Hello Sebastien, thanks for your reply. >>Are you going with a 10Gb network? It’s not an issue for IOPS but more for the bandwidth. If so read the following: Currently it's planned to use 1gb network for the public network (vm-->rbd cluster). Maybe 10gbe for cluster network replication is possible. (I'll use replication x 3) My workload is a lot random (around 400vms with small ios) , so I don't use too much bandwith. (I'm currently using a netapp san, with 2 x 24 sas disk array, with 4 gigabit lacp links by array, and I'm far to saturate them) >>I personally won’t go with a ratio of 1:6 for the journal. I guess 1:5 (or even 1:4) is preferable. >>SAS 10K gives you around 140MB/sec for sequential writes. >>So if you use a journal with an SSD, you expect at least 140MB if you don’t want to slow things down. >>If you do so 140*10 (disks): fulfil your 10GB bandwidth already. So either you don’t need that much disks either you don’t need SSDs. >>It depends on the performance that you want to achieve. Ok, I understand. (But I think for my random workload it should be ok) >>Another thing, I also won’t use the DC S3700 since this disk was definitely made for IOPS intensive applications. The journal is purely sequential (small seq block, IIRC Stephan mentioned 370k blocks). >>I will instead use with a SSD with large sequential capabilities like 525 series 120GB. Ok, thanks! ----- Mail original ----- De: "Sebastien Han" <sebastien.han@xxxxxxxxxxxx> À: "Alexandre DERUMIER" <aderumier@xxxxxxxxx> Cc: "ceph-users" <ceph-users@xxxxxxxxxxxxxx> Envoyé: Mercredi 15 Janvier 2014 13:55:39 Objet: Re: servers advise (dell r515 or supermicro ....) Hi Alexandre, Are you going with a 10Gb network? It’s not an issue for IOPS but more for the bandwidth. If so read the following: I personally won’t go with a ratio of 1:6 for the journal. I guess 1:5 (or even 1:4) is preferable. SAS 10K gives you around 140MB/sec for sequential writes. So if you use a journal with an SSD, you expect at least 140MB if you don’t want to slow things down. If you do so 140*10 (disks): fulfil your 10GB bandwidth already. So either you don’t need that much disks either you don’t need SSDs. It depends on the performance that you want to achieve. Another thing, I also won’t use the DC S3700 since this disk was definitely made for IOPS intensive applications. The journal is purely sequential (small seq block, IIRC Stephan mentioned 370k blocks). I will instead use with a SSD with large sequential capabilities like 525 series 120GB. Cheers. –––– Sébastien Han Cloud Engineer "Always give 100%. Unless you're giving blood.” Phone: +33 (0)1 49 70 99 72 Mail: sebastien.han@xxxxxxxxxxxx Address : 10, rue de la Victoire - 75009 Paris Web : www.enovance.com - Twitter : @enovance On 15 Jan 2014, at 12:47, Alexandre DERUMIER <aderumier@xxxxxxxxx> wrote: > Hello List, > > I'm going to build a build a rbd cluster this year, with 5 nodes > > I would like to have this kind of configuration for each node: > > - 2U > - 2,5inch drives > > os : 2 disk sas drive > journal : 2 x ssd intel dc s3700 100GB > osd : 10 or 12 x sas Seagate Savvio 10K.6 900GB > > > > I see on the mailing that intank use dell r515. > I currently own a lot of dell servers and I have good prices. > > But I have also see on the mailing that dell perc H700 can have some performance problem, > and also it's not easy to flash the firmware for jbod mode. > http://www.spinics.net/lists/ceph-devel/msg16661.html > > I don't known if theses performance problem has finally been solved ? > > > > Another option could be to use supermicro server, > they have some 2U - 16 disks chassis + one or two lsi jbod controller. > But, I have had in past really bad experience with supermicro motherboard. > (Mainly firmware bug, ipmi card bug,.....) > > Does someone have experience with supermicro, and give me advise for a good motherboard model? > > > Best Regards, > > Alexandre Derumier > > > > > > > > _______________________________________________ > ceph-users mailing list > ceph-users@xxxxxxxxxxxxxx > http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com _______________________________________________ ceph-users mailing list ceph-users@xxxxxxxxxxxxxx http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com
 |