Does adding SSD to a server make sense?

[paulmo82]

Hello all,

Hardware: Quad Core Xeon X3350 2x250GB HDD 8GB DDR2 RAM

Software: Vicidial VERSION: 2.8-404a BUILD: 130605-0841

Our ‎Campaign: we are running 3 campaigns, with three remote agents assigned to each campaign, and a dial level of 18. we have the server setup to allow 150 max trunks @ 20 calls per second and are running at about 50 concurrent calls per campaign. The dropped percentage is high (10-30%) on all campaigns. If we increase the calls per second or max trunks, many of the calls will ring and disconnect after one ring. We are looking to stabilize these campaigns and add on more.

we would like to increase our dial levels to 1000 concurrent calls, and we are not sure what the best hardware configuration is.

does it make sense to replace the SATA's with a single SSD drive. Will that increase capacity?

We are currently using the above mentioned hardware in a colo and it's pretty inexpensive (around $100 / month). Adding SSD Drive is an additional $25.

or should we just add a 2nd, 3rd and 4th server etc.

is using a cluster going to get us more call efficiency per hardware … we are able to manage separate campaigns on different servers so there isn't really a business need to have a cluster… only performance

if adding SSD is just an incremental improvement -- we would rather get a new server.


thanks in advance for your feedback.


Paul M

[rrb555]

if would be best if u add another telephony to increase concurrent calls.

[williamconley]

when you reach capacity on a single server environment, it is better to add a second server rather than attempting to "beef up" the single server.

Vicibox 4.0.3 is quite capable of adding a second server to your system which would make it a "cluster". The Vicidial Group reports clusters well in excess of 10 machines.

Clustered servers must be on the same physical network, and it is best to have them share a "local" intranet (and have all the servers use the Local IP for Vicidial operations). They would use the public IPs for external access (agents and admins) in addition to trunking. This allows extremely fast interserver communications (no security needed on the local net) and secure access to the external IPs.

[paulmo82]

Thanks Gents, we did end up going with a 5 machine cluster and it seems to be working very well.

[williamconley]

Excellent. Please tell me you have learned to use the backup routine ... and that your database backups are being pushed to an ftp server NOT in the same physical location. LOL

Next, of course, a nice audio archive server with both ftp and http so your recordings do not reside on your dialers forever (hd failure = lost recordings ...).

And then go through your "crontab -e" on the db server and activate pruning/archiving routines for long-term usage to avoid bloated log tables.

[paulmo82]

We have all those milestones in place now we are primarily doing outbound survey campaigns broadcasting a 10-12 second informational message to our clients. we are running 90 calls per server, with three remote agents (30 dial level) per server and getting a very high dropped call percentage. Should I use more remote agents and lower the dial level? Would that reduce the dropped percentage?

thanks,
Paul M

[williamconley]

reduce your calls per second. watch your call stream and set it up for a constant, slow, call stream instead of "bunch of call" ... dead zone ... "bunch of call".

also be sure your "available only tally" is set to YES (this shuts down the dialer when agents are on calls!)

[paulmo82]

awesome, thanks for the feedback. I have made the adjustments (running 10 CPS per server & Turned on Available only tally). I will send feedback on the campaigns later today!

[paulmo82]

Here is a snapshot of one of the campaigns. 8,974 calls made, looks like only 9 calls were dropped and 23 answered.the percentage could be a little misleading.

DIAL LEVEL: 30 TRUNK SHORT/FILL: 0 / 0 FILTER: NONE TIME: 2013-07-31 07:19:24
DIALABLE LEADS: 104523 CALLS TODAY: 8974 AVG AGENTS: 3 DIAL METHOD: RATIO
HOPPER LEVEL: 200 DROPPED / ANSWERED: 9 / 23 DL DIFF: -11.25 STATUSES: NEW
LEADS IN HOPPER: 266 DROPPED PERCENT: 39.13% DIFF: -375.00% ORDER: RANDOM
88 calls being placed 69 calls ringing 19 calls waiting for agents 0 calls in IVR
3 agents logged in 0 agents in calls 3 agents waiting 0 paused agents

[williamconley]

misleading? 23 calls answered and 9 dropped. that's not misleading, that's apalling.

look at it from the FCC perspective: you annoyed 40% of the public you interacted with. The max limit is 3% and they are tasked with finding and prosecuting you.

look at it from the public perspective: nearly half of all the people you called had their time wasted.

look at it from your accountant's perspective: he paid for leads, you "burned" 40% of them. Wasted money.

40% dropped is way "not cool".

Just sayin' ...

[paulmo82]

lol, no no no... dont worry about that... these are all opted in subscribers to a free service. we are simply delivering free informational messages. there is no up sell what so ever....

[paulmo82]

And i completely agree it about 40% being way not cool, was trying to figure out how to reduce that amount. wasn't sure the best method as of now, didn't know if it was an issue on the telephony side or the dialer side. have been adjusting the campaigns in attempt to reduce it...

[williamconley]

look at your asterisk cli.

if you see "bursts" of "dials", followed by long period of activity that does not include dialing, then another burst of dialing ....

set your max calls per second to a lower number. if you get it set well, it will continually dial with only a couple seconds of "no dial" between bursts.

that will mean that instead of "dial bunch" followed by "answer bunch" which results in "not enough agents!" and dropped calls, you could have "slow dial ..." and "slow answers" and rotate your agents in and out of calls to catch incoming calls more randomly instead of in bunches.

and as i said earlier, be sure your available only tally is set to "Y"!

and upgrade. later versions are smoother because of the congestion detection system.