My waitstats reveal a high percentage of networkIO. Currently my adapter is
a 100Mbps. Do I need a 1Gbps adapter ? How can I find out why theres a high
wait of networkIO and what can I do to resolve it ?
ThanksNot to say that you are not maxed out on your NIC's throughput, but measure
it with perf mon.
Typically, the Network IO wait states are due to the client's ability to
process the packets, especially when there is a VERY large TDS stream.
Take a look at the clients' CPU usage. If it is high, they are not
processing the results quick enough.
Sincerely,
Anthony Thomas
"Hassan" <fatima_ja@.hotmail.com> wrote in message
news:uvvi1Aj6EHA.3944@.TK2MSFTNGP12.phx.gbl...
My waitstats reveal a high percentage of networkIO. Currently my adapter is
a 100Mbps. Do I need a 1Gbps adapter ? How can I find out why theres a high
wait of networkIO and what can I do to resolve it ?
Thanks|||I agree with Anthony on this. The client plays a large role is how
efficiently the packets get processed. Especially with heavy replication. I
have seen very high network waits on systems where the subscribers or the
distributor can not keep up with the requests.
--
Andrew J. Kelly SQL MVP
"AnthonyThomas" <Anthony.Thomas@.CommerceBank.com> wrote in message
news:eSVDgdl6EHA.3708@.TK2MSFTNGP14.phx.gbl...
> Not to say that you are not maxed out on your NIC's throughput, but
> measure
> it with perf mon.
> Typically, the Network IO wait states are due to the client's ability to
> process the packets, especially when there is a VERY large TDS stream.
> Take a look at the clients' CPU usage. If it is high, they are not
> processing the results quick enough.
> Sincerely,
>
> Anthony Thomas
>
> --
> "Hassan" <fatima_ja@.hotmail.com> wrote in message
> news:uvvi1Aj6EHA.3944@.TK2MSFTNGP12.phx.gbl...
> My waitstats reveal a high percentage of networkIO. Currently my adapter
> is
> a 100Mbps. Do I need a 1Gbps adapter ? How can I find out why theres a
> high
> wait of networkIO and what can I do to resolve it ?
> Thanks
>|||Hi
As an aside from what the others have said, if you wanted to improve the
server's network IO then you may be better off adding a second 100MB card in
the server first before investing in a more expensive option..
John
"Hassan" <fatima_ja@.hotmail.com> wrote in message
news:uvvi1Aj6EHA.3944@.TK2MSFTNGP12.phx.gbl...
> My waitstats reveal a high percentage of networkIO. Currently my adapter
> is
> a 100Mbps. Do I need a 1Gbps adapter ? How can I find out why theres a
> high
> wait of networkIO and what can I do to resolve it ?
> Thanks
>|||So what can we do in this case ? Is it something that we can control from
the SQL Server end or the client ? So are you saying that even if we had a
1Gbps NIC card, the problem still remains. Yes this server is a subscriber
and is receiving replicated transactions
"Andrew J. Kelly" <sqlmvpnooospam@.shadhawk.com> wrote in message
news:uwHRbap6EHA.2016@.TK2MSFTNGP15.phx.gbl...
> I agree with Anthony on this. The client plays a large role is how
> efficiently the packets get processed. Especially with heavy replication.
I
> have seen very high network waits on systems where the subscribers or the
> distributor can not keep up with the requests.
> --
> Andrew J. Kelly SQL MVP
>
> "AnthonyThomas" <Anthony.Thomas@.CommerceBank.com> wrote in message
> news:eSVDgdl6EHA.3708@.TK2MSFTNGP14.phx.gbl...
> > Not to say that you are not maxed out on your NIC's throughput, but
> > measure
> > it with perf mon.
> >
> > Typically, the Network IO wait states are due to the client's ability to
> > process the packets, especially when there is a VERY large TDS stream.
> >
> > Take a look at the clients' CPU usage. If it is high, they are not
> > processing the results quick enough.
> >
> > Sincerely,
> >
> >
> > Anthony Thomas
> >
> >
> > --
> >
> > "Hassan" <fatima_ja@.hotmail.com> wrote in message
> > news:uvvi1Aj6EHA.3944@.TK2MSFTNGP12.phx.gbl...
> > My waitstats reveal a high percentage of networkIO. Currently my adapter
> > is
> > a 100Mbps. Do I need a 1Gbps adapter ? How can I find out why theres a
> > high
> > wait of networkIO and what can I do to resolve it ?
> >
> > Thanks
> >
> >
>|||So having 2 100Mbps card ... Do we need to do anything special as far as
configuring them ? IF we can afford the 1Gbps card, should we go with that ?
Would that help ?
Thanks
"John Bell" <jbellnewsposts@.hotmail.com> wrote in message
news:egGkB8p6EHA.844@.TK2MSFTNGP14.phx.gbl...
> Hi
> As an aside from what the others have said, if you wanted to improve the
> server's network IO then you may be better off adding a second 100MB card
in
> the server first before investing in a more expensive option..
> John
> "Hassan" <fatima_ja@.hotmail.com> wrote in message
> news:uvvi1Aj6EHA.3944@.TK2MSFTNGP12.phx.gbl...
> > My waitstats reveal a high percentage of networkIO. Currently my adapter
> > is
> > a 100Mbps. Do I need a 1Gbps adapter ? How can I find out why theres a
> > high
> > wait of networkIO and what can I do to resolve it ?
> >
> > Thanks
> >
> >
>|||Hi
Having a 1GB card is not giving you anything extra if you don't connect it
to a 1GB port on the hub/switch
I think these should help setting up multiple nics.
http://support.microsoft.com/default.aspx?scid=kb;en-us;197991
http://support.microsoft.com/default.aspx?scid=kb;en-us;131736
John
"Hassan" <fatima_ja@.hotmail.com> wrote in message
news:%23yhMz0r6EHA.2552@.TK2MSFTNGP09.phx.gbl...
> So having 2 100Mbps card ... Do we need to do anything special as far as
> configuring them ? IF we can afford the 1Gbps card, should we go with that
> ?
> Would that help ?
> Thanks
> "John Bell" <jbellnewsposts@.hotmail.com> wrote in message
> news:egGkB8p6EHA.844@.TK2MSFTNGP14.phx.gbl...
>> Hi
>> As an aside from what the others have said, if you wanted to improve the
>> server's network IO then you may be better off adding a second 100MB card
> in
>> the server first before investing in a more expensive option..
>> John
>> "Hassan" <fatima_ja@.hotmail.com> wrote in message
>> news:uvvi1Aj6EHA.3944@.TK2MSFTNGP12.phx.gbl...
>> > My waitstats reveal a high percentage of networkIO. Currently my
>> > adapter
>> > is
>> > a 100Mbps. Do I need a 1Gbps adapter ? How can I find out why theres a
>> > high
>> > wait of networkIO and what can I do to resolve it ?
>> >
>> > Thanks
>> >
>> >
>>
>|||If the bottleneck is that the clients or subscribers can not process the
packets fast enough it won't do any good to try and send them even faster.
The ability to process the requests fromt he server are not just network
card related. What ever process is reading those packets and subsequently
doing something with that data needs to be optimized to handle the load. It
may be something as simple as the disk drives on the clients can not handle
the load and that is backing everything up from there. But you need to
investigate each machine to see where the bottleneck is before you can
address it. Simply throwing hardware at a problem is not good unless you
know what the actual bottleneck is so you can throw the right hardware at
it. Or better yet tune the process so it is more efficient. If you are
doing a lot of replication you need to ensure your log files are on a fast
drive array that is not encumbered by other processes or files and that the
disk queues are almost non-existent. Once you are sure that is not an issue
you can work your way up to the other drives, cpu etc to ensure you don't
have any bottlenecks. It is certainly possible that the network has issues
and can't handle the load but it is usually the case that the clients are
the source of the problems and not the server in cases like this.
--
Andrew J. Kelly SQL MVP
"Hassan" <fatima_ja@.hotmail.com> wrote in message
news:uPWcO0r6EHA.2192@.TK2MSFTNGP14.phx.gbl...
> So what can we do in this case ? Is it something that we can control from
> the SQL Server end or the client ? So are you saying that even if we had a
> 1Gbps NIC card, the problem still remains. Yes this server is a subscriber
> and is receiving replicated transactions
>
> "Andrew J. Kelly" <sqlmvpnooospam@.shadhawk.com> wrote in message
> news:uwHRbap6EHA.2016@.TK2MSFTNGP15.phx.gbl...
>> I agree with Anthony on this. The client plays a large role is how
>> efficiently the packets get processed. Especially with heavy
>> replication.
> I
>> have seen very high network waits on systems where the subscribers or the
>> distributor can not keep up with the requests.
>> --
>> Andrew J. Kelly SQL MVP
>>
>> "AnthonyThomas" <Anthony.Thomas@.CommerceBank.com> wrote in message
>> news:eSVDgdl6EHA.3708@.TK2MSFTNGP14.phx.gbl...
>> > Not to say that you are not maxed out on your NIC's throughput, but
>> > measure
>> > it with perf mon.
>> >
>> > Typically, the Network IO wait states are due to the client's ability
>> > to
>> > process the packets, especially when there is a VERY large TDS stream.
>> >
>> > Take a look at the clients' CPU usage. If it is high, they are not
>> > processing the results quick enough.
>> >
>> > Sincerely,
>> >
>> >
>> > Anthony Thomas
>> >
>> >
>> > --
>> >
>> > "Hassan" <fatima_ja@.hotmail.com> wrote in message
>> > news:uvvi1Aj6EHA.3944@.TK2MSFTNGP12.phx.gbl...
>> > My waitstats reveal a high percentage of networkIO. Currently my
>> > adapter
>> > is
>> > a 100Mbps. Do I need a 1Gbps adapter ? How can I find out why theres a
>> > high
>> > wait of networkIO and what can I do to resolve it ?
>> >
>> > Thanks
>> >
>> >
>>
>|||"Hassan" <fatima_ja@.hotmail.com> wrote in message
news:%23yhMz0r6EHA.2552@.TK2MSFTNGP09.phx.gbl...
> So having 2 100Mbps card ... Do we need to do anything special as far as
> configuring them ? IF we can afford the 1Gbps card, should we go with that
?
> Would that help ?
Well, w/o 1 gig cards at the other ends, it may not help much.
And to ask the obvious question, this is 100Mbs switched network right? No
hubs?|||We got a couple of independent threads going on here; so, it looked like a
better idea to start another than to get buried in either.
First question is one of architecture. Second is of capacity.
Is this installation stricly Client/Server (2-tier) or are most of the
requests being fed through an application or middle-ware server (n-tier)?
If this is most likely a client [that is a workstation(s)], then looking at
upgrading all of the client NICs, memory, cpu, and/or OS might be cost
prohibitive. A look at an n-tier design might yield better performance
increase due to distribution efficiencies.
If this is already an n-tier architecture, then it is probably a cpu and/or
memory issue on the application/web/middle-ware server, or all three,
perhaps. Check out the Bytes and Packets Sent/second statistics in
Performance Monitor on all hosts in the system. The one with the lowest
throughput in the chain is probably your bottleneck.
Sincerely,
Anthony Thomas
"Hassan" <fatima_ja@.hotmail.com> wrote in message
news:uvvi1Aj6EHA.3944@.TK2MSFTNGP12.phx.gbl...
My waitstats reveal a high percentage of networkIO. Currently my adapter is
a 100Mbps. Do I need a 1Gbps adapter ? How can I find out why theres a high
wait of networkIO and what can I do to resolve it ?
Thanks|||Its a combination of middle tier + replication i.e. this SQL Server is a
subscriber that receives quite a bit of transactions
You mentioned Bytes and Packets Sent/sec from all the client ... But is it
the problem with the clients processing the results fast enough .. which
means do i need to look at Bytes and Packets received/sec at the client ?
Also another question is will the duration in Profiler for the completion of
say a sproc being executed from the client also include the networkio waits
in them .. I mean say the networkio wait contributes 3 secs .. will this 3
secs be accounted for in the duration for the sproc completion or will it be
excluded. ?
Thanks
"AnthonyThomas" <Anthony.Thomas@.CommerceBank.com> wrote in message
news:eC6qD%2386EHA.2180@.TK2MSFTNGP12.phx.gbl...
> We got a couple of independent threads going on here; so, it looked like a
> better idea to start another than to get buried in either.
> First question is one of architecture. Second is of capacity.
> Is this installation stricly Client/Server (2-tier) or are most of the
> requests being fed through an application or middle-ware server (n-tier)?
> If this is most likely a client [that is a workstation(s)], then looking
at
> upgrading all of the client NICs, memory, cpu, and/or OS might be cost
> prohibitive. A look at an n-tier design might yield better performance
> increase due to distribution efficiencies.
> If this is already an n-tier architecture, then it is probably a cpu
and/or
> memory issue on the application/web/middle-ware server, or all three,
> perhaps. Check out the Bytes and Packets Sent/second statistics in
> Performance Monitor on all hosts in the system. The one with the lowest
> throughput in the chain is probably your bottleneck.
> Sincerely,
>
> Anthony Thomas
>
> --
> "Hassan" <fatima_ja@.hotmail.com> wrote in message
> news:uvvi1Aj6EHA.3944@.TK2MSFTNGP12.phx.gbl...
> My waitstats reveal a high percentage of networkIO. Currently my adapter
is
> a 100Mbps. Do I need a 1Gbps adapter ? How can I find out why theres a
high
> wait of networkIO and what can I do to resolve it ?
> Thanks
>|||Hi
You will need to use perfmon on the client(s) to see what workload they are
processing. I can't see how profiler could include the network time as can
only be accounted for on the client.
One thing no-one seems to have suggested that poor SQL that returns too much
data may also be having an impact.
If you are replicating, then if possible try and separate the servers onto
their own dedicated segment.
John
"Hassan" <fatima_ja@.hotmail.com> wrote in message
news:OLBRFr96EHA.2012@.TK2MSFTNGP15.phx.gbl...
> Its a combination of middle tier + replication i.e. this SQL Server is a
> subscriber that receives quite a bit of transactions
> You mentioned Bytes and Packets Sent/sec from all the client ... But is it
> the problem with the clients processing the results fast enough .. which
> means do i need to look at Bytes and Packets received/sec at the client ?
> Also another question is will the duration in Profiler for the completion
> of
> say a sproc being executed from the client also include the networkio
> waits
> in them .. I mean say the networkio wait contributes 3 secs .. will this 3
> secs be accounted for in the duration for the sproc completion or will it
> be
> excluded. ?
> Thanks
>
> "AnthonyThomas" <Anthony.Thomas@.CommerceBank.com> wrote in message
> news:eC6qD%2386EHA.2180@.TK2MSFTNGP12.phx.gbl...
>> We got a couple of independent threads going on here; so, it looked like
>> a
>> better idea to start another than to get buried in either.
>> First question is one of architecture. Second is of capacity.
>> Is this installation stricly Client/Server (2-tier) or are most of the
>> requests being fed through an application or middle-ware server (n-tier)?
>> If this is most likely a client [that is a workstation(s)], then looking
> at
>> upgrading all of the client NICs, memory, cpu, and/or OS might be cost
>> prohibitive. A look at an n-tier design might yield better performance
>> increase due to distribution efficiencies.
>> If this is already an n-tier architecture, then it is probably a cpu
> and/or
>> memory issue on the application/web/middle-ware server, or all three,
>> perhaps. Check out the Bytes and Packets Sent/second statistics in
>> Performance Monitor on all hosts in the system. The one with the lowest
>> throughput in the chain is probably your bottleneck.
>> Sincerely,
>>
>> Anthony Thomas
>>
>> --
>> "Hassan" <fatima_ja@.hotmail.com> wrote in message
>> news:uvvi1Aj6EHA.3944@.TK2MSFTNGP12.phx.gbl...
>> My waitstats reveal a high percentage of networkIO. Currently my adapter
> is
>> a 100Mbps. Do I need a 1Gbps adapter ? How can I find out why theres a
> high
>> wait of networkIO and what can I do to resolve it ?
>> Thanks
>>
>|||Hassan wrote:
> Its a combination of middle tier + replication i.e. this SQL Server
> is a subscriber that receives quite a bit of transactions
> You mentioned Bytes and Packets Sent/sec from all the client ... But
> is it the problem with the clients processing the results fast enough
> .. which means do i need to look at Bytes and Packets received/sec at
> the client ?
> Also another question is will the duration in Profiler for the
> completion of say a sproc being executed from the client also include
> the networkio waits in them .. I mean say the networkio wait
> contributes 3 secs .. will this 3 secs be accounted for in the
> duration for the sproc completion or will it be excluded. ?
> Thanks
>
> "AnthonyThomas" <Anthony.Thomas@.CommerceBank.com> wrote in message
> news:eC6qD%2386EHA.2180@.TK2MSFTNGP12.phx.gbl...
>> We got a couple of independent threads going on here; so, it looked
>> like a better idea to start another than to get buried in either.
>> First question is one of architecture. Second is of capacity.
>> Is this installation stricly Client/Server (2-tier) or are most of
>> the requests being fed through an application or middle-ware server
>> (n-tier)?
>> If this is most likely a client [that is a workstation(s)], then
>> looking at upgrading all of the client NICs, memory, cpu, and/or OS
>> might be cost prohibitive. A look at an n-tier design might yield
>> better performance increase due to distribution efficiencies.
>> If this is already an n-tier architecture, then it is probably a cpu
>> and/or memory issue on the application/web/middle-ware server, or
>> all three, perhaps. Check out the Bytes and Packets Sent/second
>> statistics in Performance Monitor on all hosts in the system. The
>> one with the lowest throughput in the chain is probably your
>> bottleneck.
>> Sincerely,
>>
>> Anthony Thomas
>>
>> --
>> "Hassan" <fatima_ja@.hotmail.com> wrote in message
>> news:uvvi1Aj6EHA.3944@.TK2MSFTNGP12.phx.gbl...
>> My waitstats reveal a high percentage of networkIO. Currently my
>> adapter is a 100Mbps. Do I need a 1Gbps adapter ? How can I find out
>> why theres a high wait of networkIO and what can I do to resolve it ?
>> Thanks
Profiler generally takes into account the time the client takes to fetch
all the data from the result set. If a result set is fetched slowly, you
will see high duration times in Profiler. You can see this easily if you
open a large table in SQL EM (All Rows Option). Run Profiler just prior
and watch the RPC:Starting/Completed and SQL:StmtStarting/StmtCompleted
events. When you issue the select, you'll see the starting event but no
complete. If you then CTRL+END to move to the end of the result set,
you'll eventually get the completed event.
Using Profiler in this way can give you an idea of the procedures that
are potentially leaving locks on the server because the results are
fetched slowly. This is especially true if the CPU is low.
David Gugick
Imceda Software
www.imceda.com|||Which spids in sysprocesses are showing the NETWORK IO wait states? Is it
the replication agents?
Think about it, in a replication environment, who is the client? That's
right, the susbscriper. So, in this case, it could very well be that the
SQL Server itself is getting its NIC(s) congested if the replication is too
large and/or too frequent for the network architecture employed.
Add to this that SQL Server's main purpose is to answer queries. If it is
spending most of its time keeping up with the replication, then how is it
suppose to respond to users' requests?
You should consider having seperate NIC(s) for the connection to the
Distributer server and then other NIC(s) for end-user queries and dedicate
those activities to those NIC(s).
You will also need to monitor your CPU, Memory, and Disk I/O counters to
make sure you've got a high enough capacity box.
Sincerely,
Anthony Thomas
"Hassan" <fatima_ja@.hotmail.com> wrote in message
news:OLBRFr96EHA.2012@.TK2MSFTNGP15.phx.gbl...
Its a combination of middle tier + replication i.e. this SQL Server is a
subscriber that receives quite a bit of transactions
You mentioned Bytes and Packets Sent/sec from all the client ... But is it
the problem with the clients processing the results fast enough .. which
means do i need to look at Bytes and Packets received/sec at the client ?
Also another question is will the duration in Profiler for the completion of
say a sproc being executed from the client also include the networkio waits
in them .. I mean say the networkio wait contributes 3 secs .. will this 3
secs be accounted for in the duration for the sproc completion or will it be
excluded. ?
Thanks
"AnthonyThomas" <Anthony.Thomas@.CommerceBank.com> wrote in message
news:eC6qD%2386EHA.2180@.TK2MSFTNGP12.phx.gbl...
> We got a couple of independent threads going on here; so, it looked like a
> better idea to start another than to get buried in either.
> First question is one of architecture. Second is of capacity.
> Is this installation stricly Client/Server (2-tier) or are most of the
> requests being fed through an application or middle-ware server (n-tier)?
> If this is most likely a client [that is a workstation(s)], then looking
at
> upgrading all of the client NICs, memory, cpu, and/or OS might be cost
> prohibitive. A look at an n-tier design might yield better performance
> increase due to distribution efficiencies.
> If this is already an n-tier architecture, then it is probably a cpu
and/or
> memory issue on the application/web/middle-ware server, or all three,
> perhaps. Check out the Bytes and Packets Sent/second statistics in
> Performance Monitor on all hosts in the system. The one with the lowest
> throughput in the chain is probably your bottleneck.
> Sincerely,
>
> Anthony Thomas
>
> --
> "Hassan" <fatima_ja@.hotmail.com> wrote in message
> news:uvvi1Aj6EHA.3944@.TK2MSFTNGP12.phx.gbl...
> My waitstats reveal a high percentage of networkIO. Currently my adapter
is
> a 100Mbps. Do I need a 1Gbps adapter ? How can I find out why theres a
high
> wait of networkIO and what can I do to resolve it ?
> Thanks
>
No comments:
Post a Comment