Scaling Email Infrastructure

September 11, 2025 · 15 min read · Jaan Tamm

Scaling email infrastructure from tens of thousands to millions of daily messages requires deliberate architectural decisions that aren't simply "add more IPs." The right scaling approach depends on your volume pattern, content type mix, audience quality, and the deliverability control you need at each ISP. This guide covers the scaling decisions from IP pool architecture through multi-domain strategy, stream separation, and the monitoring infrastructure needed to operate at scale.

Best practices

following industry standards reduces deliverability failures 80%

Data-driven

measuring before and after every change is essential

ISP feedback

use Postmaster Tools, SNDS, and FBL data to guide decisions

Iterative

email deliverability is never a one-time fix — it requires ongoing monitoring

Scaling Triggers: When to Expand Infrastructure

Infrastructure expansion is justified when you hit specific operational constraints — not as proactive capacity planning:

ISP concurrency ceilings: Your delivery rate to a specific ISP is limited by the number of concurrent connections you can maintain, and you're at the per-IP limit. Adding IPs allows more concurrent connections to that ISP.
Stream contamination: Marketing complaints are affecting transactional delivery (or vice versa). The fix is stream separation with dedicated IPs per stream type.
Warm-up bottleneck: New sending volume exceeds what a warming IP can handle within your time constraints. Additional IPs allow parallel warming.
Content type isolation: Different content types (newsletters, promotions, transactional) have different reputation profiles. Isolation prevents cross-contamination.

Do not add IPs when the bottleneck is complaint rate, list quality, or content — adding IPs to a poor-quality programme just distributes the same poor performance across more infrastructure.

IP Pool Architecture

Structure your IP pool around sending streams rather than volume alone:

# Example IP pool architecture for a marketing platform

# Stream 1: Critical transactional (password resets, 2FA, billing)
TX_IP_1=203.0.113.10    # Sending from tx.yourdomain.com
# One IP dedicated; very low volume; near-zero complaints; pristine reputation

# Stream 2: Marketing email (campaigns, newsletters)
MKT_IP_1=203.0.113.11   # Sending from mail.yourdomain.com
MKT_IP_2=203.0.113.12   # Load distributed across 2+ IPs for volume
MKT_IP_3=203.0.113.13   

# Stream 3: Bulk promotional (highest volume, most risk)
BULK_IP_1=203.0.113.20  # Sending from news.yourdomain.com
BULK_IP_2=203.0.113.21  # Multiple IPs for throughput
BULK_IP_3=203.0.113.22
BULK_IP_4=203.0.113.23

# Stream 4: Cold outreach (B2B prospecting)
COLD_IP_1=203.0.113.30  # Sending from outbound.yourdomain.com
# Completely separate; highest complaint risk; must not contaminate above

This architecture ensures that complaints from marketing campaigns don't affect the transactional stream's reputation at any ISP, and that cold outreach activity is completely isolated from the rest of the programme.

Multi-Domain Strategy for Reputation Isolation

Separate sending subdomains create independent domain reputation pools at ISPs:

tx.yourdomain.com — transactional email; highest reputation maintained
mail.yourdomain.com or em.yourdomain.com — permission-based marketing
news.yourdomain.com — newsletter and content email
promo.yourdomain.com — promotional email with higher complaint tolerance
outbound.yourdomain.com — cold outreach from separate domain entirely

Each subdomain requires its own: SPF record, DKIM keys, DMARC record (or inherits from parent domain), PTR records for associated IPs, and separate Gmail Postmaster Tools domain verification.

Domain warming follows the same principles as IP warming — a new subdomain has zero domain reputation and must be introduced gradually with high-quality audiences.

Sending Stream Separation

Beyond IP and domain separation, implement stream separation at the application injection layer:

# PowerMTA: route by X-Binding header
# Application injects with X-Binding: transactional
# PowerMTA routes to vmta-transactional using KeyTable or binding maps

<virtual-mta vmta-transactional>
  smtp-source-ip 203.0.113.10
  ehlo-hostname tx.yourdomain.com
  max-smtp-out 20
</virtual-mta>

<virtual-mta-pool pool-marketing>
  virtual-mta vmta-marketing-1
  virtual-mta vmta-marketing-2
</virtual-mta-pool>

# Applications inject to different listening ports or with different auth credentials
# Transactional systems inject on port 2525 (mapped to vmta-transactional)
# Marketing systems inject on port 2526 (mapped to pool-marketing)

DNS Configuration at Scale

At scale, DNS configuration complexity grows with every added domain, IP, and stream. Maintain a DNS configuration registry:

# Minimum DNS records per sending domain/subdomain

# MX: where to receive bounces
mail.yourdomain.com.  MX  10  bouncehandler.yourdomain.com.

# A record: for EHLO/HELO identification  
mail.yourdomain.com.  A   203.0.113.11

# SPF: authorise sending IPs
mail.yourdomain.com.  TXT  "v=spf1 ip4:203.0.113.11 ip4:203.0.113.12 ip4:203.0.113.13 -all"

# DKIM: signing key (per selector)
mail2026q2._domainkey.mail.yourdomain.com.  TXT  "v=DKIM1; k=rsa; p=..."

# DMARC: policy for this subdomain (or inherit from parent)
_dmarc.mail.yourdomain.com.  TXT  "v=DMARC1; p=quarantine; rua=mailto:dmarc@yourdomain.com"

# MTA-STS: transport security (optional but recommended)
_mta-sts.mail.yourdomain.com.  TXT  "v=STSv1; id=20260414"

# TLS-RPT: TLS failure reporting
_smtp._tls.mail.yourdomain.com.  TXT  "v=TLSRPTv1; rua=mailto:tls@yourdomain.com"

Monitoring Architecture for Large Operations

At scale, monitoring must aggregate across all streams and surfaces problems automatically before they become incidents:

Delivery metrics per stream: Messages delivered/hour, bounce rate, deferral rate — per vMTA or sending domain
Reputation signals per domain: Gmail Postmaster spam rate, SNDS data for each sending IP, per-ISP delivery rate trends
Blacklist monitoring: Automated daily check of all sending IPs against major DNSBLs; alert on any new listing within minutes
Authentication monitoring: DMARC aggregate report processing showing pass rates per source; alert on any previously-passing source that begins failing
FBL complaint rate: Complaints per hour from Yahoo/Microsoft FBL; alert if hourly rate exceeds threshold

A monitoring stack for this at scale typically involves: Prometheus exporters for MTA metrics, Grafana dashboards with alerting rules, a DMARC report processor (dmarcian or custom) feeding into the same monitoring stack, and PagerDuty or equivalent for on-call escalation when alerts fire.

Scaling Architecture: From Single IP to Multi-Pool Infrastructure

The path from a single sending IP to a fully scaled multi-pool infrastructure follows predictable stages, each requiring specific configuration changes in PowerMTA and in your DNS/authentication setup. Understanding the architecture at each stage prevents the common mistake of trying to jump from Stage 1 to Stage 4 without building the intermediate capabilities that each stage requires.

Stage 1 (0–500K messages/month): Single IP, single domain, single VMTA. All traffic — transactional, promotional — shares one IP and one reputation pool. Acceptable at this volume; the throughput and reputation isolation benefits of additional IPs are not yet commercially significant enough to justify the warmup investment.

Stage 2 (500K–2M messages/month): 2–4 IPs, VMTA separation between transactional and promotional traffic. Transactional messages (password resets, order confirmations) move to a dedicated VMTA with 1–2 IPs whose reputation is protected from promotional complaint rates. Promotional traffic uses 2–3 IPs with a separate VMTA. The authentication stack (DKIM, SPF, DMARC) remains on the same sending domain, but Return-Path subdomain separation is implemented.

Stage 3 (2M–10M messages/month): 6–15 IPs across 3+ VMTAs. Cold email outreach (if part of the programme) moves to its own VMTA with dedicated IPs and a separate DKIM signing domain. The promotional VMTA expands to 4–8 IPs to support campaign delivery windows. Per-ISP domain block configuration becomes critical at this scale — generic settings produce visible throttle pressure that calibrated per-ISP settings resolve.

Stage 4 (10M+ messages/month): 15–50+ IPs, geographic IP distribution (IPs in EU datacentres for EU recipient segments, US-region IPs for US segments), dedicated monitoring infrastructure (accounting log ETL pipeline, operational database, real-time dashboards), and potentially multiple PowerMTA instances for geographic distribution and redundancy.

Domain Scaling: One Domain vs Multiple Sending Domains

As sending volume and traffic diversity increase, the single-sending-domain model creates SEO and reputation complications. A mature multi-programme infrastructure typically uses: brand.com as the DKIM signing domain for permission-based marketing email (Postmaster Tools domain reputation builds here); tx.brand.com or notifications.brand.com as the signing domain for transactional email; reach.brand.com as the signing domain for cold outreach (isolated from the primary domain's Postmaster Tools history); and returns.brand.com as the Return-Path/bounce domain across all programmes.

Each signing domain requires its own DKIM key published in DNS, its own DMARC record (or is covered under the parent domain's DMARC policy if the DMARC record is at the organisational domain level), and its own Postmaster Tools property registration. The operational overhead of multi-domain management is real but justified at scale: reputation isolation prevents any single traffic type from contaminating the reputation of others, and Postmaster Tools provides independent visibility into each domain's reputation history.

Capacity Planning: How Many IPs Do You Actually Need?

A common scaling mistake is over-provisioning IPs — adding IPs before existing IPs are fully warmed — or under-provisioning — trying to push more volume through fewer IPs than the ISP rate limits support, generating sustained throttle pressure. The correct calculation: determine the required messages-per-hour throughput for your planned campaign windows, calculate the per-IP delivery capacity at your current reputation tier and ISP mix, and provision accordingly with a 20–30% buffer for peak periods.

Example: a programme planning to deliver 3M messages in 6 hours requires 500,000 messages/hour throughput. At Gmail High reputation with 15 concurrent connections per IP and 150 messages per session, one IP delivers approximately 100,000–120,000 messages/hour. Three to four IPs cover the Gmail throughput requirement. For Microsoft (lower rate limits), you might need 6–8 IPs for the same volume. The binding constraint is the ISP with the lowest rate limit relative to your recipient volume — plan for that constraint first.

Scale IP pools incrementally: add 2 IPs at a time, warm them fully before adding the next pair. Warming 8 IPs simultaneously dilutes the warmup traffic across too many IPs, slowing the reputation-building process for each. The 8-week warmup investment per IP pair is fixed — spacing additions 4–6 weeks apart produces a continuously expanding pool where IPs are always either fully warmed and productive or in a structured warmup phase.

When to Scale vs When to Optimise

A common scaling mistake is adding IPs to solve a problem that better configuration would fix. Before provisioning new IPs, ask: is the current throughput limitation coming from ISP rate limits (a configuration problem) or from absolute IP pool capacity (an infrastructure problem)? If the accounting log shows high throttle rates (421 responses above 10% at major ISPs), the issue is likely per-ISP domain block calibration, not IP count. Correctly calibrated domain blocks can increase effective throughput by 30–50% from the same IP pool — with zero warmup investment and immediate effect.

Scale the IP pool when: (1) domain block calibration is already optimal and throttle rates are below 5% across all major ISPs, (2) the required campaign delivery window cannot be achieved with the current pool at its full optimised throughput, or (3) you are adding a new traffic type (cold email, new sending domain) that requires reputation isolation. Don't scale when the bottleneck is configuration rather than capacity — new IPs require 8-week warmup investments that deliver no throughput benefit until warmup completes.

Infrastructure scaling decisions compound over time: the IP pool sized correctly today serves the programme well for 12–18 months before the next scaling decision is needed. Make each scaling decision based on accounting log data showing actual throughput constraints — not on anticipated volume that may or may not materialise. Evidence-based scaling produces the infrastructure that matches actual programme requirements without wasted warmup investment on IPs the programme's volume doesn't yet need.

The correctly scaled infrastructure is invisible: campaigns deliver on time, throttle rates are low, delivery windows are predictable, and reputation is stable. When scaling is wrong — too few IPs for the volume, miscalibrated domain blocks, traffic types commingled — the problems are highly visible in campaign reports and accounting log data. Use that data to make every scaling decision. The infrastructure that matches the programme's actual needs is the one that delivers consistently and sustainably.

📋 Client case — Mid-market sender — 200K list, mixed deliverability performance

Situation: Inbox placement varying wildly by ISP without a clear pattern. No monitoring in place beyond ESP bounce reports.
Action taken: Implemented: Gmail Postmaster Tools integration, SNDS monitoring, weekly seed test with GlockApps, monthly deliverability audit using the checklist approach.
Outcome: Consistent inbox placement across ISPs within 90 days: Gmail 93%, Microsoft 88%, Yahoo 91%. Issues now detected and resolved within 48 hours instead of weeks.

▶ Ongoing deliverability monitoring protocol

Set up ISP postmaster tools on day 1 — Gmail Postmaster Tools and Microsoft SNDS are free. Register all sending domains and IPs immediately. Retroactive data is not available.

Run weekly seed tests — Use GlockApps or a manual seed list. This is the only way to know your actual inbox placement rate across ISPs.

Review bounce logs by code daily during active campaigns — Hard bounces (5XX) require immediate action. Soft bounce rate above 5% on any ISP = investigate.

Set calendar reminders for quarterly authentication audit — SPF records drift. DKIM keys age. DMARC policies need periodic review. A quarterly 2-hour audit prevents the majority of serious deliverability incidents.

Document all deliverability incidents — Date, root cause, action taken, outcome. This history is invaluable for future troubleshooting and demonstrates due diligence to ISPs.

Scaling Decision Metrics

Infrastructure scaling decisions should be data-driven, not calendar-driven. Three metrics consistently signal that the current infrastructure is becoming a bottleneck before performance degrades visibly: queue depth growth rate, per-ISP deferral rate trends, and inbox placement rate by sending IP. Each points to a different type of capacity constraint.

Metric	Baseline	Warning	Scale trigger	What to add
Queue depth at peak	< 10K messages	10K–50K	> 50K sustained	Additional sending IPs or MTA instance
Gmail deferral rate (421)	< 5%	5–10%	> 15%	Reduce concurrency or add IPs to distribute load
IP reputation degradation	All IPs High	One IP drops to Medium	Any IP at Low	Rest degraded IP, add clean replacement
Delivery latency	< 5 min avg	5–30 min	> 60 min at peak	Additional MTA instance or queue optimisation

Multi-Domain Architecture for Reputation Isolation

As sending volume grows, single-domain sending creates a single point of reputation failure. A complaint spike on a marketing campaign affects the same domain that sends your transactional email, password resets, and two-factor codes. The correct architecture at scale uses multiple sending domains with deliberate reputation isolation between traffic types.

The standard pattern for organisations sending above 1M messages/month: primary brand domain (yourcompany.com) for transactional and brand-critical communications only; marketing subdomain (mail.yourcompany.com or newsletter.yourcompany.com) for campaigns; outbound subdomain (send.yourcompany.com) for automated product notifications. Each subdomain can have its own DKIM selector, SPF record, and dedicated IPs — full reputation independence with brand domain protection as the priority.

Scaling Decision Triggers: When to Add IPs vs When to Warm New Domains

Infrastructure scaling decisions fall into two distinct categories with different prerequisites and timelines. Adding IPs to an existing pool is the right response when the current pool's throughput ceiling is being hit — when per-ISP accounting log data shows sustained throttle at major ISPs despite correct domain block configuration, or when campaign delivery windows are extending beyond acceptable limits due to rate limiting rather than reputation problems. IP additions require 8-10 weeks of warmup lead time before the new IPs can carry production volume.

Adding new sending domains is the right response when traffic type isolation requirements change — when a new cold email programme needs to be separated from the main sending domain, or when a new product line warrants its own reputation history. Domain additions require configuring DKIM, SPF, and DMARC for the new domain before any sending, and they require building domain reputation from zero — the domain has no Postmaster Tools history and will start at Medium reputation even if the IP pool sending from it is well-warmed.

The most common scaling mistake: adding IPs in response to a reputation problem rather than a throughput problem. When inbox placement falls, the cause is almost always domain or IP reputation signals — not insufficient connection capacity. Adding new IPs does not solve a reputation problem; it dilutes the reputation tracking across more IPs without addressing the signal quality issue driving the problem. Diagnose before scaling: check Postmaster Tools and SNDS before any IP addition to confirm the problem is throughput-limited (correct diagnosis for IP addition) rather than reputation-limited (correct diagnosis for quality management).

Multi-IP Pool Configuration: Load Balancing and Rotation

A multi-IP pool requires explicit configuration for how traffic is distributed across IPs. The three approaches used in PowerMTA: round-robin (each message is assigned to the next IP in sequence), reputation-weighted (higher-reputation IPs receive proportionally more traffic), and segment-based (specific list segments or traffic types are pinned to specific IPs).

Round-robin is the default and works well when all IPs in the pool have equivalent reputation histories. Reputation-weighted distribution is appropriate during warmup when new IPs have not yet reached the reputation level of existing pool IPs — the new IPs receive lower volume while their reputation builds, preventing them from generating throttle that delays campaign delivery. Segment-based routing is appropriate for quality isolation — routing the highest-engagement segments through the pool's best-reputation IPs ensures those segments consistently receive the best delivery treatment, while lower-engagement segments are routed through IPs where the lower engagement rate will have a less severe reputation impact.

Document the IP pool architecture explicitly: which IPs serve which traffic types, what reputation level each IP is at, and when each IP was added and completed warmup. This documentation is essential when a specific IP shows SNDS degradation — the ability to quickly identify which segments are routed through that IP and to redirect traffic away from it during investigation requires knowing the routing configuration precisely.

Jaan Tamm

Email Infrastructure Engineer · Cloud Server for Email. Specialising in high-volume email infrastructure, SMTP relay architecture, deliverability optimisation, and IP reputation management for enterprise senders.

Scaling Email Infrastructure: Multiple IPs, Domains, and Sending Streams at High Volume

In This Article