|
Agreed about the real servers.
We use LVS to load balance VoIP signaling, and so I don't have a payload size,
but each call brings at least two TCP sockets through the LVS, not to mention
the numerous UDP requests that get sent for each call.
Rob.
>-----Original Message-----
>From: lvs-users-admin@xxxxxxxxxxxxxxxxxxxxxx
>[mailto:lvs-users-admin@xxxxxxxxxxxxxxxxxxxxxx]On Behalf Of Wayne
>Sent: Thursday, November 01, 2001 3:26 PM
>To: lvs-users@xxxxxxxxxxxxxxxxxxxxxx
>Subject: RE: NAT Performance and what is a good bench mark?
>
>
>Depends on what kind of real server you have, it could take
>half the world to get them fully exhausted. I run a test on
>a box with 500 new connection per second, total 50,000
>connection on it, it only used about 25MB memory out of
>128MB, and CPU usage is about 8% of a Celeron 600MHz.
>
>At 03:16 PM 11/1/2001 -0800, you wrote:
>>I actually have the same question. I have seen on the web site
>that it only recommends 20 real servers behind the LVS. What is
>this based upon?
>>
>>Rob.
>>
>>>-----Original Message-----
>>>From: lvs-users-admin@xxxxxxxxxxxxxxxxxxxxxx
>>>[mailto:lvs-users-admin@xxxxxxxxxxxxxxxxxxxxxx]On Behalf Of Wayne
>>>Sent: Thursday, November 01, 2001 2:51 PM
>>>To: lvs-users@xxxxxxxxxxxxxxxxxxxxxx
>>>Subject: RE: NAT Performance and what is a good bench mark?
>>>
>>>
>>>You bottleneck is not CPU or memory. It is the network cards.
>>>If you have heavy outgoing connections, the number of ports
>>>could be masqueraded could be a limiting factor, too.
>>>
>>>At 02:33 PM 11/1/2001 -0800, you wrote:
>>>>I too have been looking to benchmark an LVS-NAT setup using the
>2.4 kernel
>>>>but I'm not sure what a good benchmark is? What do people
>suggest for the
>>>>following setup?
>>>>
>>>># | |
>>>># | client |
>>>># |________|
>>>># CIP=eth0 192.168.1.13
>>>># |
>>>># |
>>>># VIP=eth0:0 192.168.1.200/32
>>>># __________
>>>># | |
>>>># | director |
>>>># |__________|
>>>># DIP=eth2:254 10.10.10.200/24
>>>># |
>>>># |
>>>># |
>>>># --------------------------------------------
>>>># | | |
>>>># | | |
>>>># RIP1=eth0 RIP2=eth0 RIP=eth0
>>>># 10.10.10.10 10.10.10.20 10.10.10.30
>>>># ______________ ______________ ________________
>>>># | | | | | |
>>>># | rs1 | | rs2 | | rs3 |
>>>># |______________| |______________| |________________|
>>>>
>>>>Director = 2x 1GHz 1Gig RAM 2 onboard Pro100's and 1 64bit/66MHz
>>>>Pro1000(talks to real server)
>>>>
>>>>Realservers = 3X same config as above but no Pro1000 (could add though)
>>>>
>>>>All running the 2.4-9 linux kernel with that level of LVS code.
>>>The network
>>>>is fully switched 100/1000.
>>>>
>>>>Any suggestions would be helpful. Definitely looking to
>benchmark web or
>>>>streaming type content.
>>>>
>>>>--Ben Odom
>>>>
>>>>
>>>>
>>>>
>>>>
>>>>
>>>>
>>>>
>>>>
>>>>
>>>>-----Original Message-----
>>>>From: Peter Mueller [mailto:pmueller@xxxxxxxxxxxx]
>>>>Sent: Wednesday, October 31, 2001 1:08 PM
>>>>To: 'lvs-users@xxxxxxxxxxxxxxxxxxxxxx'
>>>>Subject: RE: NAT Performance
>>>>
>>>>|I'm considering LVS as a replacement to Cisco LocalDirectors to front
>>>>|several SMTP server farms. Due to customer-end constraints we
>>>>|need to do
>>>>|this via NAT. Target throughput is up to 40 Mbps. Is this
>>>>|realistic? LVS
>>>>|hardware is likely to be twin 1GHz Pentiums.
>>>>|
>>>>
>>>>40mbps sounds all right on dual-1ghz box, assuming you use later
>>>2.4 kernel
>>>>on your director. Be very stingy on your NICs.. for example
>>>eepro100's seem
>>>>to historically be 'uncertain' with high bandwidth..
>>>>
>>>>I haven't seen a "gauranteed" figure with newer 2.4 kernels. However,
>>>>lurking on the mailing list for a year or so now has led me to
>>>believe that
>>>>the key LVS people now believe NAT (with 2.4 kernel) to perform
>similar to
>>>>DR. LVS-DR easily exceeds 40mbs, assuming you have decent hardware.
>>>>
>>>>some probably outdated performance evaluations:
>>>>http://www.linuxvirtualserver.org/Joseph.Mack/performance/single_r
>>ealserver_
>>>performance.html
>>>and
>>>http://www.linuxvirtualserver.org/Documents.html#manuals (look under
>>>performance)
>>>
>>>===
>>>
>>>bottom line : if I were you, I would have a basic assumption that it will
>>>perform and proceed directly to your testing of LVS-NAT and see
>if it does.
>>>It would only take 2-3 days and could possibly save you lots of $.
>>>
>>>Peter
>>>
>>>_______________________________________________
>>>LinuxVirtualServer.org mailing list - lvs-users@xxxxxxxxxxxxxxxxxxxxxx
>>>Send requests to lvs-users-request@xxxxxxxxxxxxxxxxxxxxxx
>>>or go to http://www.in-addr.de/mailman/listinfo/lvs-users
>>>
>>>_______________________________________________
>>>LinuxVirtualServer.org mailing list - lvs-users@xxxxxxxxxxxxxxxxxxxxxx
>>>Send requests to lvs-users-request@xxxxxxxxxxxxxxxxxxxxxx
>>>or go to http://www.in-addr.de/mailman/listinfo/lvs-users
>>
>>
>>_______________________________________________
>>LinuxVirtualServer.org mailing list - lvs-users@xxxxxxxxxxxxxxxxxxxxxx
>>Send requests to lvs-users-request@xxxxxxxxxxxxxxxxxxxxxx
>>or go to http://www.in-addr.de/mailman/listinfo/lvs-users
>>
>>
>>_______________________________________________
>>LinuxVirtualServer.org mailing list - lvs-users@xxxxxxxxxxxxxxxxxxxxxx
>>Send requests to lvs-users-request@xxxxxxxxxxxxxxxxxxxxxx
>>or go to http://www.in-addr.de/mailman/listinfo/lvs-users
>
>
>_______________________________________________
>LinuxVirtualServer.org mailing list - lvs-users@xxxxxxxxxxxxxxxxxxxxxx
>Send requests to lvs-users-request@xxxxxxxxxxxxxxxxxxxxxx
>or go to http://www.in-addr.de/mailman/listinfo/lvs-users
>
|
