I was mistakened.. the SCSI layout is like so..
- robot is on its own single ended scsi bus
- there are two drives per controller, of which the controller is a
dual-channel with each drive on its own channel
- The trunking software doesn't seem to make a difference as far as the
througput goes. We are considering changing it out for a Fibre-Channel cards
instead (perhaps two 2gb Fibre Channel Qlogic cards) At that point it may be
worth while to configure the client as a storage node and do san backups.
- I'll have to check out the other stuff ..
Thanks !
On Sun, 11 Jan 2004 22:20:44 +0000
"Mark Bradshaw (BTOpenWorld)" <notthehoople AT BTOPENWORLD DOT COM> wrote:
> Alex,
>
> A few things to check:
>
> - Are you using CDI under NetWorker 7.1? If so, turn it off - this seems to
> introduce some performance problems with LTO drives.
>
> - Could be wrong here but isn't the X6541A an HVD SCSI controller? Are your
> drives HVD? I wasn't aware that HP LTO2 drives came in HVD! If everything
> checks out and *is* the same type of SCSI and you've gone for HVD (rather
> than LVD) then this will be a major bottleneck. HVD usually sustains around
> 20MB/s in total - just under half of what you can expect from a single LTO2
> drive. You need to move your drives so they are each on a single SCSI
> channel to get the best throughput in your config.
>
> - What disk units are you using with your V880 and E4500s? Are these units
> capable of the data rates you are trying to extract from them? If you are
> just using normal, internal disks then you are not going to achieve the
> throughput you want.
>
> - Are you using tape media compatible with your tape drives? Check you are
> using LTO2 tapes *and* that they are certified for use with HP drives. Using
> the wrong tapes can cause slow performance and poor reliability.
>
> - What number and type of CPUs have you installed in the V880? If you are
> trunking your gigabit connections then you will lose a significant amount of
> CPU which will be used just talking to the network. Monitor the CPU load on
> the V880 during backup and see if that is causing a bottleneck. Also monitor
> the CPU load on the E4500 to see what effect the trunking is having on this
> server.
>
> - Can you make the drives run at a good speed using bigasm? This removes
> disk traffic and network traffic from the equation and shows you
> (effectively) what can be written down to the tape via CPU. A good start
> when using new tape drives!
>
> After checking all of this I'd start with a single drive and load it until
> you get a good throughput. If you can't produce enough data to achieve good
> throughput on 1 drive you'll have no chance when using lots of drives! With
> the data rates you have seen so far you really should be able to beat your
> total data rate seen using just a single drive!
>
> A final note - I don't believe the IO architecture of the V880 will manage
> to run all 6 LTO2 drives at full tilt. I haven't done all the sums for this
> box but would expect, based on previous tests done using LTO1, that the
> server will max out at between 3-4 LTO2 drives.
>
> Good luck
>
> Mark
>
> > I tried local backups and I get the same story 3-6 mb/s.
> >
> > The L700 is connected SCSI configuration is like so:
> >
> > - 2 drives per scsi bus (X6541 Dual Differential Ultra/Wide SCSI - pci)
> > thus
> > 3 controllers
> >
> > I'll look into the whole i/o contention theory with the nsr db by using
> > iostat
> > on the box while backing it up.
> >
> > Thanks,
> >
> > Alex
> >
> > On Sun, 11 Jan 2004 12:45:57 -0500
> > Paul Galjan <pgaljan AT YAHOO DOT COM> wrote:
> >
> >> I don't know if you're going to be able to get the advertised 180 MB/s for
> >> all six drives. But the place to start would be to see if you can get
> >> anywhere close to that rate backing stuff up without going over the
> >> network.
> >> See if you can write two data streams at 30 MB/s each, step it up from
> >> there. See when it breaks.
> >>
> >> Another area to check would be to make sure that you are effectively load
> >> balancing over both the NICs.
> >>
> >> Yet another suspect would be the Networker database. You've got a
> >> partition
> >> with 3 MB files, so each drive should be backing up 10 records per second.
> >> That 60 inodes per second, each updating the database. I'm no Networker
> >> database expert, so I can't say for sure if the software can handle that
> >> (perhaps someone on this list can), but you can definitely tell if it's a
> >> hardware bottleneck.
> >>
> >> How is the SCSI bus configured (how many channels are the drives spread
> >> over?).
> >>
> >> --paul
> >>
> >>
> >> ----- Original Message -----
> >> From: "Alex Theodore" <alex AT FUZZYCHEESE DOT COM>
> >> To: <NETWORKER AT LISTMAIL.TEMPLE DOT EDU>
> >> Sent: Sunday, January 11, 2004 11:19 AM
> >> Subject: [Networker] Performace isssues with LTO 2 Ultrium and unable to
> >> write to all tapes at same time
> >>
> >>
> >>> I am only getting about 3-6 megabytes/s when backing up to LTO 2 tape
> >> drives. I would expect that I should get somewhere near the advertised
> >> value, but I am not. Here are the details.
> >>>
> >>> - Legato Network 6.7.1
> >>> - Solaris 8 (latest patches)
> >>> - Sun L700 with 6 HP LTO 2 tape drives
> >>> - Backup Server is a Sun Fire V880
> >>> - Backup Client is Sun Enterprise 4500
> >>> - Network pipe for backups is a dedicated, directly connected (no switch),
> >> dual gigabit ethernet, which is trunked (glued together into 1 virtual
> >> pipe)
> >>>
> >>>
> >>> The data on the client which is being backed up are images (photos) which
> >> are a few megabytes in size. I decided to split up the save sets into 6
> >> different client configurations with the parallelism set to 6 so that all 6
> >> tape drives would be used at the same time during backups but this isn't
> >> happening either.. I can only seem to write to a few tape drives at once..
> >>>
> >>> Any ideas would be greatly appreciated.
> >>>
> >>> TIA,
> >>>
> >>> Alex
> >>>
> >>> --
> >>> Alex Theodore
> >>> alex AT fuzzycheese DOT com
> >>> Boca Raton, FL USA
> >>>
> >>> --
> >>> Note: To sign off this list, send a "signoff networker" command via email
> >>> to listserv AT listmail.temple DOT edu or visit the list's Web site at
> >>> http://listmail.temple.edu/archives/networker.html where you can
> >>> also view and post messages to the list.
> >>> =*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=
> >>
> >> --
> >> Note: To sign off this list, send a "signoff networker" command via email
> >> to listserv AT listmail.temple DOT edu or visit the list's Web site at
> >> http://listmail.temple.edu/archives/networker.html where you can
> >> also view and post messages to the list.
> >> =*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=
> >>
> >
> >
> > --
> > Alex Theodore
> > alex AT fuzzycheese DOT com
> > Boca Raton, FL USA
> >
> > --
> > Note: To sign off this list, send a "signoff networker" command via email
> > to listserv AT listmail.temple DOT edu or visit the list's Web site at
> > http://listmail.temple.edu/archives/networker.html where you can
> > also view and post messages to the list.
> > =*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=
>
> --
> Note: To sign off this list, send a "signoff networker" command via email
> to listserv AT listmail.temple DOT edu or visit the list's Web site at
> http://listmail.temple.edu/archives/networker.html where you can
> also view and post messages to the list.
> =*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=
>
--
Alex Theodore
alex AT fuzzycheese DOT com
Boca Raton, FL USA
--
Note: To sign off this list, send a "signoff networker" command via email
to listserv AT listmail.temple DOT edu or visit the list's Web site at
http://listmail.temple.edu/archives/networker.html where you can
also view and post messages to the list.
=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=
|
