Re: Strange FTP Transfer Rates

From: Bill (bverzal_at_komatsuna.com)
Date: 03/02/04 

Date: 2 Mar 2004 06:20:27 -0800

Bela Gazdy <bela@bluedog.cc.emory.edu> wrote in message news:<c21r9s$74r$1@puck.cc.emory.edu>...
> Bill <bverzal@komatsuna.com> wrote:
> B> Bela Gazdy <bela@bluedog.cc.emory.edu> wrote in message news:<c1niq5$bqd$1@puck.cc.emory.edu>...
> >> FTP transfer between two AIX boxen, both at AIX 5100-05 (ftp server v4.1).
> >> The NIC is the same, IBM 10/100 Mbps Ethernet PCI Adapter, on both, with
> >> 100_Half_Duplex on Box A and 10_Half_Duplex on Box B. Same subnet.
> >>
> >> I ftp-ed the same, ~4.5 MB file back and forth between A and B. Here are
> >> some of the rates I've recorded
> >>
> >> A -> B B -> A
> >> ====== ======
> >> 212.8 5009
> >>
> >> Does anyone have an explanation of 1) the oscillation between two
> >> consecutive, identical transfers, and 2) the dependence (difference) of
> >> the rates on the direction of transfer?
> >>
> >> Any idea would be greatly appreciated!
>
> B> Assuming that you are ftp'ing the same file each time:
>
> B> Pass #1, the ftp is reading the file from disk.
> B> Pass #2, the ftp is reading the file from the FS cache.
>
> B> I assume this is 4 iterations of different files.
>
> B> CPU speed and architecture will affect speed, as will disk config.
> B> When I FTP across my SP switch from the MCA 200MHz node to my S80, it
> B> runs at xxx speed. If I go from the S80 to my 200MHz SP node, it goes
> B> xxxxxxxx fast.
>
> Thanks Bill!
> Okay, cashing makes sense in explaining why two consecutive transfers differs;
> however, how would you explain the vast (10x-40x) discrepancy between the same
> two machines (p170 and p620) in send vs. receive? FTPing the same file from
> the 170 to the 620 is very slow (under 100 Kb/s) while the reverse is
> reasonable (over 1000 Kb/s). It does not really matter which machine's ftp
> server is used, speed seems to be dependent entirely on direction. Could this be
> simply disk I/O-related? The p620 is faster, but not by an order of magnitude.

The 620 would be an order of magnitude faster if it contains the
faster network card. Regular Ethernet (10Mb) usually has a max usable
bandwidth of 80%. So, right there, you are not going to push more
than somewhere aroung 1MB/sec. However, with Fast Ethernet, I've seen
it run in the 12-14MB/sec sustained range, provided it is configured
FULL_DUPLEX. I'd get both servers and switch ports set to FULL_DUPLEX
ASAP.

BV

Relevant Pages

  • Re: What tool to use for processing large documents
    ... cannot parse faster than the disk can read the XML data. ... Reading 10 GB off a disk will take around 3 to 5 minutes ... I forgot to mention that my logs are in zipped xml. ... Get the set of nodes matching an XPath expression. ...
    (comp.text.xml)
  • Re: *Fast* way to process large files line by line
    ... line and parse it. ... reading file loop> marshalling> parsing. ... the disk I am using is an LVM mapped ext3 local disk. ... Btw will using something like an mmap extension for ruby speed things ...
    (comp.lang.ruby)
  • Re: Poor performance from file system
    ... find a specific address on the disk. ... (I believe most hard drives now have multiple platters and multiple ... last reading ... disk temperature. ...
    (comp.os.linux.misc)
  • Re: Scanf and number formats
    ... > before the decimal-point character (which is nonzero if the ... > largely dominated by the disk I/O rate... ... > reading the data from memory, as if it were using getcto fetch ...
    (comp.lang.c)
  • Re: First-timer buying AS/400 and have questions
    ... (writes memory to disk for the next boot). ... I use the Internet to get PTFs and FTP them to the AS/400. ... IBM was thinking bubble memory or some form of solid-state ... Running FTP from the AS/400 you use NAMEFMT 0 to mean the ...
    (comp.sys.ibm.as400.misc)