Re: [OmniOS-discuss] CIFS ignores TCP Buffer Settings

[Mini Trader]

Bad

root@storage1:/root# dtrace -s tcp_tput.d
^C

  unacked(bytes)  10.250.0.3
   2049
   value  - Distribution - count
  32 | 0
  64 | 11
 128 | 0

  unacked(bytes)  10.250.0.2
   2049
   value  - Distribution - count
  -1 | 0
   0 | 63
   1 | 0
   2 | 0
   4 | 0
   8 | 0
  16 | 0
  32 | 0
  64 |@5
 128 |@@@  195
 256 |@@   54
 512 |@@   13
1024 |@5
2048 | 0

  unacked(bytes)  10.255.0.55
   445
   value  - Distribution - count
  -1 | 0
   0 | 7
   1 | 0
   2 | 0
   4 | 0
   8 | 0
  16 | 0
  32 | 10
  64 | 48
 128 | 9
 256 | 0
 512 | 0
1024 | 1
2048 | 0
4096 | 0
8192 | 0
   16384 | 2
   32768 | 46626
   65536 | 6
  131072 | 0

  SWND(bytes) 10.255.0.55
22
   value  - Distribution - count
   16384 | 0
   32768 | 1
   65536 | 0

  SWND(bytes) 10.250.0.3
   2049
   value  - Distribution - count
   32768 | 0
   65536 | 17
  131072 | 0

  SWND(bytes) 10.250.0.2
   2049
   value  - Distribution - count
  131072 | 0
  262144 | 369
  524288 | 0

  SWND(bytes) 10.255.0.55
   445
   value  - Distribution - count
   16384 | 0
   32768 | 367
   65536 | 49052
  131072 | 0

Good

root@storage1:/root# svcadm restart smb/server
root@storage1:/root# dtrace -s tcp_tput.d
^C

  unacked(bytes)  10.255.0.55
22
   value  - Distribution - count
  32 | 0
  64 | 1
 128 | 0

  unacked(bytes)  10.250.0.3
   2049
   value  - Distribution - count
  32 | 0
  64 

Re: [OmniOS-discuss] CIFS ignores TCP Buffer Settings

[Mini Trader]

Running the following dtrace.

#!/usr/sbin/dtrace -s

#pragma D option quiet

tcp:::send
/ (args[4]->tcp_flags & (TH_SYN|TH_RST|TH_FIN)) == 0 /
{
@unacked["unacked(bytes)", args[2]->ip_daddr, args[4]->tcp_sport] =
quantize(args[3]->tcps_snxt - args[3]->tcps_suna);
}

tcp:::receive
/ (args[4]->tcp_flags & (TH_SYN|TH_RST|TH_FIN)) == 0 /
{
@swnd["SWND(bytes)", args[2]->ip_saddr, args[4]->tcp_dport] =
quantize((args[4]->tcp_window)*(1 << args[3]->tcps_snd_ws));

}


Is showing that the windows sizes are not going above 64k when things are
not working properly.

On Tue, Mar 8, 2016 at 7:56 PM, Mini Trader 
wrote:

> If it helps.  This doesn't happen on NFS from the exact same client.  How
> do I file a bug?
>
> On Tue, Mar 8, 2016 at 1:51 PM, Mini Trader 
> wrote:
>
>> Simple example.
>>
>> 1 Server 1 client.
>>
>> Restart service everything is fast.  A few hours later from same client
>> (nothing happening concurrently) speed is slow.  Restart service again,
>> speed is fast.
>>
>> Its like CIFS starts off fast than somehow for whatever reason if it is
>> not used, the connection for my CIFS drives to the server becomes slow.
>> Also this only happens when the client is downloading.  Not when uploading
>> to the server that is always fast.
>>
>> On Tue, Mar 8, 2016 at 1:42 AM, Jim Klimov  wrote:
>>
>>> 8 марта 2016 г. 6:42:13 CET, Mini Trader 
>>> пишет:
>>> >Is it possible that CIFS will ignore TCP buffer settings after a while?
>>> >
>>> >I've confirmed my systems max transfer rate using iperf and have tuned
>>> >my
>>> >buffers accordingly. For whatever reason CIFS seems to forget these
>>> >settings after a while as speed drops significantly. Issuing a restart
>>> >of
>>> >the service immediately appears to restore the setting as transfer
>>> >speed
>>> >becomes normal again.
>>> >
>>> >Any ideas why this would happen?
>>> >
>>> >
>>> >
>>> >
>>> >___
>>> >OmniOS-discuss mailing list
>>> >OmniOS-discuss@lists.omniti.com
>>> >http://lists.omniti.com/mailman/listinfo/omnios-discuss
>>>
>>> As a random guess from experience with other network stuff - does the
>>> speed-drop happen on a running connection or new ones too? Do you have
>>> concurrent transfers at this time?
>>>
>>> Some other subsystems (no idea if this one too) use best speeds for new
>>> or recently awakened dormant connections, so short-lived bursts are fast -
>>> at expence of long-running active bulk transfers (deemed to be bulk because
>>> they run for a long time).
>>>
>>> HTH, Jim
>>> --
>>> Typos courtesy of K-9 Mail on my Samsung Android
>>>
>>
>>
>
___
OmniOS-discuss mailing list
OmniOS-discuss@lists.omniti.com
http://lists.omniti.com/mailman/listinfo/omnios-discuss

Re: [OmniOS-discuss] CIFS ignores TCP Buffer Settings

[Mini Trader]

If it helps.  This doesn't happen on NFS from the exact same client.  How
do I file a bug?

On Tue, Mar 8, 2016 at 1:51 PM, Mini Trader 
wrote:

> Simple example.
>
> 1 Server 1 client.
>
> Restart service everything is fast.  A few hours later from same client
> (nothing happening concurrently) speed is slow.  Restart service again,
> speed is fast.
>
> Its like CIFS starts off fast than somehow for whatever reason if it is
> not used, the connection for my CIFS drives to the server becomes slow.
> Also this only happens when the client is downloading.  Not when uploading
> to the server that is always fast.
>
> On Tue, Mar 8, 2016 at 1:42 AM, Jim Klimov  wrote:
>
>> 8 марта 2016 г. 6:42:13 CET, Mini Trader 
>> пишет:
>> >Is it possible that CIFS will ignore TCP buffer settings after a while?
>> >
>> >I've confirmed my systems max transfer rate using iperf and have tuned
>> >my
>> >buffers accordingly. For whatever reason CIFS seems to forget these
>> >settings after a while as speed drops significantly. Issuing a restart
>> >of
>> >the service immediately appears to restore the setting as transfer
>> >speed
>> >becomes normal again.
>> >
>> >Any ideas why this would happen?
>> >
>> >
>> >
>> >
>> >___
>> >OmniOS-discuss mailing list
>> >OmniOS-discuss@lists.omniti.com
>> >http://lists.omniti.com/mailman/listinfo/omnios-discuss
>>
>> As a random guess from experience with other network stuff - does the
>> speed-drop happen on a running connection or new ones too? Do you have
>> concurrent transfers at this time?
>>
>> Some other subsystems (no idea if this one too) use best speeds for new
>> or recently awakened dormant connections, so short-lived bursts are fast -
>> at expence of long-running active bulk transfers (deemed to be bulk because
>> they run for a long time).
>>
>> HTH, Jim
>> --
>> Typos courtesy of K-9 Mail on my Samsung Android
>>
>
>
___
OmniOS-discuss mailing list
OmniOS-discuss@lists.omniti.com
http://lists.omniti.com/mailman/listinfo/omnios-discuss

Re: [OmniOS-discuss] CIFS ignores TCP Buffer Settings

[Mini Trader]

Simple example.

1 Server 1 client.

Restart service everything is fast.  A few hours later from same client
(nothing happening concurrently) speed is slow.  Restart service again,
speed is fast.

Its like CIFS starts off fast than somehow for whatever reason if it is not
used, the connection for my CIFS drives to the server becomes slow.  Also
this only happens when the client is downloading.  Not when uploading to
the server that is always fast.

On Tue, Mar 8, 2016 at 1:42 AM, Jim Klimov  wrote:

> 8 марта 2016 г. 6:42:13 CET, Mini Trader  пишет:
> >Is it possible that CIFS will ignore TCP buffer settings after a while?
> >
> >I've confirmed my systems max transfer rate using iperf and have tuned
> >my
> >buffers accordingly. For whatever reason CIFS seems to forget these
> >settings after a while as speed drops significantly. Issuing a restart
> >of
> >the service immediately appears to restore the setting as transfer
> >speed
> >becomes normal again.
> >
> >Any ideas why this would happen?
> >
> >
> >
> >
> >___
> >OmniOS-discuss mailing list
> >OmniOS-discuss@lists.omniti.com
> >http://lists.omniti.com/mailman/listinfo/omnios-discuss
>
> As a random guess from experience with other network stuff - does the
> speed-drop happen on a running connection or new ones too? Do you have
> concurrent transfers at this time?
>
> Some other subsystems (no idea if this one too) use best speeds for new or
> recently awakened dormant connections, so short-lived bursts are fast - at
> expence of long-running active bulk transfers (deemed to be bulk because
> they run for a long time).
>
> HTH, Jim
> --
> Typos courtesy of K-9 Mail on my Samsung Android
>
___
OmniOS-discuss mailing list
OmniOS-discuss@lists.omniti.com
http://lists.omniti.com/mailman/listinfo/omnios-discuss

Re: [OmniOS-discuss] CIFS ignores TCP Buffer Settings

[Jim Klimov]

8 марта 2016 г. 6:42:13 CET, Mini Trader  пишет:
>Is it possible that CIFS will ignore TCP buffer settings after a while?
>
>I've confirmed my systems max transfer rate using iperf and have tuned
>my
>buffers accordingly. For whatever reason CIFS seems to forget these
>settings after a while as speed drops significantly. Issuing a restart
>of
>the service immediately appears to restore the setting as transfer
>speed
>becomes normal again.
>
>Any ideas why this would happen?
>
>
>
>
>___
>OmniOS-discuss mailing list
>OmniOS-discuss@lists.omniti.com
>http://lists.omniti.com/mailman/listinfo/omnios-discuss

As a random guess from experience with other network stuff - does the 
speed-drop happen on a running connection or new ones too? Do you have 
concurrent transfers at this time?

Some other subsystems (no idea if this one too) use best speeds for new or 
recently awakened dormant connections, so short-lived bursts are fast - at 
expence of long-running active bulk transfers (deemed to be bulk because they 
run for a long time).

HTH, Jim
--
Typos courtesy of K-9 Mail on my Samsung Android
___
OmniOS-discuss mailing list
OmniOS-discuss@lists.omniti.com
http://lists.omniti.com/mailman/listinfo/omnios-discuss

[OmniOS-discuss] CIFS ignores TCP Buffer Settings

[Mini Trader]

Is it possible that CIFS will ignore TCP buffer settings after a while?

I've confirmed my systems max transfer rate using iperf and have tuned my
buffers accordingly. For whatever reason CIFS seems to forget these
settings after a while as speed drops significantly. Issuing a restart of
the service immediately appears to restore the setting as transfer speed
becomes normal again.

Any ideas why this would happen?
___
OmniOS-discuss mailing list
OmniOS-discuss@lists.omniti.com
http://lists.omniti.com/mailman/listinfo/omnios-discuss