Optimize File Transfer on a 100Mbps LAN

newbie123:

Maybe the hard disks can't read faster than 25 Mbps?

newbie123 wrote in part:

No, it is not. 10 Mbit/s is the only drop possible. The interface either runs at 10 or it runs at 100. It cannot run at 20 or any intermediate speed. You have 100 and a bottleneck elsewhere.

There is overhead in all transfers -- checking data for corruption and keeping track of packets. I'm doubt TCP receive window size [wiki't] would come into pure netbios.

That is very old. Modern drives should be at least 7200, and be capable of above that. However, writes are usually slower than reads, and OSes often make them much slower by metadata updates.

-- Robert

You mean a modern drive such as the Western Digital Cavier GP "1 Terabyte" drive, whose spindle speed is somewhere between

The Cavier GP was released less than a year ago. Is that "very old"?

Walter Roberson wrote: (snip)

Higher speeds are mostly to reduce latency. For a given bit density it does also increase transfer rate. In most cases, transfer rate is out of the cache anyway.

The only way to reduce latency (rotational delay) is a faster rotation rate. There are other ways to increase the transfer rate if needed.

-- glen

In article , glen herrmannsfeldt wrote: [re: hard discs]

Increasing the bit density would allow reduction of the platter size, which would result in an average reduction in rotational latency. Using more physically smaller platters has the same effect.

Walter Roberson wrote in part:

No. Density and size are irrelevant. Spin rate matters: For a random seek (which first settles on the track), there's still on average one half a rotation before the desired sector comes under the read head. It doesn't matter how many sectors are on that track.

For track-to-track, the start of next track can be offset by the seek distance from the end of the previous track, cutting the rotational latency down to whatever jitter needs.

-- Robert

Hi All,

Thx for the replies. However I th> Walter Roberson wrote in part:

There is one way to reduce latency without speeding up the disk, which is more than one head per track.

I believe the IBM 2305-1 did that. 2.5ms average latency.

-- glen

glen herrmannsfeldt wrote in part:

Certainly. Some extra trouble moving the longer arms and alignment problems from thermal expansion.

-- Robert

newbie123 wrote in part:

Well, if you think it is the MS-Win TCP/IP stack (and no reason it could not be), then find and run TTCP.EXE on both machines.

It is an old-time bandwidth utility that runs from memory, so disk speed and buffering isn't a variable.

I suspect Ixia Qcheck might have better buffering to improve disk performance.

-- Robert

The 2305 is a fixed head disk. The tracks may be wide enough not to worry about thermal expansion. Then again, they may specify the allowed temperature range. They were mostly used as a paging device for S/370.

-- glen

the modern fix would be use a "flash disk" - no seek time, and no rotational delay.

i saw a 1 or 2U server touted at IBC in Amsterdam last year which claimed up to a Terabyte of flash could be installed.....

Hi Robert,

I ran my tests using TTCP. Here's the results

C:\\>ttcp -t -l500000 10.10.1.10 ttcp-t: local 10.10.1.6 -> remote 10.10.1.10 ttcp-t: buflen=3D500000, nbuf=3D2048, align=3D16384/+0, port=3D5001 tcp ->=

Unless I am reading this wrong it seems to be telling me that I am transferring data at approx 91Mbps. Please correct me if I am wrong.

Any suggesti> newbie123 wrote in part: