Class 4 and class 5: when one switch is not enough

7 min read

A dialer campaign and a retail customer share one softswitch. The campaign opens, calls per second jump, and the retail calls start to fail. Wholesale price lists change every week on that same box, while the PBX customers need recordings, resellers, and phones behind NAT.

This is for the owner or CTO who sells retail or hosted PBX and also carries wholesale minutes or call-centre traffic. MOR is the retail side. M4 is wholesale and dialer SIP. One traffic type has a shorter answer at the end: stay on that switch. Features and capacity are on the MOR vs M4 comparison. Definitions are in Class 4 vs Class 5 softswitch: two jobs. Split when the three signs below show up.

Three signs one switch is now a risk

A dialer burst beside retail customers

Call-centre traffic has a different shape from a phone on a desk. A predictive dialer can send a pile of new attempts in one second, and many of them fail before anyone answers. The concurrent-call counter can look calm while the switch sets up calls, writes CDRs, and checks balances. Call-center traffic on a softswitch walks through that shape. Which dialer customer burned your CPS route follows the shared trunk, where one originator spends the attempt budget and the quiet customers inherit the failures.

An operator running retail and conversational wholesale on one softswitch, at about 200 calls per second, looked at moving the call-centre traffic onto a second platform. The question was whether the second platform bills, routes, manages price lists, and raises alerts. M4 does those four. Billing and SIP routing are the wholesale job on the comparison page. Price lists are under Billing → Tariffs. Alerts are under SECURITY → Alerts: answer-seizure ratio, average call duration, post-dial delay, then a block or a notification. Price-file updates use the automatic tariff import add-on, and that add-on works the same way on MOR and on M4.

About 200 attempts per second is past the published MOR estimate. The multi-server MOR row tops out at 100 calls per second under ideal conditions, and the ordinary figure on that row is closer to 60. If dialer bursts are part of that 200, move the bursts onto M4.

The recording does not move with the burst. M4 has no call recordings. A dialer that sends SIP only to M4 keeps the recording on the dialer platform. MOR records a call that still passes through MOR. M4 bills the originator that sent the SIP. The campaign name stays in the dialer.

Wholesale price lists that keep changing

Carriers send a new file, then another. One destination prefix can carry two prices, and the caller’s number picks the row. That is origin-based rating. MOR stores a CLI group on the wholesale rate. M4 matches the rate when the destination prefix matches and the caller ID matches a prefix in the CLI group. An empty CLI group matches on the destination prefix. The M4 example is prefix 370 for caller IDs that start with 30767 or 3706. The longest suitable prefix wins.

Load those files by hand and a Monday price change becomes a Friday dispute. Automatic tariff import watches a mailbox, maps the attachment to the provider, and waits for a person to accept the import before new rates replace the old ones. Run it on the switch that owns that provider’s buy rates. Two prices on one prefix, worked through on MOR, is Origin-based rating: two prices, one prefix.

Retail price lists move on a slower clock. Keep the mid-week wholesale file on the wholesale switch.

PBX features on transit traffic

Transit has no use for voicemail, a ring group, or a recording of every call. Those belong to the retail customer. Recording and codec transcoding also take the CPU that call setup needs. The capacity notes treat both as outside ideal conditions.

PBX functions, recordings, resellers, live call spy, and the auto-dialer are MOR features. Call-centre agent screens sit on an external platform. M4 carries and bills the SIP. Its business models are wholesale transit, DID forwarding, SIP trunking, and GSM termination. MOR is the retail switch: PBX, recordings, resellers, a statement of account, and phones that register from behind NAT. M4 can sit behind NAT. Many phones registering from one address stay on MOR.

If the wholesale routes stay on MOR so the PBX can share the box, that transit runs on Asterisk, at the lower calls-per-second estimate.

What each side needs

Job MOR (Class 5) M4 (Class 4)
Retail customers, hosted PBX PBX functions, recordings, resellers, statement of account Not on M4
Phones behind NAT Built for many registrations from one address The wrong box when those registrations pile up on one address
Wholesale and dialer SIP Can carry transit. The published calls-per-second estimate is the lower one Wholesale billing, bilateral balances, and the higher attempt rates
Attempt cap per originator Per device and per provider: a call limit inside a number of seconds Per origination point and per termination point, in the same form
Two prices for one prefix CLI group on the rate CLI group on the rate
A supplier who is also a customer Assign a provider to a user. Buy and sell balances stay separate. Grouping providers under a terminator is a report, and it does not merge the supplier balance One user can be customer and supplier, with independent buy and sell balances. Connections of the same supplier share one supplier balance
New carrier price files Automatic tariff import add-on The same add-on, plus rate notifications

M4 is not an upgrade from MOR. The architecture is different, so an operator who leaves an all-in-one billing platform and wants both retail and wholesale installs two systems.

Calls per second decide the split

Concurrent calls are the calls up at one moment. Calls per second are the new attempts in that second, including attempts that do not connect. Dialer and call-centre traffic can need more processing at a low concurrent-call count. Short calls and failed attempts create more setup work than the same number of long conversations.

The figures below are estimates for suitable hardware under ideal conditions, current as of November 2026. Read them on the live pages before you order a server: Softswitch performance, Recommended hardware, and the MOR vs M4 comparison. The estimates are not guarantees. The highest concurrent-call figure and the highest calls-per-second figure are separate limits. A licence is a third ceiling. The plan can allow fewer concurrent calls than the hardware can run, and the traffic has to fit both.

Ideal conditions mean no codec transcoding and no call recording, a quiet web interface, and no heavy reports or invoice runs beside the calls. MOR runs on Asterisk. M4 runs on Kamailio, SEMS, and RTPEngine. The comparison page records that split. Server count for a calls-per-second target is in How big a system: concurrent calls, CPS and servers.

Platform Concurrent calls Calls per second
MOR, one server up to 500 closer to 20 in practice, up to 50 under ideal conditions
MOR, several servers up to 2,000 closer to 60 in ordinary use, up to 100 under ideal conditions
M4, one server up to 500 up to 100
M4, several servers up to 10,000 up to 1,000

One observed MOR deployment used recommended hardware: a single server with 24 CPU cores and almost no web interface use. It carried wholesale transit at up to 500 concurrent calls and up to 1,500 call attempts per minute, an average of 25 attempts per second. Peaks inside the minute can sit higher. Treat the 25-per-second figure as that one deployment.

If the 200 attempts per second from the mixed switch are on MOR, they are eight times that 25-per-second average, and they sit above the multi-server MOR estimate of 100 under ideal conditions. They sit inside the multi-server M4 estimate of 1,000. Retail, with recordings and registrations, stays on MOR. The attempt-heavy traffic moves to M4.

Production hardware on that page starts at 8 CPU cores, 32 GB of RAM or more, and 500 GB or more of SSD or NVMe, on Rocky 9 minimal with the timezone set to UTC. On M4, more than 700 concurrent calls needs an Intel I210, I350, or X550T. A standby server is for redundancy. Extra active call-processing servers raise capacity.

A pilot checks the estimate against peak calls per second for each traffic type, average duration, and whether recording or transcoding will be on.

Two balances, two invoices, one SIP trunk

Each product keeps its own administration, its own users, and its own balances. The retail customer receives MOR’s invoice for seats and retail minutes. Statement of account is a MOR screen. The wholesale customer receives M4’s invoice for the minutes M4 carried. A customer who buys both receives those two invoices. A single combined bill is a separate step. Confirm it before you promise one.

For retail traffic that must leave on a wholesale route, the call starts on MOR and hands off to M4 over SIP. M4 rates the call and sends it to the carrier. A dialer hands its attempts to M4, with a calls-per-second cap on that origination point, so the burst stops there instead of on the retail server. A call that starts on M4 and must reach a MOR user is a trunk in the other direction. Match the trunk to the side that starts the call. Keep the balance on the switch that owns the customer.

A junior follows a failed retail call by the minute, the caller number, and the dialed number. On MOR, open STATISTICS → Calls → Last Calls and search that Source and Destination. Read the hangup cause and the provider. When that provider is the trunk toward M4, the call left MOR. On M4, open REPORTS → Calls List, search the same Source and Destination in that minute, and open Call Info on the row. M4 Call Tracing places a fake call and does not ring the carrier, so a call that already failed belongs on Call Info. The two switches do not share one call id.

Origin-based rating stays on the switch that rates the call. Automatic tariff import stays on the switch that receives that provider’s file. If both switches buy from carriers, run the add-on on both. Load a carrier file on the switch that will rate those calls.

Stay on one switch in these cases

The business has one traffic type. Retail, resellers, and recordings point to MOR. Wholesale transit and a high attempt rate point to M4. Use the MOR vs M4 comparison as the checklist, and Class 4 vs Class 5 softswitch: two jobs as the definition.

The second traffic type is small, and peak calls per second for the whole business fit one row in the table once you count the features you will turn on. Recording and transcoding move you off the ideal-conditions row. Stay on one switch while that row covers the peak.

Two administration screens, two backups, and two sets of price lists need an engineer. With nobody to run the pair, cap calls per second on the dialer on the switch you run today. Which dialer customer burned your CPS route is that cap and the CDR check.

FAQ

Can MOR be upgraded to M4?

No. The architecture is different. M4 is its own system.

Does M4 include billing, routing, price lists, and alerts?

Yes. Wholesale billing and SIP routing are the product. Price lists are Billing → Tariffs, including CLI groups for origin-based rating. Alerts are SECURITY → Alerts. Rate notifications are an M4 feature. Automatic tariff import, the add-on that turns a carrier spreadsheet into rates, is the same add-on on MOR.

Does one customer get one invoice?

The retail invoice comes from MOR. The wholesale invoice comes from M4. A customer who buys both receives both. Confirm a single combined bill before you write it into a contract.

Does M4 include the dialer?

The dialer and the agent desktop stay outside. M4 carries the SIP and bills the originator. The recording of a burst that only touches M4 stays on the dialer.

Three replies. If both traffic types are on your network and someone can run two systems, reply with your peak CPS per traffic type and we’ll propose a split, then prove it in a pilot. If both are real and nobody can run the pair, cap calls per second on the dialer on the switch you run today, and do not open a second system yet. If you have one traffic type, stay on that switch. The comparison page is the checklist. The second and third replies are not a request to open another system.


Send peak CPS for each traffic type

Reply with your peak CPS per traffic type and we'll propose a split, then prove it in a pilot.

Contact us

Leave a Reply

Your email address will not be published. Required fields are marked *