Poor performance with Star

Poor performance with Star

Post by Stu » Sat, 24 Jun 2006 23:29:46



Hi all I was hoping you can help.

I'm getting poor performance using star to copy some large files from a
remote NFS server over a wan (speed 100 mb/s) to local ufs partitions.

My hardware is (inherited from previous admin who clearly didn't like
upgrades) a Sun E3500 with dual USII cpu & 2GB RAM  running:

Release: Solaris 8 s28_38shwp2 SPARC
Kernel Rev: 108528-14

The locally attached storage is a:

SENA (not sure of model, any suggestions?) with a mixture of 36gb, 18gb
and 9gb 10000rpm disks.  The firmware is at 1.09.
HBA=Sun 501-3060 1GB SBus Dual FC's.
I'm using Veritas VxVM 3.2, the disks in the SENA are striped and
mirrored.

I'm using the command:

star -copy -no-fsync -nowarn -time -fs=128M -fifostats <file> <target
dir>, but am only getting approx ~900kb/s written, here are some
examples:

+ star -copy -no-fsync -nowarn -time -fs=128M -fifostats p1_index01.dbf
/serva/disk_4/oracle/oradata/IRSU1
star: fifo had 11 puts 11 gets.
star: fifo was 4 times empty and 3 times full.
star: fifo held 134225920 bytes max, size was 134225920 bytes
star: 0 blocks + 419442176 bytes (total of 419442176 bytes =
409611.50k).
star: Total time 457.739sec (894 kBytes/sec)
+ star -copy -no-fsync -nowarn -time -fs=128M -fifostats p2_index01.dbf
/serva/disk_4/oracle/oradata/IRSU1
star: fifo had 11 puts 11 gets.
star: fifo was 4 times empty and 3 times full.
star: fifo held 134225920 bytes max, size was 134225920 bytes
star: 0 blocks + 419442176 bytes (total of 419442176 bytes =
409611.50k).
star: Total time 463.958sec (882 kBytes/sec)
+ star -copy -no-fsync -nowarn -time -fs=128M -fifostats p3_index01.dbf
/serva/disk_4/oracle/oradata/IRSU1
star: fifo had 11 puts 11 gets.
star: fifo was 4 times empty and 3 times full.
star: fifo held 134225920 bytes max, size was 134225920 bytes
star: 0 blocks + 419442176 bytes (total of 419442176 bytes =
409611.50k).
star: Total time 457.527sec (895 kBytes/sec)

I'm getting lots of these errors in /var/adm/messages

Transport error on target=0x53:  Fibre Channel CRC Error on frames

but none of the disks are reporting as errored in iostat or format.

I know I'm down rev'd on Solaris and most likely firmware revisions on
the disks, but can anyone see if I'm doing something really, really
wrong before I embark on a big upgrade plan?

Thanks

Stuart

 
 
 

Poor performance with Star

Post by Bruce Esquibe » Sun, 25 Jun 2006 00:35:39


: I'm getting poor performance using star to copy some large files from a
: remote NFS server over a wan (speed 100 mb/s) to local ufs partitions.

Well, my opening question is, are you implying that it's just star that is
the problem? Like a plain old cp from one disk to another in the SENA is
much faster or what?

: SENA (not sure of model, any suggestions?) with a mixture of 36gb, 18gb
: and 9gb 10000rpm disks.  The firmware is at 1.09.
: HBA=Sun 501-3060 1GB SBus Dual FC's.
: I'm using Veritas VxVM 3.2, the disks in the SENA are striped and
: mirrored.

I'm not an expert with FC drives but have been using them for a few years on
Sun hardware and the above is likely the problem.

The mixture of drive sizes is ok, but if it's mix and match non-sun firmware
on them, just isn't going to work well. Those frame errors you saw in the
syslogs are pointing to that.

I'm not sure if FC is like the older SCSI where the bus slows down to the
lowest common denominator, but it seems something like that is going on.
I've bought and tinkered with dozens of ebay specials (shit, for under $40 a
pop for a 73GB with shipping, why not) but haven't determined any proof yet.

I mean for example, one batch of 10 I bought there seemed to be these 3
Seagate OEM's, firmware was 0003. Those 3 drives by themselves were reaching
the limit (I think) of the whole works, close to 20Mb/sec in raid5
configuration, over 35Mb/sec stripped. The problem was, even if I just stuck
in one of the majority NA07 firmware ones, everything went to shit (under
1Mb/sec even stripped).

And I'm talking about just sticking it in to the array, not configuring it
or anything. The NA07's by themselves work fine, not as good as the 0003's
but acceptable for what we're using them for (mostly webhosting).

Keep in mind I am talking about exactly the same drive type (like ST373405FC
or something) and size. Just the firmware is different but seems to vary by
a factor of 10x~15x performance depending "how well they play with others".

By the way, as long as this popped up, has anyone ever tinkered with
changing the firmware, besides Sun upgrades?

There seems to be this program distributed by Seagate (seaview enterprise or
something) but it's for Windows or Linux only. Supposed to be like a OEM
firmware upgrade/editor/formatter for FC drives.

I know it borders on piracy issues but was wondering if on FC it was
possible to "suck out" one firmware and stick it in another one that has
something else in it. Or if there was more to it than that. Can't really
find any good midnight engineering groups anymore.

Anyway, is my two cents on your problem, if the drives are a random set that
ended up in your SENA, it just isn't going to work well. If they are all Sun
branded/Sun firmware, will have to let others chime in.

But I do have to say that star is a good test for this, seems to find
bottlenecks better and faster than test suites. JS did a pretty good thing
there in creating it.

-bruce


 
 
 

1. XP -> linux: poor performance

Hi all,

I was wondering if anyone could help me with the following problem. I
have two x86 machine connected
to eachother. One is running Windows XP and the other Redhat 7.3. Both
machines are connected
through a crosscable. The Windows machine has a 3Com 3c905c-tx card
while the Linux machine used a
Intel Pro/100+ NIC. I have used the latest and greatest drivers (version
4.16.0.0 for the 3com NIC and the
2.1.15 e100 module from Intel). When I transfer a file using FTP from
the Linux box to the Windows
box, it reaches 85 Mbps, which is what I would expect. When I do the
opposite, from the Windows
box to the Linux FTP server, it reaches only about 25 Mbps. CPU
utilization is under 100% and both
cards are configured at 100full. Does anyone have a clue of what's going
on here?
Thanks in advance.

Regards,

Vincent Bodenstab

2. How to use UserDir directive in CERN WWW Server

3. very poor performance in 2.5.66[-mm1]

4. blocking gator and it's kin at the firewall

5. Poor disk performance sparc20 2.4

6. simple,short script i dont understand

7. DDS-2 tape drive poor performance or malfunction

8. problems installing gcombust (error message about gtk)

9. poor TCP performance between Sol2.4 and Sunos4.x

10. FreeBSD Gateway POOR performance

11. Poor Internet performance under kppp

12. Poor networking performance

13. Poor Ethernet Performance with Linux