Martin,

 

  We were trying to do some performance tuning to see if we could get
our LTO2 and LTO3 drives working more effectively. The most interesting
thing I ran across was setting the fragment size on the storage unit. If
you don't have any fragment size, that is the best. Given our
circumstances we need to run with fragment sizes on our storage units.
For testing purposes, the system was configured with NET_BUFFER_SZ =
262144, SIZE_DATA_BUFFERS = 262144 and NUMBER_DATA_BUFFERS = 16. The
test data was generated by the HPCreateData tool and is supposed to be
2-1 compressible data.

 

  What I found was that a 1 GB fragment size does not even give an LTO2
drive enough time to come up to its rated speed. It writes for about 6 -
10 seconds, then has to write the EOF marker, then repositions the tape.
Similarly the LTO3 drive does not get up to speed either.

 

  Here's a quick chart of findings. The media server was an IBM P5/550
with 4 CPUs 8GB RAM with 2 LPARs, each getting half the resources.

 

LTO2

Fragment Size         Speed

1 GB                  35 MB/s

2 GB                  40 MB/s

4 GB                  43 MB/s

6 GB                  53 MB/s

7 GB                  52 MB/s

8 GB                  44 MB/s

10 GB                 41 MB/s

12 GB                 49 MB/s

No fragment           58 MB/s   (approximately LTO2 max speed at 2-1
compression)

 

LTO3

1 GB                  41 MB/s

2000 MB               44 MB/s

2 GB                  55 MB/s

4 GB                  51 MB/s

5000 MB               49 MB/s

6 GB                  58 MB/s

8 GB                  63 MB/s

10000 MB              79 MB/s

10 GB                 81 MB/s

12 GB                 61 MB/s

No fragment           82 MB/s

 

  I think the upper limits on my LTO3 speed are due to my inability to
get the AIX tuning correct. My disk subsystem should be able to supply
data at roughly 130 MB/s and the LTO3 drives should be able to handle
that. Something is not right on my tuning.

 

  But I did find it very interesting as to the impact that fragment size
can have on the tape performance.

 

Bryan Bahnmiller

ISD Business Continuity

Pier 1 Imports, Inc

Fort Worth, TX

 

________________________________

From: veritas-bu-bounces at mailman.eng.auburn.edu
[mailto:veritas-bu-bounces at mailman.eng.auburn.edu] On Behalf Of Martin,
Jonathan (Contractor)
Sent: Friday, August 25, 2006 11:58 AM
To: veritas-bu at mailman.eng.auburn.edu
Subject: [Veritas-bu] Maximzing Gigabit Performance

 

All, we're tying to maximize our throughput across gigabit Ethernet and
playing with the buffers etc to try and get faster numbers.

 

Client Buffer

NET_BUFFER_SIZE

NUMBER_DATA_BUFFERS_DISK

SIZE_DATA_BUFFERS_DISK

KB/sec

256

262144

---

---

34000

256

2097152

---

---

39500

256

2097152

16

---

37000

256

2097152

16

262144

36000

512

2097152

64

262144

40480

1024

2097152

64

262144

37608

1024

4194304

64

262144

37473

 

 

Netbackup 5.1 Client --> Gigabit Copper --> Cisco Switch --> Gigabit
Fiber --> Media Server --> Direct Attached SCSI 320 Drive

 

Here's the results of some testing from this morning on one of my
fastest servers (as far as disk/processor/memory) are concerned.  It
looks like my "sweet spot" is about 40MB/sec through Netbackup.  If I
try and FTP several files directly from this server to the Media Server
via the same network config to disk I can get close to 60MB/sec.  Am I
missing some setting here?  Why can't I get even close to duplicating
the 60MB/sec I get via FTP?  My perfmon shows the server's nic only
pushing 45MB/sec or less with drops in communication every 10 seconds or
so.  I'm missing something, any ideas?

 

-Jonathan

-------------- next part --------------
An HTML attachment was scrubbed...
URL: 
http://mailman.eng.auburn.edu/pipermail/veritas-bu/attachments/20060826/e8311f18/attachment-0001.html