There are bold sysadmins and there are old sysadmins, but funny, no bold
and old sysadmins :)
Other than tuning the MTU on the separate NetApp NIC (which turns on
jumbo frames), I verified that the window scale option, etc, were
already turned on.
We have two gigabit ethernet interfaces servicing regular clients (in
Solaris Multi-path redudancy config) and one interface connected to a
vlan with the two NetApps.
At gigabit speed the NIC seems to be less of a issue than the CPU
servicing the NIC's interrupts.
We eventually might move to a HP DL585 server with 4x dual core opteron
especially since HP sells Gigabit server adapters with TOE.
>
> Those are some bold tuning parameters. I like your design of
> a separate VLAN to the filers where jumbo frames can be used.
> Did you need to tune the NetApp in anyway or does it adapt
> nicely to the frame size and TCP window size?
>
> I presume you have another gigabit NIC to the regular clients
> that have no problem with these tcp parameter? Solaris 9 I assume.
>
> On a NetApp restore from several tapes I would think the NIC
> is a bottleneck; something we don't have with the filer
> owning the drives.
>
> Very interesting insight into the jumbo frames the new NDMP
> capabilities.
>
>
> Robert Maiello
> Pioneer Data Systems
>
>
> -----Original Message-----
> From: Matthew Huff [mailto:mhuff AT ox DOT com]
> Sent: Friday, September 16, 2005 12:50 PM
> To: Legato NetWorker discussion; Maiello, Robert
> Subject: RE: [Networker] Running 7.2.1 with NDMP including tuning
>
> Another one of the big advantages of using DSA is that when
> using the local network tape drives it uses variable block
> size, so it's much faster on the tapes.
>
> Just some info about the tuning. We found that we had to set
> up a separate vlan for these backups because the interupt
> overhead on the sparc was so high due to the large number of
> packets. Setting up a separate vlan allowed us to turn on
> jumbo frames, reducing the interrupts by a factor of 4.
>
> We also tuned the gigabit ethernet adapter in the sun by
> setting up an /etc/init.d/tune_ce script that does:
>
> #!/bin/sh
>
> PATH=/bin:/usr/sbin:/usr/sbin;export PATH
>
> ndd -set /dev/tcp tcp_max_buf 4194304
> ndd -set /dev/tcp tcp_cwnd_max 2097152
> ndd -set /dev/tcp tcp_xmit_hiwat 1048576 ndd -set /dev/tcp
> tcp_recv_hiwat 1048576 ndd -set /dev/tcp tcp_tstamp_always 1
> ndd -set /dev/tcp tcp_wscale_always 1
>
> ----
> Matthew Huff | One Manhattanville Rd
> Director of Operations | Purchase, NY 10577
> OTA LLC | Phone: 914-460-4039
> http://www.otaotr.com | Fax: 914-460-4139
>
> > -----Original Message-----
> > From: Legato NetWorker discussion
> > [mailto:NETWORKER AT LISTSERV.TEMPLE DOT EDU] On Behalf Of Maiello, Robert
> > Sent: Friday, September 16, 2005 10:14 AM
> > To: NETWORKER AT LISTSERV.TEMPLE DOT EDU
> > Subject: Re: [Networker] Running 7.2.1 ???
> >
> > Great, that looks like a real good design. It sounds like the V480
> > is dedicated to this?
> >
> > Do you find this was faster then sending the NetApp single
> streamed to
> > several tape drives over the SAN?
> >
> > You have other non-NDMP/regular savesets multiplexed with this?
> >
> >
> > Robert Maiello
> > Pioneer Data Systems
> >
> >
> > -----Original Message-----
> > From: Matthew Huff [mailto:mhuff AT ox DOT com]
> > Sent: Friday, September 16, 2005 9:41 AM
> > To: Legato NetWorker discussion; Maiello, Robert
> > Subject: RE: [Networker] Running 7.2.1 ???
> >
> > The NDMP device backup is sent across the network. The data
> is in the
> > format that the NDMP backup creates (dump for NetApps for example),
> > but is encapsulated in a Legato datastream. Hence you can
> save it in
> > the same pool and same media at the same time as other backups, and
> > you can get improved parallelism.
> >
> > We have two NetApp F920 clustered with about 4-5TB of data.
> > We backup across a dedicated Gig-e pipe running Jumbo
> frames and other
> > tcp tuning to a Sun V480 connected with a fibre HBA to a StorageTek
> > L180 with 6 x IBM LTO-2 drives. Since we upgrade to 7.2 and
> moved the
> > drives to the Sun, and finished tuning, our backup time
> (even over the
> > network) has improved by about 30%. The complete backup takes about
> > 6-7 hours.
> >
> > ----
> > Matthew Huff | One Manhattanville Rd
> > Director of Operations | Purchase, NY 10577
> > OTA LLC | Phone: 914-460-4039
> > http://www.otaotr.com | Fax: 914-460-4139
> >
> > > -----Original Message-----
> > > From: Legato NetWorker discussion
> > > [mailto:NETWORKER AT LISTSERV.TEMPLE DOT EDU] On Behalf Of Robert Maiello
> > > Sent: Friday, September 16, 2005 9:23 AM
> > > To: NETWORKER AT LISTSERV.TEMPLE DOT EDU
> > > Subject: Re: [Networker] Running 7.2.1 ???
> > >
> > > When you say "non-NDMP" device, does that imply that the
> data goes
> > > over the network to the storage node and then to the tape drive?.
> > >
> > > Currently our NDMP devices are owned by the filer.. if
> the devices
> > > are non-NDMP (but can take an NDMP stream) and are owned by the
> > > storage node, does that not imply the NDMP stream will go
> over the
> > > network to
> > > the storage node? Currently they go right to tape not
> > touching the
> > > network.
> > >
> > >
> > > Sorry if I mis-understood the thread.
> > >
> > >
> > > Robert Maiello
> > > Pioneer Data Systems
> > >
> > >
> > > On Thu, 15 Sep 2005 16:57:04 -0400, Stan Horwitz <stan AT TEMPLE DOT EDU>
> > > wrote:
> > >
> > > >On Sep 15, 2005, at 2:05 PM, Yura Pismerov wrote:
> > > >
> > > >>
> > > >>
> > > >> Stan Horwitz wrote:
> > > >>
> > > >>> On Sep 13, 2005, at 5:46 PM, Jimeee Womack wrote:
> > > >>> Last night, we did our first test of an NDMP recover. We
> > > >>> successfully recovered approx. 450GB worth of data to a
> > > test server
> > > >>> via NDMP with SnapImage. The test worked fine. As I
> write this
> > > >>> message, I am backing
> > > >>>
> > > >>
> > > >> Is it true that 7.2.1 can now do NDMP backups without
> > > employing the
> > > >> Snapimage ?
> > > >
> > > >I tried doing a backup of around 450GB that way last night.
> > > Although I
> > > >have not had time to test that backup yet, it seems to have
> > > run fine on
> > > >a non-NDMP device.
> > > >I am going to try cloning the data tomorrow.
> > > >
> > > >To sign off this list, send email to
> > listserv AT listserv.temple DOT edu and
> > > type "signoff networker" in the
> > > >body of the email. Please write to
> > > >networker-request AT listserv.temple DOT edu
> > > if you have any problems
> > > >wit this list. You can access the archives at
> > > http://listserv.temple.edu/archives/networker.html or
> > > >via RSS at http://listserv.temple.edu/cgi-bin/wa?RSS&L=NETWORKER
> > > >=============================================================
> > > ==========
> > > >==
> > >
> > > To sign off this list, send email to listserv AT listserv.temple DOT edu
> > > and type "signoff networker"
> > in the body
> > > of the email. Please write to
> > networker-request AT listserv.temple DOT edu if
> > > you have any problems wit this list. You can access the
> archives at
> > > http://listserv.temple.edu/archives/networker.html or via RSS at
> > > http://listserv.temple.edu/cgi-bin/wa?RSS&L=NETWORKER
> > >
> >
> > To sign off this list, send email to
> > listserv AT listserv.temple DOT edu and type "signoff networker"
> in the body
> > of the email. Please write to
> networker-request AT listserv.temple DOT edu if
> > you have any problems wit this list. You can access the archives at
> > http://listserv.temple.edu/archives/networker.html or via RSS at
> > http://listserv.temple.edu/cgi-bin/wa?RSS&L=NETWORKER
> >
>
>
>
To sign off this list, send email to listserv AT listserv.temple DOT edu and
type "signoff networker" in the
body of the email. Please write to networker-request AT listserv.temple DOT edu
if you have any problems
wit this list. You can access the archives at
http://listserv.temple.edu/archives/networker.html or
via RSS at http://listserv.temple.edu/cgi-bin/wa?RSS&L=NETWORKER
|