[Gluster-devel] Multiple NFS Servers (Gluster NFS in 3.x, unfsd, knfsd, etc.)

Thu Jan 7 09:52:53 UTC 2010

Gordan Bobic wrote:
> Gordan Bobic wrote:
> 
>>> With native NFS there'll be no need to first mount a glusterFS
>>> FUSE based volume and then export it as NFS. The way it has been 
>>> developed is that
>>> any glusterfs volume in the volfile can be exported using NFS by adding
>>> an NFS volume over it in the volfile. This is something that will become
>>> clearer from the sample vol files when 3.0.1 comes out.
>>
>> It may be worth checking the performance of that solution vs the 
>> performance of the standalone unfsd unbound to portmap/mountd over 
>> mounted glfs volumes, as I discovered today that the performance feels 
>> very similar to native knfsd and server-side AFR, but without the 
>> fuse.ko complications of the former and the buggyness of the latter 
>> (e.g. see bug 186: 
>> http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=186 - that 
>> bug has been driving me nuts since before 2.0.0 was released)
>>
>> I'd hate to see this be another wasted effort like booster when there 
>> is a solution that already works.
>>
>>> The answer to your question is, yes, it will be possible to export your
>>> local file system with knfsd and glusterfs distributed-replicated 
>>> volumes
>>> with Gluster NFS translator BUT not in the first release.
>>
>> See comment above. Isn't that all the more reason to double check 
>> performance figures before even bothering?
>>
>> In fact, I may have just convinced myself to acquire some iozone 
>> performance figures. Will report later.
> 
> OK, I couldn't get iozone to report sane results. glfs was reporting 
> things in the reasonable ball park I'd expect (between 7MB/s and 110MB/s 
> which is what I'd expect on gigabit ethernet). NFS was reporting figures 
> that look more like the memory bandwidth so I'd guess that FS-Cache was 
> taking over. With O_DIRECT and O_SYNC figures were in the 700KB/s range 
> for NFS which is clearly not sane because in actual use the two seem 
> fairly equivalent.
> 
> So - I did a redneck test instead - dd 64MB of /dev/zero to a file on 
> the mounted partition.
> 
> On writes, NFS gets 4.4MB/s, GlusterFS (server side AFR) gets 4.6MB/s. 
> Pretty even.
> On reads GlusterFS gets 117MB/s, NFS gets 119MB/s (on the first read 
> after flushing the caches, after that it goes up to 600MB/s). The 
> difference in the unbuffered readings seems to be in the sane ball park 
> and the difference on the reads is roughly what I'd expect considering 
> NFS is running UDP and GLFS is running TCP.
> 
> So in conclusion - there is no performance difference between them worth 
> speaking of. So what is the point in implementing a user-space NFS 
> handler in glusterfsd when unfsd seems to do the job as well as 
> glusterfsd could reasonably hope to?

A single dd, which is basically sequential IO is something even
an undergrad OS 101 project can optimize for. We, on the other hand,
are aiming higher. We'll be providing much better meta-data
performance, something unfsd sucks at(..not without reason, I
appreciate the measures it takes for ensuring correctness..) due to
the large number of system calls it performs, much better support for
concurrency in order to exploit the proliferating multi-cores, much
better parallelism for multiple NFS clients where all of them are
hammering away at the server, again something unfsd does not to do.

-Shehjar

> 
> There is, however a problem iozone showed up - it wouldn't complete on 
> glfs. The glusterfs client would cause it to error out before it 
> finished. iozone would report errors like this:
> 
> Error writing block 2047, fd= 3
> write: Transport endpoint is not connected
> 
> Error writing block 2047, fd= 3
> write: File descriptor in bad state
> 
> Error writing block 2047, fd= 3
> write: File descriptor in bad state
> 
> 
> 
> /home/gordan/test/f1: Transport endpoint is not connected
> /home/gordan/test/f2: Transport endpoint is not connected
> /home/gordan/test/f3: Transport endpoint is not connected
> /home/gordan/test/f4: Transport endpoint is not connected
> 
> 
> On the client, the logs show things like this:
> 
> 
> [2010-01-06 21:42:30] E 
> [client-protocol.c:457:client_ping_timer_expired] home: Server 
> 10.2.0.10:7000 has not responded in the last
> 10 seconds, disconnecting.
> [2010-01-06 21:42:30] E [saved-frames.c:165:saved_frames_unwind] home: 
> forced unwinding frame type(1) op(FSYNC)
> [2010-01-06 21:42:30] W [fuse-bridge.c:888:fuse_err_cbk] glusterfs-fuse: 
> 269780: FSYNC() ERR => -1 (Transport endpoint is not connec
> ted)
> 
> 
> Followed by lots of this:
> 
> [2010-01-06 21:46:39] W [fuse-bridge.c:1540:fuse_writev_cbk] 
> glusterfs-fuse: 532985: WRITE => -1 (Transport endpoint is not connecte
> d)
> 
> and this:
> 
> [2010-01-06 21:53:41] W [fuse-bridge.c:888:fuse_err_cbk] glusterfs-fuse: 
> 537456: FLUSH() ERR => -1 (File descriptor in bad state)
> 
> glfs seems to be rather fragile when load starts approaching wire speed. 
> My ssh sessions running top to the same machine didn't disconnect or 
> show any noticeable lag, so the disconnections are probably uncalled for 
> - there ought to be a more graceful way to deal with it. How often does 
> it send heartbeat packets and how many in a row have to go missing 
> before it decided to disconnect?
> 
> Another thing I noticed is that even though the server glusterfs process 
> was running with it's server side AFR export, the first time I tried to 
> connect to it from the client after some hours of using the NFS mount, 
> the server process appears to have crashed. This was what ended up in 
> it's log:
> 
> 
> [2010-01-06 21:35:54] N [server-protocol.c:7065:mop_setvolume] 
> server-home: accepted client from 10.2.3.1:1023
> [2010-01-06 21:35:54] N [server-protocol.c:7065:mop_setvolume] 
> server-home: accepted client from 10.2.3.1:1022
> pending frames:
> frame : type(1) op(LOOKUP)
> frame : type(1) op(LK)
> 
> patchset: v2.0.9
> signal received: 11
> time of crash: 2010-01-06 21:36:12
> configuration details:
> argp 1
> backtrace 1
> db.h 1
> dlfcn 1
> fdatasync 1
> libpthread 1
> llistxattr 1
> setfsid 1
> spinlock 1
> epoll.h 1
> xattr.h 1
> st_atim.tv_nsec 1
> package-string: glusterfs 2.0.9
> /lib64/libc.so.6[0x3f55e302d0]
> /usr/lib64/glusterfs/2.0.9/xlator/protocol/client.so(client_lookup+0xc8)[0x2afea8f89438] 
> 
> /usr/lib64/glusterfs/2.0.9/xlator/cluster/replicate.so(afr_lookup+0x226)[0x2afea97e3e66] 
> 
> /usr/lib64/glusterfs/2.0.9/xlator/protocol/server.so(server_lookup_cbk+0x513)[0x2afea95cb2a3] 
> 
> /usr/lib64/glusterfs/2.0.9/xlator/cluster/replicate.so(afr_self_heal_cbk+0x8e)[0x2afea97e46fe] 
> 
> /usr/lib64/glusterfs/2.0.9/xlator/cluster/replicate.so(afr_sh_data_done+0xbe)[0x2afea97f8bce] 
> 
> /usr/lib64/glusterfs/2.0.9/xlator/cluster/replicate.so(afr_sh_data_flush_cbk+0x44)[0x2afea97fa284] 
> 
> /usr/lib64/glusterfs/2.0.9/xlator/cluster/replicate.so(afr_sh_data_utimes_cbk+0x9)[0x2afea97fa2a9] 
> 
> /usr/lib64/glusterfs/2.0.9/xlator/protocol/client.so(client_utimens_cbk+0x14e)[0x2afea8f9152e] 
> 
> /usr/lib64/glusterfs/2.0.9/xlator/protocol/client.so(protocol_client_pollin+0xca)[0x2afea8f808aa] 
> 
> /usr/lib64/glusterfs/2.0.9/xlator/protocol/client.so(notify+0x212)[0x2afea8f874e2] 
> 
> /usr/lib64/glusterfs/2.0.9/transport/socket.so(socket_event_handler+0xd3)[0x2aaaaaaafe33] 
> 
> /usr/lib64/libglusterfs.so.0[0x3f56a27115]
> /usr/sbin/glusterfs(main+0xa06)[0x403e96]
> /lib64/libc.so.6(__libc_start_main+0xf4)[0x3f55e1d994]
> /usr/sbin/glusterfs[0x402509]
> ---------
> 
> None of the above went wrong when using the unfsd mount - and that 
> really doesn't look very confidence inspiring in a stable release (2.0.9).
> 
> Gordan
> 
> 
> _______________________________________________
> Gluster-devel mailing list
> Gluster-devel at nongnu.org
> http://lists.nongnu.org/mailman/listinfo/gluster-devel