🚀 go-pugleaf

You've got a *lot* more going on there than just "network performance". You're
getting requests, processing them, pulling bits off of those drives, then
spitting them out to the network. It's hardly a simple "network performance"
problem. It's a *server* performance problem, and from what little I know, the
disk subsystem felt like the place to look first.

You can also start using performance monitor to see what subsystems are the
limiting factor. This takes time and some insight to figure out just what to
monitor.

JimR

Moshe <moshem@email.com> wrote:
>Hi
>
>why the scsi is your main suspect ? I said that the problem I experiened was
>network performence oriented.
>
>I have a feeling your are right , but why the scsi ?
>
>Moshe
>
>In article <37c3fc0b.427220285@news.atl.bellsouth.net>,
>  noone@bellsouth.net (Jim) wrote:
>> It feels like it must be in the drive subsystem, if the rest of the hardware and
>> the software config is so identical.
>>
>> Why RAID 0 on the two 4.5G's, which I assume are system drives? Can they be set
>> up as RAID 1?
>>
>> Are all the drives on the same SCSI Channel? If so, can they be seperated onto
>> two channels? In general, I'd look closely at the SCSI config.
>>
>> JimR
>>
>> Moshe <moshem@email.com> wrote:
>>
>> >Hi
>> >
>> >
>> >I created an TCP/IP Server that accepts connections from about 300 sockets
>> >and keeps them alive , when one socket sends a request the app understande it
>> >, query a local Infomix server , takes the data and send it over back via the
>> >same socket.
>> >
>> >
>> >I have used I/O completion ports ( Windows NT ) to implement the I/O requrest
>> >queues .
>> >
>> >I developend the app on a IBM Intellistation with 2 Intel Pentium III 500
>> >cpu's and 256MB of RAM , it worked great , the entire process takes about
>> >20-45 ms for each request-query-send cycle!
>> >
>> >Now I bought a larger server configuration for my actual platform , it's an
>> >IBM Netfinity 5500 M20 , with 2 Intel Pentium III 500 XEON 512KB cpu's and
>> >same configuration for the other components as the Intellistation.
>> >
>> >the only difference is the RAID card that is suppose to be stronger on the
>> >netfinty.
>> >
>> >
>> >to my horror , I found at first that I got 500ms for the enitre process !! I
>> >reinstaled all the hardware and software and now I get ~90ms.
>> >
>> >still , I pointed it out no to be an informix problem ( at least with the
>> >query ) that is the same time.
>> >
>> >What I did notice is the recieve cycle ( actuall made out of select , recv )
>> >takes twice the time , and the send cycle take about the same .
>> >
>> >I disable the onboard adapter and installed an Intel PRO 100  , the results
>> >are the same.
>> >
>> >this is the nefinity configuration :
>> >
>> >
>> >Netfinity 5500 M20
>> >
>> >2 x CPU Pentium III 500 XEON 512K
>> >
>> >256MB RAM ECC
>> >
>> >OnBoard AMD Fault Tolorent Adapter 10/100
>> >    also tried Intel Pro 100 PCI
>> >
>> >2 x 4.5GB HDD , one logical drive ,config as Raid 0
>> >
>> >40GB Hdd's , one logical drive , config as Raid 5
>> >
>> >Windows NT 4 SP 4 ( 1-4 cpu edition )
>> >
>> >Informix 7.30
>> >
>> >IE 4
>> >
>> >ALL services are idential
>> >all system parameters are identical.
>> >
>> >
>> >
>> >
>> >can some one give me a hint of what is going on , why a huge server can't get
>> >the same results as a workstation that costs half as much ?
>> >
>> >
>> >any help is greatly appreciated !
>> >
>> >Moshe
>> >
>> >
>> >
>> >
>> >
>> >
>> >Sent via Deja.com http://www.deja.com/
>> >Share what you know. Learn what you don't.
>>
>>
>
>
>Sent via Deja.com http://www.deja.com/
>Share what you know. Learn what you don't.