Re: [Gluster-users] GlusterFS & MMAP Support?
I read somewhere that MMAP is supported in fuse in recent kernels. I am also looking for MMAP support because I want to run openvz containers hosted on glusterfs and some applications need MMAP. I am kind of stuck as using a newer kernel means I need to patch it to work with openvz, something I would like to avoid. Pablo On Sun, Nov 29, 2009 at 12:38 PM, Harshavardhana wrote: > Hi Jeffery, > >Depends on your kernel version? and which fuse ABI under use . you can > get that >info from "dmesg | grep -i fuse" > > Regards > -- > Harshavardhana > Gluster - http://www.gluster.com > > > On Sun, Nov 29, 2009 at 2:57 PM, Jeffery Soo wrote: > > > Thanks for the tip but unfortunately it doesn't seem to help. Do you > know > > if FUSE supports mmap and if there's something in FUSE that needs to be > > changed to make mmap work? > > > > > > > > Harshavardhana wrote: > > > > Hi Jeffery, > > > > try --disable-direct-io as a command line option for glusterfs. > > > > Regards > > -- > > Harshavardhana > > Gluster - http://www.gluster.com > > > > > > On Sun, Nov 29, 2009 at 11:24 AM, Jeffery Soo > wrote: > > > >> I have the gluster patched fuse and 2.0.8 GlusterFS but it still does > not > >> support mmap. I've read conflicting things that fuse itself doesn't > support > >> mmap at all. > >> Is there anyway to make mmap work in GlusterFS? > >> > >> Thanks > >> ___ > >> Gluster-users mailing list > >> Gluster-users@gluster.org > >> http://gluster.org/cgi-bin/mailman/listinfo/gluster-users > >> > > > > > > > > ___ > Gluster-users mailing list > Gluster-users@gluster.org > http://gluster.org/cgi-bin/mailman/listinfo/gluster-users > > ___ Gluster-users mailing list Gluster-users@gluster.org http://gluster.org/cgi-bin/mailman/listinfo/gluster-users
Re: [Gluster-users] DRBD like performance?
Hello, DRBD is faster because, among other things, it replicate byte blocks on disks, whatever they are. A corrupted master will be simply replicated... In order to try to improve glusterfs (metadata file level) performance one could try NUFA first at client side and then server side replication. NUFA is _very_ good to accelerate local writes and reads at distributed configuration. Maybe, such concepts could be developed to replication in future. Regards. Andre Felipe Machado http://www.techforce.com.br ___ Gluster-users mailing list Gluster-users@gluster.org http://gluster.org/cgi-bin/mailman/listinfo/gluster-users
Re: [Gluster-users] GlusterFS Slow Performance In One Direction
Sorry to bump this, but I'm surprised that no one had any thoughts. What could cause the speed in my setup to be so slow in one direction? Thanks Jeffery Soo wrote: Does anyone have any ideas? I also notice the performance on listing and other things is not like DRBD. Is there something in my config files that I should change? Thanks Jeffery Soo wrote: Here's my setup. I have 2 servers with the exact hardware and OS configuration. Each server is also a client. I've tried to do a "DRBD" style setup using GlusterFS. Let's say I have gluserfs mounted on /mnt/gluster On Remote 1's shell I issue this command: cp /mnt/gluster/test1000mbfile /mnt/gluster/test100mbfile-- The transfer rate to Server 2 reaches about 30-40 MB/s. The problem is if I do this same thing on Server 2, the transfer rate to Server 1 is only about 10-15 MB/s Does anyone have any idea what is causing the slow performance from Server 2 to Server 1? Server 2 is capable of sending at 40MB/s to Server 1 in an SCP transfer, so I've ruled out network or hardware issues. = Both servers have the same glusterfsd.vol and glusterfs.vol files: Client File: +--+ 1: volume remote1 2: type protocol/client 3: option transport-type tcp 4: option remote-host xx.xx.xx.xx 5: option remote-subvolume brick 6: end-volume 7: 8: volume remote2 9: type protocol/client 10: option transport-type tcp 11: option remote-host xx.xx.xx.xx 17: option remote-subvolume brick 18: end-volume 19: 20: volume replicate 21: type cluster/replicate 22: subvolumes remote1 remote2 23: end-volume 24: 25: volume writebehind 26: type performance/write-behind 27: option window-size 1MB 28: subvolumes replicate 29: end-volume 30: 31: volume cache 32: type performance/io-cache 33: option cache-size 512MB 34: subvolumes writebehind 35: end-volume Server = 1: volume posix 2: type storage/posix 3: option directory /data/export 4: end-volume 5: 6: volume locks 7: type features/locks 8: subvolumes posix 9: end-volume 10: 11: volume brick 12: type performance/io-threads 13: option thread-count 8 14: subvolumes locks 15: end-volume 16: 17: volume server 18: type protocol/server 19: option transport-type tcp 20: option auth.addr.brick.allow * 21: subvolumes brick 22: end-volume ___ Gluster-users mailing list Gluster-users@gluster.org http://gluster.org/cgi-bin/mailman/listinfo/gluster-users ___ Gluster-users mailing list Gluster-users@gluster.org http://gluster.org/cgi-bin/mailman/listinfo/gluster-users ___ Gluster-users mailing list Gluster-users@gluster.org http://gluster.org/cgi-bin/mailman/listinfo/gluster-users
Re: [Gluster-users] DRBD like performance?
I'm having a similar problem, I'm looking into DRBD but the downside here will be, if the head server goes down the clients won't automatically switch over to the slave server... > -Original Message- > From: gluster-users-boun...@gluster.org > [mailto:gluster-users-boun...@gluster.org] On Behalf Of Jeffery Soo > Sent: 29 November 2009 09:30 > To: gluster-users@gluster.org > Subject: [Gluster-users] DRBD like performance? > > I had the intention of using GlusterFS to replace DRBD to setup a > clustered/redundant webserver but so far the performance is > about 7-8x > slower than native due to the live writing feature that > GlusterFS uses. > Is it possible to have a setup like DRBD to improve performance? > > Basically I want to know if I can get the same functionality and > performance of DRBD? I have 2 servers and with DRBD each > server would > perform all reads locally (giving native performance) and > does not write > data until it is fully written locally (delayed write I guess > you could > say). This way you get the replication but still get native > performance. > > Is there a current way to setup GlusterFS like this in order > to get this > 'DRBD-like' functionality? > ___ > Gluster-users mailing list > Gluster-users@gluster.org > http://gluster.org/cgi-bin/mailman/listinfo/gluster-users > ___ Gluster-users mailing list Gluster-users@gluster.org http://gluster.org/cgi-bin/mailman/listinfo/gluster-users