Named Pipes vs TCP/IP

Why would one choose Named Pipes over TCP/IP for the client net lib? We have
noticed for some of our customers that changing to named pipes speeds up the
response times of our application. But it is my understanding that a
properly configured LAN should work fine with TCP/IP.
Thanks,
Bob Castleman
SuccessWare Software
It is my understanding that named pipes have slightly less overhead than
TCP/IP but there it is usually not enough to warrant using named pipes. TCP
is much more versatile and should be encouraged in general over named pipes.
One thing to note is that WIN2003 has a 25% higher throuput in TCP over
Win2000 and DNS lookups are up to 120% faster as well. If your clients are
looking for speed improvements and are not on Win2003 they should try it
out. Another thing to note is that usually when clients tell me one
protocol is faster than another I tend to find there is too much chatter
from the client to the server in the first place. The more packets sent
back and forth the more likely you will have performance issues under heavy
load.
Andrew J. Kelly SQL MVP
"Bob Castleman" <nomail@.here> wrote in message
news:OAh7NLOlEHA.324@.TK2MSFTNGP11.phx.gbl...
> Why would one choose Named Pipes over TCP/IP for the client net lib? We
have
> noticed for some of our customers that changing to named pipes speeds up
the
> response times of our application. But it is my understanding that a
> properly configured LAN should work fine with TCP/IP.
> Thanks,
> Bob Castleman
> SuccessWare Software
>
|||A specific example of what we are running into happened this morning. A
customer called complaining of speed problems. Everything was "fine" on
Friday and today it was slow. CPU load on the database server was around 2%.
Disks weren't thrasing, etc. I had him change a workstation from TCP/IP to
Named Pipes and the time it took to load our application went from 20
seconds to 2 seconds. He reported a similar effect on the other
workstations. Nothing I have been able to find about TCP/IP and Named Pipes
would explain this behavior. This seems like some problem in network
communication, especially since the server was basically sitting idle.
I personally don't like "quick fixes". We run into this with just enough
regularity that it would be good to know what's really happening so we can
address the underlying problem.
"Andrew J. Kelly" <sqlmvpnooospam@.shadhawk.com> wrote in message
news:ed0R4WOlEHA.2892@.tk2msftngp13.phx.gbl...
> It is my understanding that named pipes have slightly less overhead than
> TCP/IP but there it is usually not enough to warrant using named pipes.
> TCP
> is much more versatile and should be encouraged in general over named
> pipes.
> One thing to note is that WIN2003 has a 25% higher throuput in TCP over
> Win2000 and DNS lookups are up to 120% faster as well. If your clients
> are
> looking for speed improvements and are not on Win2003 they should try it
> out. Another thing to note is that usually when clients tell me one
> protocol is faster than another I tend to find there is too much chatter
> from the client to the server in the first place. The more packets sent
> back and forth the more likely you will have performance issues under
> heavy
> load.
> --
> Andrew J. Kelly SQL MVP
>
> "Bob Castleman" <nomail@.here> wrote in message
> news:OAh7NLOlEHA.324@.TK2MSFTNGP11.phx.gbl...
> have
> the
>
|||Simply some facts top start with: Named Pipes is a OSI level 7 (Application)
protocol, WinSock is OSI level 5 (Session), therefore in a TCP only based
network, Named Pipes actually goes over WinSock, so it does not make any
sense that TCP Winsock is slower in that case, which is also not the
experience, and this is why we use TCP to perform client server benchmarks
for SQL Server. When running on the same machine, Named Pipes is faster then
TCP, because it becomes a Local Pipe call (this is where the magic . (dot)
notation is coming from).
Also since Named Pipes are file system objects at kernel level the
scalability compared to WinSock is much less, which is why if you need to
handle large number of connections you always have to use TCP sockets
instead of named pipes. Which each OS release thresholds move, like Andrew
said in Windows Server 2003, TCP sockets and DNS performance increase
dramatically. For example in NT 4.0 you would run out of file handles when
you would have more then 100-150 named pipes connections, this is not longer
a problem with Windows 2000 and Windows Server 2004, but this is to indicate
that each protocol have there own characteristics and scalability
thresholds.
In general the default choice is and should be TCP sockets for all remote
connections, if you are running on the same server like batch process you
might consider named pipes using a local pipe (using the . (dot) notation).
GertD@.SQLDev.Net
Please reply only to the newsgroups.
This posting is provided "AS IS" with no warranties, and confers no rights.
You assume all risk for your use.
Copyright SQLDev.Net 1991-2004 All rights reserved.
"Andrew J. Kelly" <sqlmvpnooospam@.shadhawk.com> wrote in message
news:ed0R4WOlEHA.2892@.tk2msftngp13.phx.gbl...
> It is my understanding that named pipes have slightly less overhead than
> TCP/IP but there it is usually not enough to warrant using named pipes.
> TCP
> is much more versatile and should be encouraged in general over named
> pipes.
> One thing to note is that WIN2003 has a 25% higher throuput in TCP over
> Win2000 and DNS lookups are up to 120% faster as well. If your clients
> are
> looking for speed improvements and are not on Win2003 they should try it
> out. Another thing to note is that usually when clients tell me one
> protocol is faster than another I tend to find there is too much chatter
> from the client to the server in the first place. The more packets sent
> back and forth the more likely you will have performance issues under
> heavy
> load.
> --
> Andrew J. Kelly SQL MVP
>
> "Bob Castleman" <nomail@.here> wrote in message
> news:OAh7NLOlEHA.324@.TK2MSFTNGP11.phx.gbl...
> have
> the
>
|||Anti.virus program trying to AV check the named pipes stuff?
Tibor Karaszi, SQL Server MVP
http://www.karaszi.com/sqlserver/default.asp
http://www.solidqualitylearning.com/
"Bob Castleman" <nomail@.here> wrote in message news:uZ2a5kOlEHA.3392@.TK2MSFTNGP14.phx.gbl...
>A specific example of what we are running into happened this morning. A customer called complaining
>of speed problems. Everything was "fine" on Friday and today it was slow. CPU load on the database
>server was around 2%. Disks weren't thrasing, etc. I had him change a workstation from TCP/IP to
>Named Pipes and the time it took to load our application went from 20 seconds to 2 seconds. He
>reported a similar effect on the other workstations. Nothing I have been able to find about TCP/IP
>and Named Pipes would explain this behavior. This seems like some problem in network communication,
>especially since the server was basically sitting idle.
> I personally don't like "quick fixes". We run into this with just enough regularity that it would
> be good to know what's really happening so we can address the underlying problem.
> "Andrew J. Kelly" <sqlmvpnooospam@.shadhawk.com> wrote in message
> news:ed0R4WOlEHA.2892@.tk2msftngp13.phx.gbl...
>
|||An other option we see often are a DNS name resolution problems, use tracert
to determine the route taken.
An other thought are these notebooks, that connect wireless as well, or at
home? In which case it most of the times helps to dump the DNS resolver
cache of the workstation. See IPCONFIG /flushdns
GertD@.SQLDev.Net
Please reply only to the newsgroups.
This posting is provided "AS IS" with no warranties, and confers no rights.
You assume all risk for your use.
Copyright SQLDev.Net 1991-2004 All rights reserved.
"Tibor Karaszi" <tibor_please.no.email_karaszi@.hotmail.nomail.com> wrote in
message news:e6n8opOlEHA.3372@.TK2MSFTNGP09.phx.gbl...
> Anti.virus program trying to AV check the named pipes stuff?
> --
> Tibor Karaszi, SQL Server MVP
> http://www.karaszi.com/sqlserver/default.asp
> http://www.solidqualitylearning.com/
>
> "Bob Castleman" <nomail@.here> wrote in message
> news:uZ2a5kOlEHA.3392@.TK2MSFTNGP14.phx.gbl...
>
|||"Bob Castleman" <nomail@.here> wrote in message
news:OAh7NLOlEHA.324@.TK2MSFTNGP11.phx.gbl...
> Why would one choose Named Pipes over TCP/IP for the client net lib? We
have
> noticed for some of our customers that changing to named pipes speeds up
the
> response times of our application. But it is my understanding that a
> properly configured LAN should work fine with TCP/IP.
Don't know if this is your problem, but there was a bug a while ago in SQL
Server 2000 to do with the introduction of Kerberos aware authentication
using DBLibrary connections over TCP/IP.
IIRC, the SQL Client and SQL Server would not correctly negotiate packet
sizes. This could be circumvented by explicitly overriding the default
packet size in the client app. This was resolved in SQL Server SP2.
Kind Regards, Howard
|||Yes I have seen where issues with DNS have caused situations such as this as
well.
Andrew J. Kelly SQL MVP
"Gert E.R. Drapers" <GertD@.SQLDev.Net> wrote in message
news:e4O$dsOlEHA.3712@.TK2MSFTNGP15.phx.gbl...
> An other option we see often are a DNS name resolution problems, use
tracert
> to determine the route taken.
> An other thought are these notebooks, that connect wireless as well, or at
> home? In which case it most of the times helps to dump the DNS resolver
> cache of the workstation. See IPCONFIG /flushdns
> GertD@.SQLDev.Net
> Please reply only to the newsgroups.
> This posting is provided "AS IS" with no warranties, and confers no
rights.
> You assume all risk for your use.
> Copyright SQLDev.Net 1991-2004 All rights reserved.
> "Tibor Karaszi" <tibor_please.no.email_karaszi@.hotmail.nomail.com> wrote
in[vbcol=seagreen]
> message news:e6n8opOlEHA.3372@.TK2MSFTNGP09.phx.gbl...
network[vbcol=seagreen]
enough[vbcol=seagreen]
than[vbcol=seagreen]
pipes.[vbcol=seagreen]
over[vbcol=seagreen]
clients[vbcol=seagreen]
it[vbcol=seagreen]
chatter[vbcol=seagreen]
sent[vbcol=seagreen]
We
>
|||Same here.
Basically if the DNS server is unavailable the client will wait for 20-30
seconds, and then resolve the servername in a different way.
You can do a quick check if this is the issue by connecting to the server
via both the IP-address and the servername. IP-address should behave
normally and the servername will have the 30 second delay.
Jacco Schalkwijk
SQL Server MVP
"Andrew J. Kelly" <sqlmvpnooospam@.shadhawk.com> wrote in message
news:uD8FsIPlEHA.3356@.TK2MSFTNGP14.phx.gbl...
> Yes I have seen where issues with DNS have caused situations such as this
> as
> well.
> --
> Andrew J. Kelly SQL MVP
>
> "Gert E.R. Drapers" <GertD@.SQLDev.Net> wrote in message
> news:e4O$dsOlEHA.3712@.TK2MSFTNGP15.phx.gbl...
> tracert
> rights.
> in
> network
> enough
> than
> pipes.
> over
> clients
> it
> chatter
> sent
> We
>
|||Thanks,
I'll try that.
"Jacco Schalkwijk" <jacco.please.reply@.to.newsgroups.mvps.org.invalid > wrote
in message news:uXQM9FQlEHA.3876@.TK2MSFTNGP15.phx.gbl...
> Same here.
> Basically if the DNS server is unavailable the client will wait for 20-30
> seconds, and then resolve the servername in a different way.
> You can do a quick check if this is the issue by connecting to the server
> via both the IP-address and the servername. IP-address should behave
> normally and the servername will have the 30 second delay.
> --
> Jacco Schalkwijk
> SQL Server MVP
>
> "Andrew J. Kelly" <sqlmvpnooospam@.shadhawk.com> wrote in message
> news:uD8FsIPlEHA.3356@.TK2MSFTNGP14.phx.gbl...
>