VoIP Mailing List Archives

[asterisk at tescogroup...]

zttest:

--- Results after 44 passes ---
Best: 100.000000 -- Worst: 99.987793 -- Average: 99.988350

Dell 2950
Wildcard TE410P (2nd Gen) (rev 02) (with echo cancel daughter board)
1 PRI configured to Telco
1 PRI configured to old Panasonic DBS 576 being used just as a mux for our fax machines.

--
--
Steven

http://www.connectech.org/

"broadband Voice" <broadbandvoice at gmail.com> wrote in message news:b37f000c0801260315q2699b975pfc788a8265509539 at mail.gmail.com...
How did this workout? I am considering getting the Dell PowerEdge 2950.


On 11/5/07, Steven <asterisk at tescogroup.com> wrote:
2950s work fine.

I have had the parity error for over a year with no noticable problems. It is working fine.

I did have to make some IRQ changes to clean up the system.

I did these on my Dell 1750 test machine, but have made the same changes on my production machine.
The changes basically redue the IRQ load from other cards, like the RAID card, which will reduce the bus's capacity for processing
all of the TDM IRQs.
It also allocates just one CPU full time for all of the TDM IRQs.

The changes are below:


ref:
FYI on zttool output on SMP system

--- Results after 56 passes ---
Best: 100.000000 -- Worst: 99.987793 -- Average: 99.999564
Only 2 were 99.987793, the 54 others were all 100.000000.

I got this by making the changes below on my dual proc Dell 1750.

setpci -v -s 01:08.1 LATENCY_TIMER=8
setpci -v -s 00:0f.1 LATENCY_TIMER=8
setpci -v -s 01:04.0 LATENCY_TIMER=8
setpci -v -s 01:02.0 LATENCY_TIMER=8
setpci -v -s 00:0f.2 LATENCY_TIMER=8
setpci -v -s 01:04.0 LATENCY_TIMER=8 (these are USB, SCSI HW RAID driver, Ethernet, Video, etc. I did not alter ZAP cards, nor any
bridges or buses)

echo 1 > /proc/irq/17/smp_affinity (Ethernet)
echo 1 > /proc/irq/18/smp_affinity (SCSI HW RAID Driver)
echo 2 > /proc/irq/20/smp_affinity (TDM)
echo 2 > /proc/irq/24/smp_affinity (TE411P)

I also turned of the startup of irqbalance.

The setpci changes did the most work concerning reaching 100% in zttest.

Irqbalance was causing the the processor handling the interrupts of the zap cards to change very often.
This would impose a delay during the change and cause the zttest numbers to drop/be inconsistent.

Because I turned irqbalance off, the irqs are processed round robin style, which is also not good.
Therefore, I hard coded the processor affinity for the zap cards to one proc and all other high load irqs to the other proc.
If you have more than 2 procs, you can spread them out even more. If you do not turn off irqbalance, the affinity changes will be
overwritten by it.

I made these changes on a live system without issue.
I set these changes in /etc/rc.d/rc.local to reset them after reboots.

--
--
Steven

http://www.glimasoutheast.org



"Brian Hutchinson" <b.hutchman at gmail.com> wrote in message news:3d1967ab0710250333r62643f08nc8efa99ec6c6e2a0 at mail.gmail.com...
You guys are scaring me! I'm building a 2950 with SAS RAID 5 on the new PERC and it will have 2 TE420P's. I hope it works or my
bacon will fry.


On 10/25/07, Joseph Begumisa <joe at cfi.co.ug> wrote: