Re: What would a good OSD node hardware configuration look like?
On 07-11-12 09:17, Gandalf Corvotempesta wrote: 2012/11/7 Wido den Hollander : Except that SSDs will mainly fail due to the amount of write cycles they had to endure. So in RAID-1 your SSDs will fail at almost the same time. With for example 8 OSDs in a server you better spread them out 50/50 over two SSDs. But the OS should be installed on SSD, to allow the use of the whole SATA/SAS disk as OSDs. Without RAID-1 between SSDs, a disk failure will result it the whole node down. -- True. You might want to use a small internal USB stick like those from Transcend for installing your OS. They are available in 4GB and 8GB. Let syslog handle your logging to an external device and you have almost no write I/O on that thing. This way you can use the SSDs completely for journaling and shrink them using HPA with hdparm. Still, I think that in the distributed nature it shouldn't matter if you loose a node. If loosing a node is a to big impact to your cluster, go with smaller nodes. Wido To unsubscribe from this list: send the line "unsubscribe ceph-devel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html -- To unsubscribe from this list: send the line "unsubscribe ceph-devel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
syncfs slower than without syncfs
Hello list, syncfs is much slower than without syncfs to me. If i compile latest ceph master with wip-rbd-read: with syncfs: rand 4K: write: io=1133MB, bw=12853KB/s, iops=3213, runt= 90277msec read : io=1239MB, bw=14046KB/s, iops=3511, runt= 90325msec seq 4M: write: io=37560MB, bw=423874KB/s, iops=103, runt= 90738msec read : io=103852MB, bw=1151MB/s, iops=287, runt= 90237msec without syncfs: rand 4K: write: io=3733MB, bw=42459KB/s, iops=10614, runt= 90039msec read : io=6018MB, bw=68446KB/s, iops=17111, runt= 90038msec seq 4M: write: io=51204MB, bw=577328KB/s, iops=140, runt= 90820msec read : io=150320MB, bw=1666MB/s, iops=416, runt= 90228msec I thought syncfs should boost the performance. Greets, Stefan -- To unsubscribe from this list: send the line "unsubscribe ceph-devel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: SSD journal suggestion
On Wed, 7 Nov 2012, Gandalf Corvotempesta wrote: > I'm evaluating some SSD drives as journal. > Samsung 840 Pro seems to be the fastest in sequential reads and write. > > What parameter should I consider for a journal? I think that none of > read benchmark are influent because when dumping journal to disk, the > bottleneck will always be the SAS/SATA write speed. (in this case, the > SSD will never reach it's read best performance) > So, should I evaluate only the write speed when datas are wrote to the > journal ? Sequential or random ? Sequential write is the only thing that matters for the osd journal. I'd look at both large writes and small writes. sage -- To unsubscribe from this list: send the line "unsubscribe ceph-devel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: syncfs slower than without syncfs
Am 07.11.2012 10:41, schrieb Stefan Priebe - Profihost AG: Hello list, syncfs is much slower than without syncfs to me. If i compile latest ceph master with wip-rbd-read: with syncfs: rand 4K: write: io=1133MB, bw=12853KB/s, iops=3213, runt= 90277msec read : io=1239MB, bw=14046KB/s, iops=3511, runt= 90325msec seq 4M: write: io=37560MB, bw=423874KB/s, iops=103, runt= 90738msec read : io=103852MB, bw=1151MB/s, iops=287, runt= 90237msec without syncfs: rand 4K: write: io=3733MB, bw=42459KB/s, iops=10614, runt= 90039msec read : io=6018MB, bw=68446KB/s, iops=17111, runt= 90038msec seq 4M: write: io=51204MB, bw=577328KB/s, iops=140, runt= 90820msec read : io=150320MB, bw=1666MB/s, iops=416, runt= 90228msec I thought syncfs should boost the performance. Fixed - my chroot env with glibc syncfs support didn't had the libcrypto installed and it isn't in the control dependencies. Now with libcrypto i get the same results as without syncfs. Stefan -- To unsubscribe from this list: send the line "unsubscribe ceph-devel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Unexpected behavior by ceph 0.48.2argonaut.
I am not sure about my judgments but please help me out understanding the result of following experiment carried out : - Experiment : (3 node cluster, all have ceph v.0.48.2agonaut( after building ceph from src code) + UBUNTU 12.04 + kernel 3.2.0 ) -- VM1 ( mon.0+ osd.0 +mds.0 ) VM2 (mon.1 + osd.1 + mds.1) VM3(mon.2) - Cluster is up and HEALTH_OK - Replication factor is 2. (by default all pools have replication factor set to 2) - After mounting "mount.ceph mon_addr:port :/ ~/cephfs " , I created file inside mounted Dir "cephfs" . - And able to see data on both OSD i.e. VM1(osd.0) and on VM2(osd.1) as well as file is accessible . - Then VM2 is made down & VM2 absence is verified with ceph -s . - Even after VM1(osd.0 mds.0 mon.0) + VM3 (mon.2) was live , I am unable to access the file . - I tried to remount the data on different Dir with mount.ceph currently_live_mons:/ /home/hemant/xyz - Even after that I was unable to access the file stored on cluster. -- - Hemant Surale. -- To unsubscribe from this list: send the line "unsubscribe ceph-devel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
unexpected problem with radosgw fcgi
I have realize that requests from fastcgi in nginx from radosgw returning: HTTP/1.1 200, not a HTTP/1.1 200 OK Any other cgi that i run, for example php via fastcgi return this like RFC says, with OK. Is someone experience this problem ?? I see in code: ./src/rgw/rgw_rest.cc line 36 const static struct rgw_html_errors RGW_HTML_ERRORS[] = { { 0, 200, "" }, What if i change this into: { 0, 200, "OK" }, -- - Regards Sławek Skowron -- To unsubscribe from this list: send the line "unsubscribe ceph-devel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: SSD journal suggestion
On 11/07/2012 06:28 AM, Gandalf Corvotempesta wrote: 2012/11/7 Sage Weil : On Wed, 7 Nov 2012, Gandalf Corvotempesta wrote: I'm evaluating some SSD drives as journal. Samsung 840 Pro seems to be the fastest in sequential reads and write. The 840 Pro seems to reach 485MB/s in sequential write: http://www.storagereview.com/samsung_ssd_840_pro_review -- To unsubscribe from this list: send the line "unsubscribe ceph-devel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html I'm using Intel 510s in a test node and can do about 450MB/s per drive. Right now I'm doing 3 journals per SSD, but topping out at about 1.2-1.4GB/s from the client perspective for the node with 15+ drives and 5 SSDs. It's possible newer versions of the code and tuning may increase that. TV pointed me at the new Intel DC S3700 which looks like a very interesting option (the 100GB model for $240). http://www.anandtech.com/show/6432/the-intel-ssd-dc-s3700-intels-3rd-generation-controller-analyzed Mark -- To unsubscribe from this list: send the line "unsubscribe ceph-devel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: syncfs slower than without syncfs
Whew, glad you found the problem Stefan! I was starting to wonder what was going on. :) Do you mind filling a bug about the control dependencies? Mark On 11/07/2012 07:31 AM, Stefan Priebe - Profihost AG wrote: Am 07.11.2012 10:41, schrieb Stefan Priebe - Profihost AG: Hello list, syncfs is much slower than without syncfs to me. If i compile latest ceph master with wip-rbd-read: with syncfs: rand 4K: write: io=1133MB, bw=12853KB/s, iops=3213, runt= 90277msec read : io=1239MB, bw=14046KB/s, iops=3511, runt= 90325msec seq 4M: write: io=37560MB, bw=423874KB/s, iops=103, runt= 90738msec read : io=103852MB, bw=1151MB/s, iops=287, runt= 90237msec without syncfs: rand 4K: write: io=3733MB, bw=42459KB/s, iops=10614, runt= 90039msec read : io=6018MB, bw=68446KB/s, iops=17111, runt= 90038msec seq 4M: write: io=51204MB, bw=577328KB/s, iops=140, runt= 90820msec read : io=150320MB, bw=1666MB/s, iops=416, runt= 90228msec I thought syncfs should boost the performance. Fixed - my chroot env with glibc syncfs support didn't had the libcrypto installed and it isn't in the control dependencies. Now with libcrypto i get the same results as without syncfs. Stefan -- To unsubscribe from this list: send the line "unsubscribe ceph-devel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html -- To unsubscribe from this list: send the line "unsubscribe ceph-devel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: Mons network
On Wed, Nov 7, 2012 at 1:31 PM, Gandalf Corvotempesta wrote: > Which kind of network should I plain for 3 or 4 MONs nodes? > Are these node called by the ceph client (RBD, RGW, and so on) or only by OSD? > > Can I use some virtual machines distributed across multiple Xen nodes > on a 100mbit/s netoworks reachable by the whole storage cluster > (client, server, ods, rbd, rgw) on should I put a gigabit network > reachable only by OSD (for performance and security purposes) The mons need to be reachable by everybody. They don't do a ton of network traffic, but 100Mb/s might be pushing it a bit low... -- To unsubscribe from this list: send the line "unsubscribe ceph-devel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: Mons network
On Wed, Nov 7, 2012 at 4:20 PM, Gandalf Corvotempesta wrote: > 2012/11/7 Gregory Farnum : >> The mons need to be reachable by everybody. They don't do a ton of >> network traffic, but 100Mb/s might be pushing it a bit low... > > Some portion of my network are 10/100 with gigabit uplink. > In this portion we have many XenServer nodes that we would like to use > for mons, but maximum speed of each is 100mbit > Moving mons to the SAN will result in new hardware to buy. Well, you can experiment with it and see what happens. Or just colocate the monitors with some of your OSDs. -- To unsubscribe from this list: send the line "unsubscribe ceph-devel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
trying to import crushmap results in max_devices > osdmap max_osd
Hello, i've added two nodes with 4 devices each and modified the crushmap. But importing the new map results in: crushmap max_devices 55 > osdmap max_osd 35 What's wrong? Greets Stefan -- To unsubscribe from this list: send the line "unsubscribe ceph-devel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: SSD journal suggestion
On 11/07/2012 10:12 AM, Atchley, Scott wrote: On Nov 7, 2012, at 10:01 AM, Mark Nelson wrote: On 11/07/2012 06:28 AM, Gandalf Corvotempesta wrote: 2012/11/7 Sage Weil : On Wed, 7 Nov 2012, Gandalf Corvotempesta wrote: I'm evaluating some SSD drives as journal. Samsung 840 Pro seems to be the fastest in sequential reads and write. The 840 Pro seems to reach 485MB/s in sequential write: http://www.storagereview.com/samsung_ssd_840_pro_review -- To unsubscribe from this list: send the line "unsubscribe ceph-devel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html I'm using Intel 510s in a test node and can do about 450MB/s per drive. Is that sequential read or write? Intel lists them at 210-315 MB/s for sequential write. The 520s are rated at 475-520 MB/s seq. write. Doh, wrote that too early in the morning after staying all night watching the elections. :) You are correct, it's the 520, not the 510. Right now I'm doing 3 journals per SSD, but topping out at about 1.2-1.4GB/s from the client perspective for the node with 15+ drives and 5 SSDs. It's possible newer versions of the code and tuning may increase that. What interconnect is this? 10G Ethernet is 1.25 GB/s line rate and I would expect your Sockets and Ceph overhead to eat into that. Or is it dual 10G Ethernet? Scott This is 8 concurrent instances of rados bench running on localhost. Ceph is configured with 1x replication. 1.2-1.4GB/s is the aggregate throughput of all of the rados bench instances. TV pointed me at the new Intel DC S3700 which looks like a very interesting option (the 100GB model for $240). http://www.anandtech.com/show/6432/the-intel-ssd-dc-s3700-intels-3rd-generation-controller-analyzed Mark -- To unsubscribe from this list: send the line "unsubscribe ceph-devel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html -- To unsubscribe from this list: send the line "unsubscribe ceph-devel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: SSD journal suggestion
On Nov 7, 2012, at 10:01 AM, Mark Nelson wrote: > On 11/07/2012 06:28 AM, Gandalf Corvotempesta wrote: >> 2012/11/7 Sage Weil : >>> On Wed, 7 Nov 2012, Gandalf Corvotempesta wrote: I'm evaluating some SSD drives as journal. Samsung 840 Pro seems to be the fastest in sequential reads and write. >> >> The 840 Pro seems to reach 485MB/s in sequential write: >> http://www.storagereview.com/samsung_ssd_840_pro_review >> -- >> To unsubscribe from this list: send the line "unsubscribe ceph-devel" in >> the body of a message to majord...@vger.kernel.org >> More majordomo info at http://vger.kernel.org/majordomo-info.html >> > > I'm using Intel 510s in a test node and can do about 450MB/s per drive. Is that sequential read or write? Intel lists them at 210-315 MB/s for sequential write. The 520s are rated at 475-520 MB/s seq. write. > Right now I'm doing 3 journals per SSD, but topping out at about > 1.2-1.4GB/s from the client perspective for the node with 15+ drives and > 5 SSDs. It's possible newer versions of the code and tuning may > increase that. What interconnect is this? 10G Ethernet is 1.25 GB/s line rate and I would expect your Sockets and Ceph overhead to eat into that. Or is it dual 10G Ethernet? Scott > TV pointed me at the new Intel DC S3700 which looks like a very > interesting option (the 100GB model for $240). > > http://www.anandtech.com/show/6432/the-intel-ssd-dc-s3700-intels-3rd-generation-controller-analyzed > > Mark > -- > To unsubscribe from this list: send the line "unsubscribe ceph-devel" in > the body of a message to majord...@vger.kernel.org > More majordomo info at http://vger.kernel.org/majordomo-info.html -- To unsubscribe from this list: send the line "unsubscribe ceph-devel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: syncfs slower than without syncfs
Am 07.11.2012 16:04, schrieb Mark Nelson: Whew, glad you found the problem Stefan! I was starting to wonder what was going on. :) Do you mind filling a bug about the control dependencies? Sure where should i fill it in? Stefan -- To unsubscribe from this list: send the line "unsubscribe ceph-devel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: SSD journal suggestion
On Nov 7, 2012, at 11:20 AM, Mark Nelson wrote: >>> Right now I'm doing 3 journals per SSD, but topping out at about >>> 1.2-1.4GB/s from the client perspective for the node with 15+ drives and >>> 5 SSDs. It's possible newer versions of the code and tuning may >>> increase that. >> >> What interconnect is this? 10G Ethernet is 1.25 GB/s line rate and I would >> expect your Sockets and Ceph overhead to eat into that. Or is it dual 10G >> Ethernet? > > This is 8 concurrent instances of rados bench running on localhost. > Ceph is configured with 1x replication. 1.2-1.4GB/s is the aggregate > throughput of all of the rados bench instances. Ok, all local with no communication. Given this level of local performance, what does that translate into when talking over the network? Scott -- To unsubscribe from this list: send the line "unsubscribe ceph-devel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: SSD journal suggestion
On 11/07/2012 10:35 AM, Atchley, Scott wrote: On Nov 7, 2012, at 11:20 AM, Mark Nelson wrote: Right now I'm doing 3 journals per SSD, but topping out at about 1.2-1.4GB/s from the client perspective for the node with 15+ drives and 5 SSDs. It's possible newer versions of the code and tuning may increase that. What interconnect is this? 10G Ethernet is 1.25 GB/s line rate and I would expect your Sockets and Ceph overhead to eat into that. Or is it dual 10G Ethernet? This is 8 concurrent instances of rados bench running on localhost. Ceph is configured with 1x replication. 1.2-1.4GB/s is the aggregate throughput of all of the rados bench instances. Ok, all local with no communication. Given this level of local performance, what does that translate into when talking over the network? Scott Well, local, but still over tcp. Right now I'm focusing on pushing the osds/filestores as far as I can, and after that I'm going to setup a bonded 10GbE network to see what kind of messenger bottlenecks I run into. Sadly the testing is going slower than I would like. Mark -- To unsubscribe from this list: send the line "unsubscribe ceph-devel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: Ubuntu 12.04.1 + xfs + syncfs is still not our friend
On 11/07/2012 12:14 AM, Gandalf Corvotempesta wrote: 2012/11/7 Dan Mick : Resolution: installing the packages built for precise, rather than squeeze, got versions that use syncfs. Which packages, ceph or libc? The ceph packages. -- To unsubscribe from this list: send the line "unsubscribe ceph-devel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
rbd striping format v1 / v2 benchmark
Hello list, i've done some benchmarks regarding striping / v1 / v2. Results: format 1: write: io=5739MB, bw=65278KB/s, iops=16319, runt= 90029msec read : io=5771MB, bw=65636KB/s, iops=16408, runt= 90030msec write: io=77224MB, bw=874044KB/s, iops=213, runt= 90473msec read : io=178840MB, bw=1983MB/s, iops=495, runt= 90168msec format 2: --stripe-count 8 --stripe-unit 524288 -s 4194304 write: io=5377MB, bw=61147KB/s, iops=15286, runt= 90041msec read : io=5332MB, bw=60617KB/s, iops=15154, runt= 90067msec write: io=75136MB, bw=849285KB/s, iops=207, runt= 90593msec read : io=160292MB, bw=1777MB/s, iops=444, runt= 90226msec --stripe-count 4 --stripe-unit 1048576 -s 4194304 write: io=5301MB, bw=60281KB/s, iops=15070, runt= 90046msec read : io=5367MB, bw=61031KB/s, iops=15257, runt= 90057msec write: io=74448MB, bw=840293KB/s, iops=205, runt= 90724msec read : io=170616MB, bw=1891MB/s, iops=472, runt= 90227msec So it seems right now that striping doesn't improve performance. Greets, Stefan -- To unsubscribe from this list: send the line "unsubscribe ceph-devel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: rbd striping format v1 / v2 benchmark
On Wed, 7 Nov 2012, Stefan Priebe wrote: > Hello list, > > i've done some benchmarks regarding striping / v1 / v2. > > Results: > format 1: > > write: io=5739MB, bw=65278KB/s, iops=16319, runt= 90029msec > read : io=5771MB, bw=65636KB/s, iops=16408, runt= 90030msec > write: io=77224MB, bw=874044KB/s, iops=213, runt= 90473msec > read : io=178840MB, bw=1983MB/s, iops=495, runt= 90168msec > > format 2: > > --stripe-count 8 --stripe-unit 524288 -s 4194304 > > write: io=5377MB, bw=61147KB/s, iops=15286, runt= 90041msec > read : io=5332MB, bw=60617KB/s, iops=15154, runt= 90067msec > write: io=75136MB, bw=849285KB/s, iops=207, runt= 90593msec > read : io=160292MB, bw=1777MB/s, iops=444, runt= 90226msec > > --stripe-count 4 --stripe-unit 1048576 -s 4194304 > > write: io=5301MB, bw=60281KB/s, iops=15070, runt= 90046msec > read : io=5367MB, bw=61031KB/s, iops=15257, runt= 90057msec > write: io=74448MB, bw=840293KB/s, iops=205, runt= 90724msec > read : io=170616MB, bw=1891MB/s, iops=472, runt= 90227msec > > So it seems right now that striping doesn't improve performance. It's mainly going to help when you have a deep queue of small sequential IOs, and the fact that they are all piling up on a single rbd block is turning into a bottleneck. The rest of the time the overhead of splitting things into smaller pieces will slow you down. The intended use case is a database journal or something similar, where the latency requirements prevent us from making larger IOs, but things are still sequential. rbd bench-write ... will generate this workload. sage > > Greets, > Stefan > -- > To unsubscribe from this list: send the line "unsubscribe ceph-devel" in > the body of a message to majord...@vger.kernel.org > More majordomo info at http://vger.kernel.org/majordomo-info.html > > -- To unsubscribe from this list: send the line "unsubscribe ceph-devel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
extreme ceph-osd cpu load for rand. 4k write
Hello list, whiling benchmarking i was wondering, why the ceph-osd load is so extreme high while having random 4k write i/o. Here an example while benchmarking: random 4k write: 16.000 iop/s 180% CPU Load in top from EACH ceph-osd process random 4k read: 16.000 iop/s 19% CPU Load in top from EACH ceph-osd process seq 4M write: 800MB/s 14% CPU Load in top from EACH ceph-osd process seq 4M read: 1600MB/s 9% CPU Load in top from EACH ceph-osd process I can't understand why in this single case the load is so EXTREMELY high. Greets Stefan -- To unsubscribe from this list: send the line "unsubscribe ceph-devel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: SSD journal suggestion
Hi, I have 16 SAS disk on a LSI 9266-8i and 4 Intel 520 SSD on a HBA, the node has dual 10G Ethernet. The clients are 4 nodes with dual 10GeB, as test I use rados bench on each client. The aggregated write speed is around 1,6GB/s with single replication. In the first configuration, I had the SSDs on the raidcontroller as well, but then I saturated the PCIe 2.0 x8 interface of the raidcontroller, therefore I use a second controller for the SSDs. -martin Am 07.11.2012 17:41, schrieb Mark Nelson: Well, local, but still over tcp. Right now I'm focusing on pushing the osds/filestores as far as I can, and after that I'm going to setup a bonded 10GbE network to see what kind of messenger bottlenecks I run into. Sadly the testing is going slower than I would like. -- To unsubscribe from this list: send the line "unsubscribe ceph-devel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: SSD journal suggestion
Hi, I tested a Arista 7150S-24, a HP5900 and in a few weeks I will get a Mellanox MSX1016. ATM the Arista is may favourite. For the dual 10GeB NICs I tested the Intel X520-DA2 and the Mellanox ConnectX-3. My favourite is the Intel X520-DA2. -martin Am 07.11.2012 22:14, schrieb Gandalf Corvotempesta: 2012/11/7 Martin Mailand : I have 16 SAS disk on a LSI 9266-8i and 4 Intel 520 SSD on a HBA, the node has dual 10G Ethernet. The clients are 4 nodes with dual 10GeB, as test I use rados bench on each client. The aggregated write speed is around 1,6GB/s with single replication. Just for curiosity, which switches do you have? -- To unsubscribe from this list: send the line "unsubscribe ceph-devel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: SSD journal suggestion
Am 07.11.2012 22:35, schrieb Martin Mailand: Hi, I tested a Arista 7150S-24, a HP5900 and in a few weeks I will get a Mellanox MSX1016. ATM the Arista is may favourite. For the dual 10GeB NICs I tested the Intel X520-DA2 and the Mellanox ConnectX-3. My favourite is the Intel X520-DA2. That's pretty interesting i'll get the HP5900 and HP5920 in a few weeks. HP told me the deep packet buffers of the HP5920 will burst the performance and should be used for storage related stuff. Greets, Stefan -- To unsubscribe from this list: send the line "unsubscribe ceph-devel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: SSD journal suggestion
Hi Stefan, deep buffers means latency spikes, you should go for fast switching latency. The HP5900 has a latency of 1ms, the Arista and Mellanox of 250ns. And I you should think at the price the HP5900 cost 3 times of the Mellanox. -martin Am 07.11.2012 22:44, schrieb Stefan Priebe: Am 07.11.2012 22:35, schrieb Martin Mailand: Hi, I tested a Arista 7150S-24, a HP5900 and in a few weeks I will get a Mellanox MSX1016. ATM the Arista is may favourite. For the dual 10GeB NICs I tested the Intel X520-DA2 and the Mellanox ConnectX-3. My favourite is the Intel X520-DA2. That's pretty interesting i'll get the HP5900 and HP5920 in a few weeks. HP told me the deep packet buffers of the HP5920 will burst the performance and should be used for storage related stuff. Greets, Stefan -- To unsubscribe from this list: send the line "unsubscribe ceph-devel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html -- To unsubscribe from this list: send the line "unsubscribe ceph-devel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: SSD journal suggestion
Am 07.11.2012 22:55, schrieb Martin Mailand: Hi Stefan, deep buffers means latency spikes, you should go for fast switching latency. The HP5900 has a latency of 1ms, the Arista and Mellanox of 250ns. HP told me they all use the same ships and Arista measures latency while only one port is in use. HP guarentees the latency when all ports are in use. If this is correct or just somehing hp told me - i don't know. They told me the arista is slower and the statistics are not comporable... > And I you should think at the price the HP5900 cost 3 times of the > Mellanox. Don't know what the Mellanox coests. I get the HP for a really good price below 10.000 €. Greets, Stefan -- To unsubscribe from this list: send the line "unsubscribe ceph-devel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
less cores more iops / speed
Hello again, I've noticed something really interesting. I get 5000 iops / VM for rand. 4k writes while assigning 4 cores on a 2.5 Ghz Xeon. When i move this VM to another kvm host with 3.6Ghz i get 8000 iops (still 8 cores) when i then LOWER the assigned cores from 8 to 4 i get 14.500 iops. If i assign only 2 cores i get 16.000 iops... Why does less kvm cores mean more speed? Greets, Stefan -- To unsubscribe from this list: send the line "unsubscribe ceph-devel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: SSD journal suggestion
Hi, I *think* the HP is Broadcom based, the Arista is Fulcrum based, and I don't know which chips Mellanox is using. Our NOC tested both of them, an the Arista was the clear winner, at least in our workload. -martin Am 07.11.2012 22:59, schrieb Stefan Priebe: HP told me they all use the same ships and Arista measures latency while only one port is in use. HP guarentees the latency when all ports are in use. If this is correct or just somehing hp told me - i don't know. They told me the arista is slower and the statistics are not comporable... -- To unsubscribe from this list: send the line "unsubscribe ceph-devel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: SSD journal suggestion
good question, probably we do not have enough experience with IPoIB. But it looks good on paper, so it's definitely a try worth. -martin Am 07.11.2012 23:28, schrieb Gandalf Corvotempesta: 2012/11/7 Martin Mailand : I tested a Arista 7150S-24, a HP5900 and in a few weeks I will get a Mellanox MSX1016. ATM the Arista is may favourite. Why not infiniband? -- To unsubscribe from this list: send the line "unsubscribe ceph-devel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: SSD journal suggestion
On 11/07/2012 04:51 PM, Gandalf Corvotempesta wrote: 2012/11/7 Martin Mailand : But it looks good on paper, so it's definitely a try worth. is at least 4x times faster than 10gbe and AFAIK should have a lower latency. I'm planning to use infiniband as backend storage network, used for OSD replication. 2 HBA for each OSD should give me 80Gbps and full redundancy I haven't done much with IPoIB (just RDMA), but my understanding is that it tends to top out at like 15Gb/s. Some others on this mailing list can probably speak more authoritatively. Even with RDMA you are going to top out at around 3.1-3.2GB/s. This thread may be helpful/interesting: http://comments.gmane.org/gmane.linux.drivers.rdma/12279 Mark -- To unsubscribe from this list: send the line "unsubscribe ceph-devel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: less cores more iops / speed
On 11/07/2012 10:02 PM, Stefan Priebe wrote: > Hello again, > > I've noticed something really interesting. > > I get 5000 iops / VM for rand. 4k writes while assigning 4 cores on a > 2.5 Ghz Xeon. > > When i move this VM to another kvm host with 3.6Ghz i get 8000 iops > (still 8 cores) when i then LOWER the assigned cores from 8 to 4 i get > 14.500 iops. If i assign only 2 cores i get 16.000 iops... > > Why does less kvm cores mean more speed? Totally going on a limb here, but might be related to the cache maybe? When you have more cores your threads may bounce around the cores and invalidate cache entries as they go by; will less cores you might end up with some sort of twisted, forced cpu affinity that allows you to take advantage of caching. But I don't know, really. I would be amazed if what I just wrote had an ounce of truth, and would be completely astonished if that was the cause for such a sudden increase on iops. -Joao > > Greets, > Stefan > -- > To unsubscribe from this list: send the line "unsubscribe ceph-devel" in > the body of a message to majord...@vger.kernel.org > More majordomo info at http://vger.kernel.org/majordomo-info.html -- To unsubscribe from this list: send the line "unsubscribe ceph-devel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Openstack - Boot From New Volume
Hi All, I've been looking for this bit of code for awhile how to make open stack create a VM with the dashboard with a attached/imaged volume to avoid the current multistep process of creating a vm with ceph volume. Here's the code, thanks goes out to vishy on openstack-dev's and I've added a couple of bits to allow for downloading status and better vol naming. Enjoy! https://github.com/qgrasso/nova/commit/340210a7f62e9e34bcc15972d97e81eb282fddd6 https://github.com/qgrasso/nova/commit/31f6839a2c2dd72e2cc44c37cb316610b42598bd ** use at your own risk as always etc etc, :) Regards, Quenten -- To unsubscribe from this list: send the line "unsubscribe ceph-devel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: less cores more iops / speed
On 11/07/2012 06:00 PM, Joao Eduardo Luis wrote: On 11/07/2012 10:02 PM, Stefan Priebe wrote: Hello again, I've noticed something really interesting. I get 5000 iops / VM for rand. 4k writes while assigning 4 cores on a 2.5 Ghz Xeon. When i move this VM to another kvm host with 3.6Ghz i get 8000 iops (still 8 cores) when i then LOWER the assigned cores from 8 to 4 i get 14.500 iops. If i assign only 2 cores i get 16.000 iops... Why does less kvm cores mean more speed? Totally going on a limb here, but might be related to the cache maybe? When you have more cores your threads may bounce around the cores and invalidate cache entries as they go by; will less cores you might end up with some sort of twisted, forced cpu affinity that allows you to take advantage of caching. There's also the context switching overhead. It'd be interesting to know how much the writer processes were shifting around on cores. Stefan, what tool were you using to do writes? But I don't know, really. I would be amazed if what I just wrote had an ounce of truth, and would be completely astonished if that was the cause for such a sudden increase on iops. Yeah, it's seems pretty surprising that there would be any significant effect at this level of performance. -Joao Greets, Stefan -- To unsubscribe from this list: send the line "unsubscribe ceph-devel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html -- To unsubscribe from this list: send the line "unsubscribe ceph-devel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html -- To unsubscribe from this list: send the line "unsubscribe ceph-devel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: less cores more iops / speed
On 11/07/2012 06:00 PM, Joao Eduardo Luis wrote: On 11/07/2012 10:02 PM, Stefan Priebe wrote: Hello again, I've noticed something really interesting. I get 5000 iops / VM for rand. 4k writes while assigning 4 cores on a 2.5 Ghz Xeon. When i move this VM to another kvm host with 3.6Ghz i get 8000 iops (still 8 cores) when i then LOWER the assigned cores from 8 to 4 i get 14.500 iops. If i assign only 2 cores i get 16.000 iops... Why does less kvm cores mean more speed? Totally going on a limb here, but might be related to the cache maybe? When you have more cores your threads may bounce around the cores and invalidate cache entries as they go by; will less cores you might end up with some sort of twisted, forced cpu affinity that allows you to take advantage of caching. There's also the context switching overhead. It'd be interesting to know how much the writer processes were shifting around on cores. Stefan, what tool were you using to do writes? But I don't know, really. I would be amazed if what I just wrote had an ounce of truth, and would be completely astonished if that was the cause for such a sudden increase on iops. Yeah, it's seems pretty surprising that there would be any significant effect at this level of performance. -Joao Greets, Stefan -- To unsubscribe from this list: send the line "unsubscribe ceph-devel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html -- To unsubscribe from this list: send the line "unsubscribe ceph-devel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html -- To unsubscribe from this list: send the line "unsubscribe ceph-devel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: Unexpected behavior by ceph 0.48.2argonaut.
On 11/07/2012 05:34 AM, hemant surale wrote: I am not sure about my judgments but please help me out understanding the result of following experiment carried out : - Experiment : (3 node cluster, all have ceph v.0.48.2agonaut( after building ceph from src code) + UBUNTU 12.04 + kernel 3.2.0 ) -- VM1 ( mon.0+ osd.0 +mds.0 ) VM2 (mon.1 + osd.1 + mds.1) VM3(mon.2) - Cluster is up and HEALTH_OK - Replication factor is 2. (by default all pools have replication factor set to 2) - After mounting "mount.ceph mon_addr:port :/ ~/cephfs " , I created file inside mounted Dir "cephfs" . - And able to see data on both OSD i.e. VM1(osd.0) and on VM2(osd.1) as well as file is accessible . - Then VM2 is made down & VM2 absence is verified with ceph -s . - Even after VM1(osd.0 mds.0 mon.0) + VM3 (mon.2) was live , I am unable to access the file . Was the cluster showing all pgs active+degraded after this, or did some stay inactive? - I tried to remount the data on different Dir with mount.ceph currently_live_mons:/ /home/hemant/xyz - Even after that I was unable to access the file stored on cluster. -- - Hemant Surale. -- To unsubscribe from this list: send the line "unsubscribe ceph-devel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: syncfs slower than without syncfs
On 11/07/2012 08:26 AM, Stefan Priebe wrote: Am 07.11.2012 16:04, schrieb Mark Nelson: Whew, glad you found the problem Stefan! I was starting to wonder what was going on. :) Do you mind filling a bug about the control dependencies? Sure where should i fill it in? http://www.tracker.newdream.net/projects/ceph/issues/new -- To unsubscribe from this list: send the line "unsubscribe ceph-devel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
RE: less cores more iops / speed
> I've noticed something really interesting. > > I get 5000 iops / VM for rand. 4k writes while assigning 4 cores on a > 2.5 Ghz Xeon. > > When i move this VM to another kvm host with 3.6Ghz i get 8000 iops (still 8 > cores) when i then LOWER the assigned cores from 8 to 4 i get > 14.500 iops. If i assign only 2 cores i get 16.000 iops... > > Why does less kvm cores mean more speed? There is a serious bug in the kvm vhost code. Do you use virtio-net with vhost? see: http://lists.nongnu.org/archive/html/qemu-devel/2012-11/msg00579.html Please test using the e1000 driver instead. -- To unsubscribe from this list: send the line "unsubscribe ceph-devel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
RE: less cores more iops / speed
> > I've noticed something really interesting. > > > > I get 5000 iops / VM for rand. 4k writes while assigning 4 cores on a > > 2.5 Ghz Xeon. > > > > When i move this VM to another kvm host with 3.6Ghz i get 8000 iops > > (still 8 > > cores) when i then LOWER the assigned cores from 8 to 4 i get > > 14.500 iops. If i assign only 2 cores i get 16.000 iops... > > > > Why does less kvm cores mean more speed? > > There is a serious bug in the kvm vhost code. Do you use virtio-net with > vhost? > > see: http://lists.nongnu.org/archive/html/qemu-devel/2012- > 11/msg00579.html > > Please test using the e1000 driver instead. Or update the guest kernel (what guest kernel do you use?). AFAIK 3.X kernels does not trigger the bug. -- To unsubscribe from this list: send the line "unsubscribe ceph-devel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: less cores more iops / speed
Am 08.11.2012 um 06:42 schrieb Dietmar Maurer : >> I've noticed something really interesting. >> >> I get 5000 iops / VM for rand. 4k writes while assigning 4 cores on a >> 2.5 Ghz Xeon. >> >> When i move this VM to another kvm host with 3.6Ghz i get 8000 iops (still 8 >> cores) when i then LOWER the assigned cores from 8 to 4 i get >> 14.500 iops. If i assign only 2 cores i get 16.000 iops... >> >> Why does less kvm cores mean more speed? > > There is a serious bug in the kvm vhost code. Do you use virtio-net with > vhost? > > see: http://lists.nongnu.org/archive/html/qemu-devel/2012-11/msg00579.html > > Please test using the e1000 driver instead. Why is vhost net driver involved here at all? Kvm guest only uses ssh here. Stefan -- To unsubscribe from this list: send the line "unsubscribe ceph-devel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
RE: less cores more iops / speed
> Why is vhost net driver involved here at all? Kvm guest only uses ssh here. I though you are testing things (rdb) which depends on KVM network speed? -- To unsubscribe from this list: send the line "unsubscribe ceph-devel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: less cores more iops / speed
Am 08.11.2012 um 06:49 schrieb Dietmar Maurer : >>> I've noticed something really interesting. >>> >>> I get 5000 iops / VM for rand. 4k writes while assigning 4 cores on a >>> 2.5 Ghz Xeon. >>> >>> When i move this VM to another kvm host with 3.6Ghz i get 8000 iops >>> (still 8 >>> cores) when i then LOWER the assigned cores from 8 to 4 i get >>> 14.500 iops. If i assign only 2 cores i get 16.000 iops... >>> >>> Why does less kvm cores mean more speed? >> >> There is a serious bug in the kvm vhost code. Do you use virtio-net with >> vhost? >> >> see: http://lists.nongnu.org/archive/html/qemu-devel/2012- >> 11/msg00579.html >> >> Please test using the e1000 driver instead. > > Or update the guest kernel (what guest kernel do you use?). AFAIK 3.X kernels > does not trigger the bug. Guest and Host habe 3.6.6 installed. > -- To unsubscribe from this list: send the line "unsubscribe ceph-devel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: less cores more iops / speed
Am 08.11.2012 um 06:54 schrieb Dietmar Maurer : >> Why is vhost net driver involved here at all? Kvm guest only uses ssh here. > > I though you are testing things (rdb) which depends on KVM network speed? Kvm process uses librbd and both are running on host not in guest. Stefan > > -- > To unsubscribe from this list: send the line "unsubscribe ceph-devel" in > the body of a message to majord...@vger.kernel.org > More majordomo info at http://vger.kernel.org/majordomo-info.html -- To unsubscribe from this list: send the line "unsubscribe ceph-devel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html