Does anyone know of a little utility that would test the speed of the
connection between my computers on my network? Please don't suggest Ping,
that pings the other computer 4 times and then reports 0 milliseconds. :-)

Al

Alfred Kaufmann wrote:

> Does anyone know of a little utility that would test the speed of the
> connection between my computers on my network? Please don't suggest Ping,
> that pings the other computer 4 times and then reports 0 milliseconds. :-)
>
> Al
>
>

Speed means more than one thing:

Ping measures and reports latency. 0 mS is pretty good. ;-)

XP's task manager reports connection speed via the Network tab.

The most useful speed parameter IMHO is disk-disk STR. To measure STR,
create a shared folder on PC-A and map a network drive to it from
PC-B, then create a large contiguous file (e.g., a 1 GB file for a
100 Mb/s network link) on that shared folder, then copy it using XP's
Explorer ^C and ^V. Use a watch to measure transfer time, and then
calculate the STR.

Repeat using the shared folder on PC-B and the mapping on PC-A, and
repeat for every pair of PCs on your network. Depending on the end-nodes,
you may get four different STRs: the four combos of pushing v. pulling
data, and source on A v. source on B.

With decent XP PCs on a 100 Mb/s net via a Linksys router, I get >75%
efficiency in all four cases, but YMMV.

--
Cheers, Bob

Alfred Kaufmann wrote:
> Does anyone know of a little utility that would test the speed of the
> connection between my computers on my network? Please don't suggest Ping,
> that pings the other computer 4 times and then reports 0 milliseconds. :-)
>
> Al

http://www.bandwidthplace.com/speedtest/

This will test your bandwidth to the net. Your network card should
report what it is connected at (10/100/1000). Is this what you mean?

-Jim

"Bob Willard" <(E-Mail Removed)> wrote in message
news:(E-Mail Removed)...

> Ping measures and reports latency. 0 mS is pretty good. ;-)

Yes, I like that too. ;-)

> XP's task manager reports connection speed via the Network tab.
>
> The most useful speed parameter IMHO is disk-disk STR. To measure STR,
> create a shared folder on PC-A and map a network drive to it from
> PC-B, then create a large contiguous file (e.g., a 1 GB file for a
> 100 Mb/s network link) on that shared folder, then copy it using XP's
> Explorer ^C and ^V. Use a watch to measure transfer time, and then
> calculate the STR.

I did that and trnaferred a 3GB file from one computer to the other. The
Network Tab in the task manager showed about an average utilization of 17%.
That works out to about 0.17Gbps which is much less than what I expected.
I did it again and timed it manually and it worked out to about the same. I
thought about what could be causing this bottleneck, the ethernet cards and
router are supposed to handle 1Gbps, my cpu is operating around 13% when
this transfer is happening so it is not being overloaded, that leaves the
hard drives. I have Sata hard drives (1.5Gbps) connected into a mirrored
array on both systems so they should not be causing a bottleneck.

Then I tried to copy this 3GB file from one hard drive to another in the
same machine and I timed it manually; it took 2 minutes 13 seconds ~ or
0.18Gbps. Obviously I can't expect to transfer data from one machine to the
other faster than I can internally. Now I need to figure out why my
internal hard transfers are so low compared to the advertised maximum.

I bought this so that I could use Remote Desktop and I am very happy with
that.

Al

"jtpr" <(E-Mail Removed)> wrote in message
news:(E-Mail Removed)...

> http://www.bandwidthplace.com/speedtest/
>
> This will test your bandwidth to the net. Your network card should
> report what it is connected at (10/100/1000). Is this what you mean?
>
> -Jim
>

My speed to the net is pityful and that is not going to change anytime soon,
I was testing my home lan network. Windows XP reports I am connected at
1.0Gbps; it is nice to see but when you only get 0.18Gbps you may have
wasted your money.

Al

Alfred Kaufmann wrote:

> "Bob Willard" <(E-Mail Removed)> wrote in message
> news:(E-Mail Removed)...
>
>
>>Ping measures and reports latency. 0 mS is pretty good. ;-)
>
>
> Yes, I like that too. ;-)
>
>
>>XP's task manager reports connection speed via the Network tab.
>>
>>The most useful speed parameter IMHO is disk-disk STR. To measure STR,
>>create a shared folder on PC-A and map a network drive to it from
>>PC-B, then create a large contiguous file (e.g., a 1 GB file for a
>>100 Mb/s network link) on that shared folder, then copy it using XP's
>>Explorer ^C and ^V. Use a watch to measure transfer time, and then
>>calculate the STR.
>
>
> I did that and trnaferred a 3GB file from one computer to the other. The
> Network Tab in the task manager showed about an average utilization of 17%.
> That works out to about 0.17Gbps which is much less than what I expected.
> I did it again and timed it manually and it worked out to about the same. I
> thought about what could be causing this bottleneck, the ethernet cards and
> router are supposed to handle 1Gbps, my cpu is operating around 13% when
> this transfer is happening so it is not being overloaded, that leaves the
> hard drives. I have Sata hard drives (1.5Gbps) connected into a mirrored
> array on both systems so they should not be causing a bottleneck.
>
> Then I tried to copy this 3GB file from one hard drive to another in the
> same machine and I timed it manually; it took 2 minutes 13 seconds ~ or
> 0.18Gbps. Obviously I can't expect to transfer data from one machine to the
> other faster than I can internally. Now I need to figure out why my
> internal hard transfers are so low compared to the advertised maximum.
>
> I bought this so that I could use Remote Desktop and I am very happy with
> that.
>
> Al
>
>

Ah. You should know that 1.5 Gb/s is the speed of the SATA interconnect, not
the speed of the HDs. My pretty decent SATA HD (a 10K RPM 74GB Raptor) peaks
at ~72 MB/s STR on the outer cylinders.

Your wording makes me wonder if you are expecting mirroring to deliver better
single-stream performance than a single HD, which would be wrong. With a
decent implementation, a MirrorSet (RAID-1) should have about the same STR
as a single HD (assuming equal HDs); no better, but not much worse. But a
StripeSet (RAID-0) should deliver nearly double the single HD.

3GB/2:13 ~= 3000MB/133secs ~= 22 MB/s. That's well under what I'd expect for
current HDs, but I don't have a guess as to why. I suggest downloading HDtach
to see what your HDs actually deliver for STR on your PC; remember to run it
in the LongBench mode, with your PC as close to stand-alone as you can make it.
--
Cheers, Bob

"Bob Willard" <(E-Mail Removed)> wrote in message
news:(E-Mail Removed)...

> Ah. You should know that 1.5 Gb/s is the speed of the SATA interconnect,
> not
> the speed of the HDs. My pretty decent SATA HD (a 10K RPM 74GB Raptor)
> peaks
> at ~72 MB/s STR on the outer cylinders.

Yes the sustained read speed of my Maxtors is nowhere near the speed of
their SATA connection. :-(

> Your wording makes me wonder if you are expecting mirroring to deliver
> better
> single-stream performance than a single HD, which would be wrong. With a
> decent implementation, a MirrorSet (RAID-1) should have about the same STR
> as a single HD (assuming equal HDs); no better, but not much worse. But a
> StripeSet (RAID-0) should deliver nearly double the single HD.

HDTach tells me am getting 103.4 MB/s average read and that is almost double
the single drive sustained read .

> 3GB/2:13 ~= 3000MB/133secs ~= 22 MB/s. That's well under what I'd expect
> for
> current HDs, but I don't have a guess as to why.

The drive I was copying to was an old drive - 31.9 MB/s average read. When
I copy the file from one partition to another on the raid drive it only took
85 seconds or ... 3000MB/85 = 35.3MB/s. That is a bit faster but not near
the 118.7MB/s above. This is probably caused by the hard drive switching
from read to write mode.

Al

Alfred Kaufmann wrote:


> The drive I was copying to was an old drive - 31.9 MB/s average read. When
> I copy the file from one partition to another on the raid drive it only took
> 85 seconds or ... 3000MB/85 = 35.3MB/s. That is a bit faster but not near
> the 118.7MB/s above. This is probably caused by the hard drive switching
> from read to write mode.
>
> Al
>
>

The primary cause of STR loss when copying a large file from one part to
another part on the same HD is seeks, not R-to-W switching. Under DOS or
WinWhatever, a large file copy is a sequence of: ReadSeek, ReadChunk,
WriteSeek, WriteChunk, RepeatUntilDone. If the file size is much larger
than the size of the Chunk, then a lot of time will be "wasted" seeking;
particularly since the HD's built-in ReadAhead and WriteBehind caching
enhances the raw R/W performance. To see this, do a part=>part copy of a
large file, and then do a HD=>HD copy, and look at the STR difference.

No, I don't know what the current Chunk size is with XP. And yes, there
are lots of tricks M$ could deploy to improve copy speed: bigger Chunks,
smarter caching, and asynchronous I/O in the app to take advantage of
HD command queuing. Not clear to me, however, that copy speed is something
that M$ should focus on.
--
Cheers, Bob

Alfred Kaufmann wrote:


>>Your wording makes me wonder if you are expecting mirroring to deliver
>>better
>>single-stream performance than a single HD, which would be wrong. With a
>>decent implementation, a MirrorSet (RAID-1) should have about the same STR
>>as a single HD (assuming equal HDs); no better, but not much worse. But a
>>StripeSet (RAID-0) should deliver nearly double the single HD.
>
>
> HDTach tells me am getting 103.4 MB/s average read and that is almost double
> the single drive sustained read .

You've managed to violate the laws of physics; perpetual motion is next. 8-)

Simple mirroring just does not enhance single-stream read performance of a
sequential file. I can think of two possibilities:

1. You actually have RAID-0 (StripeSet) instead of a RAID-1 MirrorSet; or, maybe
a RAID-01 or RAID-10 combination. For a single-stream read of a large file,
RAID-0/01/10 could nearly double the STR.

2. You have a RAID-1 (MirrorSet) with enough total cache (between the two HD's
read-ahead caches and the RAIDbox's cache) to fool HDtach. To confirm or
deny this, create a HUGEFILE, then fire up a Command window (looks like
DOS under XP, but it ain't), and do COPY HUGEFILE NUL, timing the copy
with your watch to calculate the STR.
--
Cheers, Bob