Would it make sense to virtualize windows using Xen or VMware ESX and expose storage on a JBOD as NFS, instead of exposing storage as CIFS/SMB for better I/O throughput?

Is it true that the below can yield only 20MB/sec for CIFS/SMB; 80MB/sec for NFS? assuming i use 2-3 servers (1Gbps NIC), NIC teaming, switch binding, jumbo frames etc...

My test rig can only afford:

  • Storage server: HP DL370G6: 2x Xeon 55xx -CPU, 16GB RAM, 300GB 10K RPM SAS, P410i - 512MB cache
  • Windows server: HP DL580: 2x Xeon 56xx, 64GB RAM, 1G NIC
  • HP procurve 3500 layer3 switch - 1GB

Planning to use distributed file system: GlusterFS (user space) over Redhat ent Linux 5.5/ext3 FS, but I am attracted by a Solaris post about CIFS @ 1Gbytes/sec !

http://blogs.sun.com/brendan/entry/cifs_at_1_gbyte_sec

I am torn between the two... I know my hardware is not at par with above post, but does OpenSolaris offer better CIFS/SMB perf than any other?

Thanks in advance for your thoughts!

link|improve this question

77% accept rate
feedback

1 Answer

I have a few Dell 2950's with lesser specs than your machines running Debian Lenny and they can easily max two bonded gigabit links using Samba. All of them have big SATA RAID-6's attached to them so they're not IOPS kings, but with large reads they keep the pipes filled.

That blog you linked was 1 giga*byte*/s.

link|improve this answer
yup, blog was about 1GB/s performance, any computer from the past 2 years with 4 1TB SATA disks can sustain 1 gigabit/s reads – Hubert Kario Oct 4 '10 at 18:10
thank guys, corrected the post about bytes! what kind of a switch were you using? – John-ZFS Oct 5 '10 at 1:11
feedback

Your Answer

 
or
required, but never shown

Not the answer you're looking for? Browse other questions tagged or ask your own question.