REPOST: Performance problems with FTP

[Bill Moran]

Charles Swiger <cswiger at mac.com> wrote:

> On Jun 29, 2004, at 2:02 PM, Bill Moran wrote:
> > I'm having some really weird problems with ftp performance on FreeBSD 
> > 4.10
> > that I just can't seem to figure out.
> >
> > My uplink here maxes out at about 35k/sec.  If I scp to the machine in 
> > question
> > I get about 30k/sec (which is expected) but when I ftp, I never get 
> > anything
> > better than 15k/sec, and occasionally as low as 8k/sec.
> [ ... ]
> > I tried running the ftp daemon that ships with FreeBSD, as well as 
> > Proftpd and
> > they both exhibit the same performance issues.
> 
> I haven't seen any signs of such problems with 4.10.  Given that you've 
> reproduced this using different FTP servers, it seems more likely to be 
> a network issue or some hardware glitch (cables? flaky NIC?) than a 
> software issue.

I hadn't even considered hardware, becuase scp is always as fast as I
would expect.  I wouldn't suspect a NIC or HDD, unless the problem was
consistent with _all_ file transfers, and it's not.  The performance
issue only occurs with FTP.  If I'm wrong on this, I'd be happy to hear
about it, though.

> Have you asked your ISP about the issue?

The ISP is my client in this case.  He's asking _me_ about the issue ;)

> Can you reproduce by moving ftp to a different port #?  (Perhaps some 
> quality-of-service thingy is providing different bandwidth by port...)

I haven't tried that, but I forgot to mention that a Debian box located
right next to the problem box (on the same network) gets speeds equal
to what would be expected.  To me, that ruled out QOS, routing and other
beyond-my-control Internet problems.  Again, I'm happy to be corrected
if there's something here I'm not aware of.

> Does passive versus active FTP make a difference?

Doesn't seem to.  The speed seems to be about 14k/sec, regardless of
active/passive.

> Anything interesting in 'netstat -s'?

This is just a snippet ... but it looks like an awful lot of retransmits
and duplicates to me.  I hadn't looked at this before, is this indicative
of any particular problem?

tcp:
        309984 packets sent
                202559 data packets (238348935 bytes)
                1620 data packets (1751573 bytes) retransmitted
                1 resend initiated by MTU discovery
                71396 ack-only packets (1521 delayed)
                0 URG only packets
                49 window probe packets
                21125 window update packets
                13235 control packets
        286171 packets received
                135943 acks (for 236641220 bytes)
                8830 duplicate acks
                0 acks for unsent data
                142183 packets (164768204 bytes) received in-sequence
                215 completely duplicate packets (25524 bytes)
                1 old duplicate packet
                0 packets with some dup. data (0 bytes duped)
                1719 out-of-order packets (2151248 bytes)
                37 packets (37 bytes) of data after window
                37 window probes
                667 window update packets
                32 packets received after close
                5 discarded for bad checksums
                0 discarded for bad header offset fields
                0 discarded because packet too short
        4613 connection requests
        5160 connection accepts
        19 bad connection attempts
        0 listen queue overflows
        9748 connections established (including accepts)
        10039 connections closed (including 1656 drops)
                463 connections updated cached RTT on close
                463 connections updated cached RTT variance on close
                132 connections updated cached ssthresh on close
        22 embryonic connections dropped
        90592 segments updated rtt (of 76760 attempts)
        805 retransmit timeouts
                9 connections dropped by rexmit timeout
        54 persist timeouts
                0 connections dropped by persist timeout
        0 keepalive timeouts
                0 keepalive probes sent
                0 connections dropped by keepalive
        55073 correct ACK header predictions
        120750 correct data packet header predictions
        5210 syncache entries added
                61 retransmitted
                36 dupsyn
                0 dropped
                5160 completed
                0 bucket overflow
                0 cache overflow
                43 reset
                7 stale
                0 aborted
                0 badack
                0 unreach
                0 zone failures
        0 cookies sent
        0 cookies received

-- 
Bill Moran
Potential Technologies
http://www.potentialtech.com