From: Aleksander W. <ale...@mo...> - 2014-10-31 12:39:20
|
Hi Yves Actually MooseFS 2.0.x is a stable release - it means that it is used in a production by many customers all over the world. In our opinion you will get much better performance with MooseFS 2.0.40-1, since we’ve made a lot of changes since version 1.6.27 - especially in communication layer. For a few months now, our own business is relying on MooseFS 2.0.40-1 for handling aproximately 2 petabytes of data in two instances and it works very well - with over 270 simultaneousmounts! Also it’s important to note, that many problems of the previous version (1.6.x) disappeared after switching to MooseFS 2.0. Please find the following test results performer in lab environment with MooseFS CE 2.0.40-1 installed on hardware with the following configuration: 1x mfsmaster 3x chunkservers - 8GB RAM, 2xSATA 1TB 7200RPM HDD, 1Gb LAN JumboFrames, Intel Xeon CPU * *Write to /mnt/mfs/goal2 dd if=/dev/zero of=test1m.bin bs=1M count=10240 10737418240 bytes (11 GB) copied, 92.4997 s, 116 MB/s dd if=/dev/zero of=test64k.bin bs=64k count=10240 671088640 bytes (671 MB) copied, 5.76488 s, 115 MB/s Read from /mnt/mfs/goal2 dd if=test10g.bin of=/dev/null 10737418240 bytes (11 GB) copied, 104 s, 103 MB/s Of course this test are not showing the real power of MooseFS. Best Regards Aleksander Wieliczko Technical Support Engineer moosefs.com On 10/30/2014 07:20 PM, Yves wrote: > Hi Aleksander, > > for the moment I prefer to keep mfs 1.6.27, which is the latest stable > version of branch 1.6 (2.0 suppose migration and test again before > production I think ...) > > My test are very simple tests done with basics tools : > > This is my new test now : > > 2 same servers (6 core AMD FX-6300 3,5 Ghz / 8 Go RAM / sata disk 6Gbps > with no raid / gigabyte LAN CAT 6). > > I know the benefice of Jumbo frames but it is not possible to apply in > my situation at the moment. > > So when I transfert a file under SSH from server A to server B (A and B > are on RAID 1), 100 MB file (random content) : > Result : speed transfert = 100 MB/s > > Now with MFS with goal = 2, the same file : 33 MB/s > > So, it's better, few days ago, the speed was ... 10MB/s > > I continue tests ... > > Yves > > Aleksander Wieliczko a écrit : >> Hi Yves >> We are very glad that You are using MooseFS distributed file system. >> >> First of all we strongly recommended to update your instance to MooseFS >> 2.0.40-1 version. >> We fixed some bugs and add many new algorithms. All this make MooseFS >> 2.0 faster and provide better stability to cluster. >> >> >From description it appears that your LAN connection is getting only 76% >> bandwidth of 1Gb LAN. >> Can you check if you have any errors, dropped frames in INC interface? >> >> MooseFS is much more sensitive on LAN problems than nc. >> MooseFS need to communicate not only to one computer, but to master, >> chunkservers and clients, so this operations generating much more TCP >> traffic than one to one connection. >> >> By the way. >> We have similar configuration (8GB RAM, SATA 7200 RPM HDD, 1Gb LAN) in >> our development environment and we getting results for GOAL2 >> read: 80MB/s and write: 50MB/s >> >> Can you send some more informations about this tests, and what software >> are you using to measure the MooseFS speed. >> >> Best regards >> Aleksander Wieliczko >> Technical Support Engineer >> moosefs.com >> >> On 10/18/2014 12:22 PM, "Yves Réveillon - eurower.fr" wrote: >> >>> Hi everybody, >>> >>> I got poor performance with MooseFS :s >>> >>> This is my test : >>> >>> 3 servers with on each : >>> - 6 core AMD FX-6300 3,5 Ghz >>> - 8 Go RAM >>> - sata disk 6Gbps with no raid >>> - gigabyte LAN >>> >>> The switch is a HP Gigabyte, with CAT 6 >>> >>> 1 server is the mfsmaster >>> 3 servers are chunkservers >>> >>> Tests with a tmpfs randomly data of 1 Go : >>> >>> Bandwidth beetween 2 server with ncat : 780 Mbps >>> Bandwitdh to write to a single sata (RAID-less) : 2593 Mbps (324 MB/s) >>> >>> MooseFS 1.6.27 with 2x replication goal : >>> Bandwidth : 85 Mbps (10.6 MB/s). >>> >>> During test, RAM is good, no production on theses servers (no activity) >>> and CPU is enought (less than 20%) >>> >>> Network and HDD not seems to be a bottleneck, so do you have any idea of >>> why I have 10.6 MB/s and not 20 or 30 MB/s as announced ? >>> >>> Thanks you >>> >>> Yves >>> >>> >>> >>> >>> ------------------------------------------------------------------------------ >>> Comprehensive Server Monitoring with Site24x7. >>> Monitor 10 servers for $9/Month. >>> Get alerted through email, SMS, voice calls or mobile push notifications. >>> Take corrective actions from your mobile device. >>> http://p.sf.net/sfu/Zoho >>> _______________________________________________ >>> moosefs-users mailing list >>> moo...@li... >>> https://lists.sourceforge.net/lists/listinfo/moosefs-users >>> >> ------------------------------------------------------------------------------ >> Comprehensive Server Monitoring with Site24x7. >> Monitor 10 servers for $9/Month. >> Get alerted through email, SMS, voice calls or mobile push notifications. >> Take corrective actions from your mobile device. >> http://p.sf.net/sfu/Zoho >> _______________________________________________ >> moosefs-users mailing list >> moo...@li... >> https://lists.sourceforge.net/lists/listinfo/moosefs-users >> |