How to tune network I/O subsystem ?
kanthi
ASE 12.5.0.3 EBF 11442.
My sysmon shows the following and recommends to tune network I/O subsystem,
but doesn't give me a clue on how :-).
Here it is :-
Task Context Switches Due To:
Voluntary Yields 1.4 0.7 426
1.7 %
Cache Search Misses 18.1 8.5 5432
21.4 %
System Disk Writes 0.0 0.0 3
0.0 %
I/O Pacing 0.0 0.0 14
0.1 %
Logical Lock Contention 0.0 0.0 0
0.0 %
Address Lock Contention 0.0 0.0 0
0.0 %
Latch Contention 0.0 0.0 0
0.0 %
Log Semaphore Contention 0.0 0.0 0
0.0 %
PLC Lock Contention 0.0 0.0 0
0.0 %
Group Commit Sleeps 0.0 0.0 6
0.0 %
Last Log Page Writes 0.2 0.1 49
0.2 %
Modify Conflicts 0.1 0.0 15
0.1 %
I/O Device Contention 0.0 0.0 0
0.0 %
Network Packet Received 8.3 3.9 2492
9.8 %
Network Packet Sent 804.8 376.7 241449
953.4 %
Other
s -748.5 -350.3 -224561 -887 %
Tuning Recommendations for Task Management
------------------------------------------
- Consider tuning your Network I/O sub-system.
Network I/O Management
----------------------
Total Network I/O Requests 21.8 10.2 6554
n/a
Network I/Os Delayed 0.0 0.0 0
0.0 %
Total TDS Packets Received per sec per xact count % of
total
------------------------- ------------ ------------ ---------- -------
---
Engine 0 1.5 0.7 445
17.9 %
Engine 1 2.7 1.3 811
32.5 %
Engine 2 4.1 1.9 1236
49.6 %
------------------------- ------------ ------------ ----------
Total TDS Packets Rec'd 8.3 3.9 2492
Total Bytes Received per sec per xact count % of
total
------------------------- ------------ ------------ ---------- -------
---
Engine 0 133.9 62.7 40169
14.4 %
Engine 1 177.5 83.1 53258
19.1 %
Engine 2 618.9 289.7 185671
66.5 %
------------------------- ------------ ------------ ----------
Total Bytes Rec'd 930.3 435.4 279098
Avg Bytes Rec'd per Packet n/a n/a 111
n/a
--------------------------------------------------------------------------
---
Total TDS Packets Sent per sec per xact count % of
total
------------------------- ------------ ------------ ---------- -------
---
Engine 0 2.0 0.9 607
0.3 %
Engine 1 3.0 1.4 895 0.4
%
Engine 2 799.8 374.3 239947
99.4 %
------------------------- ------------ ------------ ----------
Total TDS Packets Sent 804.8 376.7 241449
Total Bytes Sent per sec per xact count % of
total
------------------------- ------------ ------------ ---------- -------
---
Engine 0 402.0 188.1 120598
0.1 %
Engine 1 338.9 158.6 101678
0.1 %
Engine 2 401813.8 188056.4 120544149
99.8 %
------------------------- ------------ ------------ ----------
Total Bytes Sent 402554.8 188403.2 120766425
Avg Bytes Sent per Packet n/a n/a 500
n/a
=============================== End of Report
=================================
(return status = 0)
Here is the configuration settings on network related parameters.
1> sp_configure "network"
2> go
Msg 17411, Level 16, State 1:
Server 's01mimas', Procedure 'sp_configure', Line 195:
Configuration option is not unique.
Parameter Name Default Memory Used Config Value
Run Value Unit Type
------------------------------ ----------- ----------- ------------
----------- -------------------- ----------
additional network memory 0 0 0
0 bytes dynamic
default network packet size 512 #453 512
512 bytes static
max network packet size 512 0 512
512 bytes static
max number network listeners 5 961 5
5 number static
(1 row affected)
(return status = 1)
Will tuning on network packet size help in reducing the "network packets
sent" in sysmon ?
Thanks & regards,
Kanthi
Sherlock, Kevin
--
Kevin Sherlock
Staff Info Systems Analyst
Omaha, NE
DexMedia/Amdocs
http://dexonline.com
[TeamSybase]
http://teamsybase.com/kevin.sherlock
Take a look at CodeXchange:
http://www.codexchange.sybase.com
ISUG Enhancement Requests
http://www.isug.com/cgi-bin/ISUG2/submit_enhancement
Mark A. Parsons
>
> Hi Gurus,
Well, while we're waiting around for one of them to show up I'll throw in
my $0.02 worth ...
> My sysmon shows the following and recommends to tune network I/O subsystem,
> but doesn't give me a clue on how :-).
...snip...
1 - to adequately diagnose the issue of tuning the network i/o subsystem
we'll really need to see more of the sp_sysmon output; tuning/modifying the
packet size may help but so could adjusting the i/o polling count (in case
ASE isn't processing i/o's fast enough), runnable search count (if ASE is
being switched out by the OS too often) and/or possibly modifying a
cache(s) (if the queries driving all this network traffic can't find much
of the desired data in cache) ... can't really determine these issues
without the full set of sp_sysmon output
2 - from a network *only* perspective it would appear that all of the
network i/o during this sp_sysmon session was due to packets *sent* by
engine 2; engine 2 certainly maxed out it's outbound packets (average size
of ~500 bytes/packet) so you should probably consider upping the max
network packet size to 1024 to start with (you may need/want to go higher
based on further monitoring as well as the nature of the query(s) that are
generating all these i/o's); you'll need to look at adjusting
max network packet size
additional network memory
The SAG and P&T (Monitoring) guides have some write-up's on configuring
these settings.
NOTE: Once you've re-configured ASE you'll also have to modify your client
application (eg, isql) to utilize the larger packet size.
3 - how many users/connections were running in the server during this time
frame? the reason I ask is that engines 0 and 1 aren't doing anything
relative to engine 2; it would appear that sp_sysmon was run during one
particular set of queries running on just one engine
The point here is that we could tune your server for the particular query
that's generating all the outbound network traffic ... but will this help
any of your other ASE activity? Only way to tell is to go through multiple
sp_sysmon sessions when the system load is more (?) representable of a
'normal' load.
--
Mark A. Parsons
Iron Horse, Inc.
iron_...@NOSPAM.compuserve.com
marka.parsons
> max network packet size to 1024 to start with
You're welcome to try higher packet sizes (eg, 4096) but
you'll still want to run some tests to see what's best for
you. In the case of 'larger' packet sizes (> 1024) you may
want to check with your network administrator to see what
your servers/gateways/hubs/network-gear have as their MTU
settings ... a 'low' MTU (eg, 1500) setting can offset some
of the benefits of a larger (eg, > 1500) ASE-network packet
size.
Jesus M. Salvo Jr.
In my experiences, these "network" issues with ASE is only a symptom, not the real problem.
The real problem may actually be a query that is returning a lot of rows, or taking a long time to finish, etc...