Re: LAN Performance Not Up to Expectations
- From: "Cliff Galiher" <cgaliher@xxxxxxxxx>
- Date: Mon, 28 Apr 2008 16:24:44 -0600
The 3GB rating on sata (or cable) is only good for burst reads where data is already in the cache.
1) You aren't reading data. You are writing.
2) You aren't bursting. 108GB backup is definitely an extended write.
The fact that your network monitor reported on 5% usage indicates that you are not saturating the network...which, as your subject implies you think is the problem. I still stand by the assertion that your bottleneck is not the network. And I still specifically feel like you aren't taking into account that the HD on your backup machine is still a workstation class HD. Even high-end server-class hard drives (10,000rpm instead of 7,200rpm) won't get to 100MB/s, no less 3GB. Just won't happen.
Now, re-read that link I sent you. Specifically look under the "File information" section. The second paragraph, right below the table of files, is a sentence talking about the /FU switch.
-Cliff
"Bilbofakeemailhal-pc.org>" <wlp<fauxatdot> wrote in message news:unpb14tib9pg4fbifsl2fkei8ns4mfahoo@xxxxxxxxxx
Hi Cliff,
Well, I had replaced the DELL HDD with a Seagate ST3320620AS ( 3.0 GB
SATA) drive running at 7200 RPM and the SATA cable is rated at 3.0 GB.
I didn't see anything about /FU in the kbid=839272 you cited.
I really wasn't expecting to saturate the Gigabit LAN but I did expect
to see more than 5 % utilization. That really floored me.
On Mon, 28 Apr 2008 00:54:46 -0600, "Cliff Galiher"
<cgaliher@xxxxxxxxx> wrote:
That doesn't sound terribly low to me. You have a few bottlenecks you
didn't take into account. The first, and most important, is the I/O on the
CLIENT'S hard drive. Even high-end workstation hard drives only handle a
sustained 60MB/s, and dell doesn't put high-end drives in their dimension
series PC's.
Tack on some network overhead, and finally just the fact that NTBackup isn't
exactly efficient, and you end up right where you are at. NTBackup was
originally written for tapes, so that is how it thinks about backups. That
can be improved, to some extend, by using the /FU flag which was added to
NTBackup as of service pack 1. Check here for more info on it...
http://support.microsoft.com/?kbid=839272
I'll feel I need to mention that a workstation PC would only compare to the
cheapest of NAS devices. Most business-level NAS devices will be using
multiple hard drives in an some form of RAID array to get better effective
I/O bandwidth, and you can always upgrade to a third-party backup app if you
need better performance, but since SBS targets the small business, NTBackup
usually does the job. I haven't had a chance to do any performance testing
on 2k8 yet, so I'll be curious to see how NTBackup's successor fixes its
problems. :)
In essence, it is difficult for a workstation to saturate a gigabit
connection for any but the smallest bursts of traffic. I don't believe your
network is the cause of the numbers you are seeing. A server with a
high-load RAID servicing multiple clients can easily use it's full gig
outbound. So I don't want you to think that deploying gigabit ethernet is a
waste, just your method of testing by writing to a single-drive client
machine was not adequate to take advantage of it.
-Cliff
"Bilbofakeemailhal-pc.org>" <wlp<fauxatdot> wrote in message
news:16la1452u7g3qrai2gr99rkaotebvmhnir@xxxxxxxxxx
A small SBS2K3 with latest service packs. Server runs a 1.8 GHz
Pentium 4 with 3 GB PC3200 DDR RAM on an Intel D865GBF Mobo.
Server has two NICs. The LAN NIC is an Intel Pro/1000 GT Desktop
Adapter. The server is running 3.0 GB SATA Hard Drives and the LAN
was mostly quiescent during the test.
One domain client and three peer clients (who merely use the routing
services of the server). All but one client has Intel Pro/1000 GT
Desktop adapter, the other having USR Gigabit Ethernet PCI Adapter.
The LAN infrastructure is CAT-6 with a HP ProCurve 1800-8G Switch.
All these connections are reported by the switch as running at nominal
Gigabit speeds. No cable run exceeds 75 feet.
I was running NTBACKUP on the SBS server with backup target on one of
the peer client PCs as an experiment into the likely performance of a
NAS unit of some kind. I feel that the client PC was effectively a
NAS device during this run. It is a DELL Dimension 4700 with 3 GHz
Pentium 4, 4 GB PC2-3200 DDR2 RAM, and the USR NIC running WinXP SP2.
The only obvious bottleneck I can think of is the fact that the server
is running Trend Micro's Client/Server Messaging Security for SMB
(v3.6). The Client Anti-Virus was disabled for this test. It seems
to me that this rig should have really performed pretty well.
The network utilization viewed from the server and the target client
was only between 5 and 6 %. The size of the backup was about 108 GB
on disk and the backup took around five to hours forty-five minutes.
I reckon the net throughput to have been about 43 Mb/sec. Now,
allowing for all kinds of overheads and the like, this seems to be
pretty pitiful backup performance. There were NO events logged
indicating any network problems on either end of the test.
Does anyone have any insights to offer? How does one go about
troublehooting this kind of issue?
Thanks,
--
BilBo
--
BilBo
.
- Follow-Ups:
- Re: LAN Performance Not Up to Expectations
- From: Bilbo
- Re: LAN Performance Not Up to Expectations
- References:
- LAN Performance Not Up to Expectations
- From: Bilbo
- Re: LAN Performance Not Up to Expectations
- From: Cliff Galiher
- Re: LAN Performance Not Up to Expectations
- From: Bilbo
- LAN Performance Not Up to Expectations
- Prev by Date: Re: Network very slow
- Next by Date: Re: FAX Service - clean installation
- Previous by thread: Re: LAN Performance Not Up to Expectations
- Next by thread: Re: LAN Performance Not Up to Expectations
- Index(es):
Relevant Pages
|
