[Gluster-users] Usage Case: just not getting the performance I was hoping for

Arnold Krille arnold at arnoldarts.de
Fri Mar 16 21:20:22 UTC 2012


On 16.03.2012 20:02, D. Dante Lorenso wrote:
> Meanwhile, I'm gonna keep working with Gluster and see if I can get the
> performance. Recently I converted to using Linux Raid 10 on 4 1TB drives
> and now I'm getting 310 MB/s write speed to my brick using "dd" test.
> That's looking better and getting closer to maxing out a 1GbE link. I
> need to put gluster on top of this and see if I can continue the
> throughput all the way to the client mount.
>
> As another test, I'm planning to buy some 480 GB SSD drives which can do
> 300+ MB/s each. I'm thinking if I build a Raid 10 configuration with
> those, I might be able to push upwards of 1000+ MB/s. Then, let Gluster
> and/or Samba sit on top of that and we'll see what's what.

I think you are confusing the ~100MB/s of (single) disks with the 1Gb/s 
of single Gigabit-ethernet. The later is 1 giga-bits per second, which 
comes to 1024 Mega-bits per second and thus to ~120 Mega-bytes per 
second aka 120 MB/s. Long story short: copying a single disk via gigabit 
lan leaves only little for protocol-overhead and your ssh connection. No 
need to go to higher disk rates when your link-speed is only 1GBits...

And your (or your users) expirience of nfs/smb over gluster will mostly 
be influenced by the seek-time for small files or the accesses of many 
users, not the transfer-time of single-files for single users. Mirroring 
across network, this seek-time (aka latency) is mostly influenced by the 
round-trip-latency of your network. And multiplying your network-rate by 
ten sadly won't give you a tenth of the latency.

Have fun,

Arnold
-- 
Dieses Email wurde elektronisch erstellt und ist ohne handschriftliche 
Unterschrift gültig.



More information about the Gluster-users mailing list