In your situation, I would start with replicated pools and move to EC once enough nodes are at hand. If you want to use the benefits of EC, you need to build large clusters. Starting with 3 nodes and failure domain disk will be a horrible experience. You will not be able to maintain, upgrade or fix anything without downtime. Plan for sleeping well in worst-case situations. Best regards, = Frank Schilder AIT Risø Campus Bygning 109, rum S14 From: ceph-users on behalf of Bastiaan Visser Sent: 17 January 2020 06:55:25 To: Dave Hall Cc: ceph-users@lists.ceph.com Subject: Re: [ceph-users] Beginner questions There is no difference in allocation between replication or EC. If failure domain is host, one osd per host ok s used for a PG. So if you use a 2+1 EC profile with a host failure domain, you need 3 hosts for a healthy cluster. The pool will go read-only when you have a failure (host or disk), or are doing maintenance on a node (reboot). On a node failure there will be no rebuilding, since there is no place to find a 3rd osd for a pg, so you'll have to fix/replace the node before any writes will be accepted. So yes, you can do a 2+1 EC pool on 3 nodes, you are paying the price in reliability, flexibility and maybe performance. Only way to really know the latter is benchmarking with your setup. I think you will be fine on the hardware side. Memory recommendations swing around between 512M and 1G per Tb storage.I usually go with 1 gig. But I never use disks larger than 4Tb. On the cpu I always try to have a few more cores than I have osd's in a machine. So 16 is fine in your case. On Fri, Jan 17, 2020, 03:29 Dave Hall mailto:kdh...@binghamton.edu>> wrote: Bastiaan, Regarding EC pools: Our concern at 3 nodes is that 2-way replication seems risky - if the two copies don't match, which one is corrupted. However, 3-way replication on a 3 node cluster triples the price per TB. Doing EC pools that are the equivalent of RAID-5 2+1 seems like the right place to start as far as maximizing capacity is concerned, although I do understand the potential time involved in rebuilding a 12 TB drive. Early on I'd be more concerned about a drive failure than about a node failure. Regarding the hardware, our nodes are single socket EPYC 7302 (16 core, 32 thread) with 128GB RAM. From what I recall reading I think the RAM, at least, is a bit higher than recommended. Question: Does a PG (EC or replicated) span multiple drives per node? I haven't got to the point of understanding this part yet, so pardon the totally naive question. I'll probably be conversant on this by Monday. -Dave Dave Hall Binghamton University kdh...@binghamton.edu<mailto:kdh...@binghamton.edu> 607-760-2328 (Cell) 607-777-4641 (Office) On 1/16/2020 4:27 PM, Bastiaan Visser wrote: Dave made a good point WAL + DB might end up a little over 60G, I would probably go with ~70Gig partitions /LV's per OSD in your case. (if the nvme drive is smart enough to spread the writes over all available capacity, mort recent nvme's are). I have not yet seen a WAL larger or even close to than a gigabyte. We don't even think about EC-coded pools on clusters with less than 6 nodes (spindles, full SSD is another story). EC pools neer more processing resources We usually settle with 1 gig per TB of storage on replicated only sluters, but whet EC polls are involved, we add at least 50% to that. Also make sure your processors are up for it. Do not base your calculations on a healthy cluster -> build to fail. How long are you willing to be in a degraded state on node failure. Especially when using many larger spindles. recovery time might be way longer than you think. 12 * 12TB is 144T
Re: [ceph-users] Beginner questions
Frank, Thank you for your input. It is good to know that the cluster will go read-only in if a node goes down. Our circumstance is probably a bit unusual, which is why I'm considering the2+1 solution. We have a researcher who will be collecting extremely large amounts of data in real time, requiring both high write and high read bandwith, but it's pretty much going to be a single user or a small research group. Right now we have 3 physical storage hosts and we need to get into production. We also need to maximize the available storage on these 3 nodes. We chose Ceph doe to scalability. As the research (and the funding) progresses we expect to add many more Ceph nodes, and to move the MONs/MGRs/MDSs off on to dedicated systems. At that time I'd likely lay out more rational pools and be more thoughtful about resiliency, understanding, of course, that I'd have to play games and migrate data around. But for now we have to make the most of the hardware we have. I'm thinking 2+1 because that gives me more usable storage that keeping copies, and much more than keeping 3 copies. -Dave Dave Hall Binghamton University kdh...@binghamton.edu 607-760-2328 (Cell) 607-777-4641 (Office) On 1/17/2020 3:50 AM, Frank Schilder wrote: I would strongly advise against 2+1 EC pools for production if stability is your main concern. There was a discussion towards the end of last year addressing this in more detail. Short story, if you don't have at least 8-10 nodes (in the short run), EC is not suitable. You cannot maintain a cluster with such EC-pools. Reasoning: k+1 is a no-go in production. You can set min_size to k, but whenever a node is down (maintenance or whatever), new writes are non-redundant. Loosing just one more disk means data loss. This is not a problem with replication x3 and min_size=2. Be aware that maintenance more often than not takes more than a day. Parts may need to be shipped. An upgrade goes wrong and requires lengthy support for fixing. Etc. In addition, admins make mistakes. You need to build your cluster such that it can survive mistakes (shut down wrong host, etc.) in degraded state. Redundancy m=1 means zero tolerance for errors. Often the recommendation therefore is m=3, while m=2 is the bare minimum. Note that EC 1+2 is equal in redundancy as replication x3, but will use more compute (hence, its useless). In your situation, I would start with replicated pools and move to EC once enough nodes are at hand. If you want to use the benefits of EC, you need to build large clusters. Starting with 3 nodes and failure domain disk will be a horrible experience. You will not be able to maintain, upgrade or fix anything without downtime. Plan for sleeping well in worst-case situations. Best regards, = Frank Schilder AIT Risø Campus Bygning 109, rum S14 From: ceph-users on behalf of Bastiaan Visser Sent: 17 January 2020 06:55:25 To: Dave Hall Cc: ceph-users@lists.ceph.com Subject: Re: [ceph-users] Beginner questions There is no difference in allocation between replication or EC. If failure domain is host, one osd per host ok s used for a PG. So if you use a 2+1 EC profile with a host failure domain, you need 3 hosts for a healthy cluster. The pool will go read-only when you have a failure (host or disk), or are doing maintenance on a node (reboot). On a node failure there will be no rebuilding, since there is no place to find a 3rd osd for a pg, so you'll have to fix/replace the node before any writes will be accepted. So yes, you can do a 2+1 EC pool on 3 nodes, you are paying the price in reliability, flexibility and maybe performance. Only way to really know the latter is benchmarking with your setup. I think you will be fine on the hardware side. Memory recommendations swing around between 512M and 1G per Tb storage.I usually go with 1 gig. But I never use disks larger than 4Tb. On the cpu I always try to have a few more cores than I have osd's in a machine. So 16 is fine in your case. On Fri, Jan 17, 2020, 03:29 Dave Hall mailto:kdh...@binghamton.edu>> wrote: Bastiaan, Regarding EC pools: Our concern at 3 nodes is that 2-way replication seems risky - if the two copies don't match, which one is corrupted. However, 3-way replication on a 3 node cluster triples the price per TB. Doing EC pools that are the equivalent of RAID-5 2+1 seems like the right place to start as far as maximizing capacity is concerned, although I do understand the potential time involved in rebuilding a 12 TB drive. Early on I'd be more concerned about a drive failure than about a node failure. Regarding the hardware, our nodes are single socket EPYC 7302 (16 core, 32 thread) with 128GB RAM. From what I recall reading I think the RAM, at least, is a bit higher than recommended. Question: Does a PG (EC or replicated) span m