|
Hi Jason,
Thanks for the info. It is mostly some race in state transition on SMP
box, which leads to those negative problem.
We added a simple spinlock to fix this problem, since the collision seldom
happens, there won't be performance drawback. Please apply the attached
patch to ipvs-0.2.8 tar ball, build and test it on your smp box.
Thanks,
Wensong
On Wed, 4 Apr 2001, Jason Steenblik wrote:
> Here is the output from ipvsadm -ln. At the time it was taken only 4
> boxes were recieving traffic (172.18.2.23, 172.18.2.18, 172.18.2.13,
> 172.18.2.8) and one had just fallen out of favor (172.18.2.3). As I
> stated earlier all recieve traffic to start then one by one they are
> "disowned" and go quiet till there is just one left recieving traffic.
>
> [root@lb /root]# ipvsadm -ln
> IP Virtual Server version 0.2.8 (size=4096)
> Prot LocalAddress:Port Scheduler Flags
> -> RemoteAddress:Port Forward Weight ActiveConn
> InActConn
> TCP 172.19.0.6:25 lc
> -> 172.18.2.23:25 Masq 1 -45 448
> -> 172.18.2.18:25 Masq 1 -48 518
> -> 172.18.2.17:25 Masq 1 -10 10
> -> 172.18.2.15:25 Masq 1 -1 1
> -> 172.18.2.14:25 Masq 1 -12 12
> -> 172.18.2.13:25 Masq 1 -46 449
> -> 172.18.2.12:25 Masq 1 -1 1
> -> 172.18.2.11:25 Masq 1 -3 3
> -> 172.18.2.9:25 Masq 1 0 0
> -> 172.18.2.8:25 Masq 1 -50 558
> -> 172.18.2.7:25 Masq 1 -4 4
> -> 172.18.2.6:25 Masq 1 -9 9
> -> 172.18.2.5:25 Masq 1 -3 3
> -> 172.18.2.4:25 Masq 1 -5 5
> -> 172.18.2.3:25 Masq 1 -31 31
>
>
> I also took time to test with a non-smp kernel and everything works
> great. All servers get traffic and all keep getting traffic for the
> short (many hour) test I did. The output of ipvsadm -ln was a little
> different so I decided to paste it below.
>
> [root@lb /root]# ipvsadm -ln
> IP Virtual Server version 0.2.8 (size=4096)
> Prot LocalAddress:Port Scheduler Flags
> -> RemoteAddress:Port Forward Weight ActiveConn
> InActConn
> TCP 172.19.0.6:25 lc
> -> 172.18.2.23:25 Masq 1 0 175
> -> 172.18.2.18:25 Masq 1 0 174
> -> 172.18.2.17:25 Masq 1 1 134
> -> 172.18.2.15:25 Masq 1 0 181
> -> 172.18.2.14:25 Masq 1 0 166
> -> 172.18.2.13:25 Masq 1 1 165
> -> 172.18.2.12:25 Masq 1 0 170
> -> 172.18.2.11:25 Masq 1 1 162
> -> 172.18.2.9:25 Masq 1 2 163
> -> 172.18.2.8:25 Masq 1 0 169
> -> 172.18.2.7:25 Masq 1 0 169
> -> 172.18.2.6:25 Masq 1 0 187
> -> 172.18.2.5:25 Masq 1 0 178
> -> 172.18.2.4:25 Masq 1 2 130
> -> 172.18.2.3:25 Masq 1 1 164
>
> This one looks better
>
> -Jason
>
ipvs_state_lock.patch
Description: Text document
|
