Simulation output files - Scala Computing

Every simulation writes record files that you list and download with the results API. This page names the files the platform documents, says what writes each one, and describes the columns of the files that several models share. Files written by a single model, such as the Chakra workload or the Scala UET NIC, are described on that model’s records page; the file types overview links to each one.

How to access output files

After a simulation ends, the platform makes every file the simulator wrote available through three endpoints. In the examples, replace {sim_id} with the simulation’s ID and {file_name} with a file name from the list.

List the files

curl -X GET "https://api.scalacomputing.com/api/v1/simulations/{sim_id}/results?type=csv" \
  -H "Authorization: Bearer $SCALA_API_TOKEN"

{
  "simulationId": "sim_3K9VG9B0TZrUrvmv9lfDopnDAMF",
  "duration": "00:12:41",
  "files": [
    {
      "type": "csv",
      "name": "sim_3K9VG9B0TZrUrvmv9lfDopnDAMF-nd-stats-0.csv",
      "size": "9510000",
      "downloadUrl": "https://..."
    }
  ],
  "pagination": {
    "count": 100,
    "hasMore": true,
    "nextCursor": "eyJrIjoi..."
  }
}

Each entry in files carries a downloadUrl that is valid for 15 minutes. Download the file from it directly.
size is the file’s size in bytes, as a string. duration is the run’s elapsed time.
type comes from the file’s extension: csv, json, log, pcap, state, or unknown. Use the type query parameter to list one type only.
A page holds at most 100 files. When pagination.hasMore is true, pass pagination.nextCursor as the next query parameter to get the next page. A simulation run on several simulator processes writes one file per process for each record, so expect several pages.
A file in a subdirectory is listed with its path, for example subdir/file.csv.

Download one file

curl -X GET "https://api.scalacomputing.com/api/v1/simulations/{sim_id}/results/download-url?fileName={file_name}" \
  -H "Authorization: Bearer $SCALA_API_TOKEN"

This endpoint issues a new 15-minute downloadUrl for the file named by fileName, for example when the link from the list has expired. Its response also gives contentType, sizeBytes, expiresAt, and, when one is available, a checksum.

Get computed results

GET /api/v1/simulations/{sim_id}/results/data computes results on the platform from the record files, so you do not have to download and combine every shard yourself. The response does not hold the results: its file.downloadUrl links to a zstd-compressed JSON document that does.

URL=$(curl -s "https://api.scalacomputing.com/api/v1/simulations/{sim_id}/results/data?metric=nd-stats&tier=all" \
  -H "Authorization: Bearer $SCALA_API_TOKEN" | jq -r '.file.downloadUrl')
curl -s "$URL" | zstd -d > nd-stats-summary.json

metric names the record to compute from, and mode the kind of result. mode defaults to summary.

`metric`	Computed from	`mode` values
`nd-stats`	`nd-stats`	`summary`, `timeSeries`, `utilizationBins`, `uplinkFairness`
`pfc`	`ecn-pfc-stats`	`summary`, `timeSeries`
`perf`	`perf`	`summary`, `completionTimeBins`
`roce-transport-stats`	`roce-transport-stats`	`summary`, `timeSeries`
`uet-transport-stats`	`uet-transport-stats`	`summary`, `timeSeries`
`ranks`	`chakra-completion-stats`	`summary`
`chakra-perf`	`chakra-perf`	`summary`
`collectives`	`chakra-comm-collective-percentiles-stats`	`summary`
`cdf`	`chakra-cdf`	`summary`

tier limits the result to one network tier: ssw, fsw, rsw, or nic (see Network tiers). The default, all, covers every tier. The Simulations API reference lists every parameter of the endpoint, including its filters.

File naming convention

Output files follow the pattern:

{simulation}-{record}-{shard-index}.csv

For example, a simulation with the ID sim_3K9VG9B0TZrUrvmv9lfDopnDAMF produces files like:

sim_3K9VG9B0TZrUrvmv9lfDopnDAMF-ecn-pfc-stats-0.csv
sim_3K9VG9B0TZrUrvmv9lfDopnDAMF-nd-stats-0.csv
sim_3K9VG9B0TZrUrvmv9lfDopnDAMF-nd-stats-1.csv

{simulation} is the same in every file of one simulation. It is usually the simulation’s ID, as in the example above, but some simulations use the name you gave the simulation instead. To find a file, match the record part of its name rather than the prefix. {shard-index} numbers the simulator processes. A simulation runs as one or more simulator processes, and each process writes its own file for each record, covering the nodes that process ran. A simulation that runs as one process writes only shard 0. To cover the whole simulation, combine the files of every shard. A few files differ from the pattern:

global-sim-report is a JSON file: {simulation}-global-sim-report-{shard-index}.json.
chakra-mapping is written in shard 0 only, and chakra-comm-collective-percentiles-stats is one file for the whole run (Chakra workload records).

File types overview

Files described on this page

File type	Name pattern	Written	Description
Network device statistics	`-nd-stats-.csv`	By default	Per-interface packet/byte counters, throughput, utilization
Performance metrics	`-perf-.csv`	By default	Application-level latency and transfer measurements
ECN and PFC statistics	`-ecn-pfc-stats-.csv`	By default	Priority Flow Control and ECN marking counters
Application report	`-application-report-.csv`	When turned on	Socket and application binding information
Beacon	`-beacon-.csv`	By default	Periodic heartbeat counters for client/server traffic
RoCE transport statistics	`-roce-transport-stats-.csv`	By default, for each Scala RoCE NIC	RDMA over Converged Ethernet transport-layer counters
Tier hop	`-tier-hop-.csv`	By default	Packet counts received in each switch tier
Topology report	`-topology-report-.csv`	By default	Static topology describing device interconnections
Switch buffer statistics	`-scala-switch-agg-buff-stats-.csv`	By default, for each Scala Switch	Aggregated switch buffer occupancy and utilization
Packet trimming statistics	`-packet-trimming-stats-.csv`	When turned on	Packets trimmed, and trimmable and trimmed packets handled, by each switch
Global simulation report	`-global-sim-report-.json`	By default	Run totals of packets, bytes, drops, and PFC frames, per tier

Files described on a model page

File type	Written by	Written	Described in
`chakra-capable-hosts`, `chakra-mapping`, `chakra-apps`	Chakra workload	By default	Placement records
`chakra-perf`	Chakra workload	By default	Transfers
`chakra-completion-stats`, `chakra-cdf`	Chakra workload	By default	Rank completion
`chakra-comm-collective-stats`, `chakra-comm-collective-percentiles-stats`	Chakra workload	By default	Collectives
`chakra-profiling` (text)	Chakra workload	When turned on	Profiling
`uet-transport-stats`	Scala UET NIC	By default	uet-transport-stats
`uet-events`, `uet-monitor`, `net-buff-stats`, `pcie-stats`	Scala UET NIC	When turned on	Records turned on by attributes
`mrc-transport-stats`	Scala MRC NIC	By default	mrc-transport-stats
`mrc-events`, `mrc-monitor`	Scala MRC NIC	When turned on	Records turned on by attributes
`net-buff-stats`	Scala RoCE NIC	When turned on	Records the buffers and PFC write
`scala-switch-load-balancing-stats`	Scala Switch	By default	What is recorded
`rdma-perf`	Scala RDMA application	By default	rdma-perf
`tcp-stats`	Scala Traffic Generator	When turned on	tcp-stats

A simulation also writes diagnostic files for the simulator’s own use.

Detailed file schemas

Network device statistics (nd-stats)

Network Device Statistics files contain per-interface, per-time-step counters for every network device in the simulated topology. Each row captures packet counts, byte volumes, throughput rates, link utilization percentages, and drop counters. The global EnableNetDeviceStatLogging (default true) turns the file on, and metric=nd-stats computes results from it (Get computed results). Filename pattern: {simulation}-nd-stats-{shard-index}.csv

Columns

Column	Type	Description
`time_utc`	String	Wall-clock timestamp (ISO 8601)
`sim_time_sec`	Float64	Simulated time in seconds
`component_name`	String	Name of the network component (e.g., `ScalaFabricSwitch-0-0-0-0-1-0`)
`nd_name`	String	Network device class name
`component_type`	String	Type of component (e.g., `Switch`)
`ifid`	Int64	Interface ID on the device
`uldl`	String	Direction: `ul` (uplink) or `dl` (downlink)
`node_id`	Int64	Unique node identifier
`tier`	Integer	Network tier of the device (see Network tiers)
`subtier`	Int64	Sub-tier identifier; `4294967295` when the device has none
`total_packets_tx`	Int64	Cumulative packets transmitted
`total_packets_rx`	Int64	Cumulative packets received
`total_bytes_tx_GB`	Float64	Cumulative bytes transmitted (GB)
`total_bytes_rx_GB`	Float64	Cumulative bytes received (GB)
`interval_bytes_tx_MB`	Float64	Bytes transmitted in the current interval (MB)
`interval_bytes_rx_MB`	Float64	Bytes received in the current interval (MB)
`tx_rate_Gbps`	Float64	Transmit rate (Gbps)
`rx_rate_Gbps`	Float64	Receive rate (Gbps)
`tx_link_percent_util`	Float64	Transmit link utilization (percent)
`rx_link_percent_util`	Float64	Receive link utilization (percent)
`tx_drops`	Int64	Cumulative transmit-side drops
`rx_drops`	Int64	Cumulative receive-side drops

Network tiers

Tier value	Abbreviation	`tier` filter	Description
0	SSW	`ssw`	Spine Switch
1	FSW	`fsw`	Fabric Switch
2	RSW	`rsw`	Rack Switch
3	NIC	`nic`	Network Interface Card
4	Host	None	Host
5	None	None	In `nd-stats`, a device with no tier. In `topology-report`, a scale-up switch (Scale-up groups)

Example row

time_utc,sim_time_sec,component_name,nd_name,component_type,ifid,uldl,node_id,tier,subtier,total_packets_tx,total_packets_rx,total_bytes_tx_GB,total_bytes_rx_GB,interval_bytes_tx_MB,interval_bytes_rx_MB,tx_rate_Gbps,rx_rate_Gbps,tx_link_percent_util,rx_link_percent_util,tx_drops,rx_drops
2026-01-13T19:11:38Z,0.003,ScalaFabricSwitch-0-0-0-0-1-0,scala_switch::ScalaSwitchEthNetDevice,Switch,0,dl,0,1,4294967295,115,112,7.3062e-05,7.4146e-05,0.073062,0.074146,0.584496,0.593168,0.073062,0.074146,0,0

Performance metrics (perf)

Performance Metrics files capture application-level latency measurements. Two applications write rows:

The Scala Traffic Generator writes a row for each transaction whose reply completes in the recording window (perf).
The Scala UDP Echo application’s client writes a row for each echoed packet.

In other simulations the file holds only its header. metric=perf computes results from it (Get computed results). Filename pattern: {simulation}-perf-{shard-index}.csv The name pattern *-perf-*.csv also matches the chakra-perf and rdma-perf files, which are other records.

Columns

Column	Type	Description
`job_id`	String	Identifier for the simulation job
`time_utc`	String	Wall-clock timestamp (ISO 8601)
`sim_time_sec`	Float64	Simulated time in seconds
`metric_name`	String	Name of the performance metric (e.g., `ScalaApp`)
`perf_metric_usec`	Float64	Measured latency in microseconds
`metric_path`	String	`NAN` from both writers
`metric_aggregation_id`	String	`NAN` from both writers
`node_id`	Int64	Node that recorded the metric
`transaction_number`	Int64	Sequential transaction number. `NAN` from the UDP Echo client
`start_time_sec`	Float64	Simulation time when the transaction started (seconds)
`transfer_size_b`	Float64	Size of the echoed packet in bytes. `NAN` from the Scala Traffic Generator
`application_name`	String	Name of the application that generated the transaction

Example row

job_id,time_utc,sim_time_sec,metric_name,perf_metric_usec,metric_path,metric_aggregation_id,node_id,transaction_number,start_time_sec,transfer_size_b,application_name
example,2026-01-13T19:11:40Z,0.00500019,ScalaApp,2.83117,NAN,NAN,478,4,0.00499736,NAN,TrafficGeneratorClient-P:0-R:3-S:24-R:0-C:0-1554

ECN and PFC statistics (ecn-pfc-stats)

ECN and PFC statistics files record Priority Flow Control (PFC) frame counters and Explicit Congestion Notification (ECN) marking counts per interface, per priority level. Use this data to analyze congestion events and flow control behavior. The global EnablePFCECNStatsLogging (default true) turns the file on, and metric=pfc computes results from it (Get computed results). An interface and priority get rows only once one of their counters is non-zero. From then on they get a row every interval. Filename pattern: {simulation}-ecn-pfc-stats-{shard-index}.csv

Columns

Column	Type	Description
`job_id`	String	Identifier for the simulation job
`time_utc`	String	Wall-clock timestamp (ISO 8601)
`sim_time_sec`	Float64	Simulated time in seconds
`node_id`	Int64	Unique node identifier
`component_name`	String	Name of the network component
`component_type`	String	Type of component (e.g., `Switch`)
`ifid`	Int32	Interface ID
`uldl`	String	Direction: `ul` (uplink) or `dl` (downlink)
`tier`	Int32	Network tier (see Network tiers)
`subtier`	Int64	Sub-tier identifier
`priority`	Int64	Traffic priority class
`pfcxoff_tx`	Int64	PFC XOFF frames transmitted
`pfcxoff_rx`	Int64	PFC XOFF frames received
`pfcxon_tx`	Int64	PFC XON frames transmitted
`pfcxon_rx`	Int64	PFC XON frames received
`pfc_time_paused_usec`	Float64	Cumulative time paused due to PFC (microseconds)
`interval_total_time_xoff_usec`	Float64	Total XOFF time in the current interval (microseconds)
`interval_max_time_xoff_usec`	Float64	Maximum single XOFF duration in the interval (microseconds)
`interval_total_time_xoff_egr_occ_usec`	Float64	Total XOFF time due to egress occupancy in the interval (microseconds)
`interval_max_time_xoff_egr_occ_usec`	Float64	Maximum single XOFF duration due to egress occupancy (microseconds)
`total_xoff_time_egr_occ_usec`	Float64	Cumulative XOFF time due to egress occupancy (microseconds)
`percent_xon`	Float64	Fraction of the time since data traffic started that the priority was not paused, from 0 to 1
`xoff_egr_occ_ratio`	Float64	Time paused due to egress occupancy, as a fraction of the time since data traffic started
`pkts_ecn_marked`	Int64	Number of packets marked with ECN

Note: The microsecond-duration columns are fractional (Float64); the simulator derives them from nanosecond timers divided by 1000. Server-side PFC summary values (e.g. totalTimePausedUsec) may therefore be non-integer.

Example header

job_id,time_utc,sim_time_sec,node_id,component_name,component_type,ifid,uldl,tier,subtier,priority,pfcxoff_tx,pfcxoff_rx,pfcxon_tx,pfcxon_rx,pfc_time_paused_usec,interval_total_time_xoff_usec,interval_max_time_xoff_usec,interval_total_time_xoff_egr_occ_usec,interval_max_time_xoff_egr_occ_usec,total_xoff_time_egr_occ_usec,percent_xon,xoff_egr_occ_ratio,pkts_ecn_marked

Application report (application-report)

Application Report files describe the socket bindings established at simulation startup. Each row maps a host component and node to its application type, socket role, DSCP/ECT settings, and peer node. The file is written only when the global EnableApplicationReport is true (default false). Filename pattern: {simulation}-application-report-{shard-index}.csv

Columns

Column	Type	Description
`job_id`	String	Identifier for the simulation job
`time_utc`	String (ISO 8601)	Wall-clock timestamp
`component_name`	String	Host component name (e.g., `ScalaHost-0-0-0-0`)
`node_id`	Integer	Unique node identifier
`socket_type`	String	Socket role: `CLIENT` or `SERVER`
`app_typeid`	String	NS3 application type (e.g., `ns3::ScalaTrafficGeneratorApplication`)
`app_name`	String	Application instance name
`dscp`	Integer	Differentiated Services Code Point value
`ect`	Integer	ECN-Capable Transport flag
`peer_node_id`	Integer	Node ID of the communication peer

Example row

job_id,time_utc,component_name,node_id,socket_type,app_typeid,app_name,dscp,ect,peer_node_id
example,2026-01-13T19:11:37Z,ScalaHost-0-0-0-0,262,CLIENT,ns3::ScalaTrafficGeneratorApplication,TrafficGeneratorClient-P:0-R:0-S:0-R:0-C:0-1,0,0,386

Beacon (beacon)

Beacon files record periodic heartbeat counters tracking client and server activity and data goodput. What the client and server columns count depends on the application: requests and replies for the Scala Traffic Generator (Beacon), work requests for the Scala RDMA application (Beacon). Filename pattern: {simulation}-beacon-{shard-index}.csv

Columns

Column	Type	Description
`job_id`	String	Identifier for the simulation job
`time_utc`	String (ISO 8601)	Wall-clock timestamp
`sim_time_sec`	Float	Simulated time in seconds
`interval_duration_sec`	Integer	Wall-clock seconds since the previous row
`tx_client`	Integer	Cumulative units sent by clients
`rx_client`	Integer	Cumulative units received by clients
`tx_server`	Integer	Cumulative units sent by servers
`rx_server`	Integer	Cumulative units received by servers
`rx_bytes_reader`	Integer	Cumulative bytes received by reader applications
`rx_bytes_writer`	Integer	Bytes of the requests the servers received in full. The Scala Traffic Generator counts only those received in the recording window (Beacon)
`goodput_reader_MB_per_sec`	Float	Reader goodput per server, in MB/s (1 MB = 10^6 bytes), averaged over the intervals recorded so far
`goodput_writer_MB_per_sec`	Float	Writer goodput per server, in MB/s (1 MB = 10^6 bytes), averaged over the intervals recorded so far

Example row

job_id,time_utc,sim_time_sec,interval_duration_sec,tx_client,rx_client,tx_server,rx_server,rx_bytes_reader,rx_bytes_writer,goodput_reader_MB_per_sec,goodput_writer_MB_per_sec
example,2026-01-13T19:11:42Z,0.006,1,5027,5027,5027,5027,0,1314708,0,52.5883

RoCE transport statistics (roce-transport-stats)

RoCE Transport Statistics files record detailed RDMA over Converged Ethernet (RoCEv2) transport-layer counters per network device. This includes packet counts by verb type, acknowledgment/NACK tracking, reordering buffer statistics, work queue entries, retransmissions, and round-trip time measurements. The file has a row for each Scala RoCE NIC every NDStatsReportInterval. The global EnableRoceTransportStats (default true) turns it on, and metric=roce-transport-stats computes results from it (Get computed results). Records the transport writes explains what the transport counts. Filename pattern: {simulation}-roce-transport-stats-{shard-index}.csv

Columns

Column	Type	Description
`time_utc`	String (ISO 8601)	Wall-clock timestamp
`sim_time_sec`	Float	Simulated time in seconds
`component_name`	String	Host component name
`node_id`	Integer	Unique node identifier
`num_pkts_sent`	Integer	Cumulative packets sent
`ivl_num_pkts_sent`	Integer	Packets sent in the current interval
`num_verb_pkts_sent`	Integer	Cumulative verb (data) packets sent
`ivl_num_verb_pkts_sent`	Integer	Verb packets sent in the interval
`num_non_verb_pkts_sent`	Integer	Cumulative non-verb (control) packets sent
`ivl_num_non_verb_pkts_sent`	Integer	Non-verb packets sent in the interval
`num_conn_pkts_sent`	Integer	Cumulative connection packets sent
`ivl_num_conn_pkts_sent`	Integer	Connection packets sent in the interval
`num_conn_pkts_recvd`	Integer	Cumulative connection packets received
`ivl_num_conn_pkts_recvd`	Integer	Connection packets received in the interval
`rdma_write_cnt`	Integer	Cumulative RDMA WRITE operations
`ivl_rdma_write_cnt`	Integer	RDMA WRITE operations in the interval
`rdma_send_cnt`	Integer	Cumulative RDMA SEND operations
`ivl_rdma_send_cnt`	Integer	RDMA SEND operations in the interval
`rdma_atomic_cnt`	Integer	Cumulative RDMA ATOMIC operations
`ivl_rdma_atomic_cnt`	Integer	RDMA ATOMIC operations in the interval
`num_acks_sent`	Integer	Cumulative ACKs sent
`ivl_num_acks_sent`	Integer	ACKs sent in the interval
`num_acks_recvd`	Integer	Cumulative ACKs received
`ivl_num_acks_recvd`	Integer	ACKs received in the interval
`num_sacks_sent`	Integer	Cumulative selective ACKs sent
`ivl_num_sacks_sent`	Integer	Selective ACKs sent in the interval
`num_sacks_recvd`	Integer	Cumulative selective ACKs received
`ivl_num_sacks_recvd`	Integer	Selective ACKs received in the interval
`num_nacks_sent`	Integer	Cumulative NACKs sent
`ivl_num_nacks_sent`	Integer	NACKs sent in the interval
`num_nacks_recvd`	Integer	Cumulative NACKs received
`ivl_num_nacks_recvd`	Integer	NACKs received in the interval
`num_writes_recvd`	Integer	Cumulative RDMA WRITE operations received
`ivl_num_writes_recvd`	Integer	RDMA WRITE operations received in the interval
`num_sends_recvd`	Integer	Cumulative RDMA SEND operations received
`ivl_num_sends_recvd`	Integer	RDMA SEND operations received in the interval
`num_cnp_sent`	Integer	Cumulative Congestion Notification Packets sent
`ivl_num_cnp_sent`	Integer	CNPs sent in the interval
`num_cnp_recvd`	Integer	Cumulative CNPs received
`ivl_num_cnp_recvd`	Integer	CNPs received in the interval
`num_seq_out_of_order`	Integer	Cumulative out-of-order sequence events
`ivl_num_seq_out_of_order`	Integer	Out-of-order events in the interval
`num_duplicate`	Integer	Cumulative duplicate packets received
`ivl_num_duplicate`	Integer	Duplicate packets received in the interval
`ivl_min_size_reorder_buf_psn`	Integer	Minimum reorder buffer size in the interval (PSN entries)
`max_size_reorder_buf_psn`	Integer	Maximum reorder buffer size observed (PSN entries)
`ivl_max_size_reorder_buf_psn`	Integer	Maximum reorder buffer size in the interval
`avg_size_reorder_buf_psn`	Float	Average reorder buffer size (PSN entries)
`ivl_avg_size_reorder_buf_psn`	Float	Average reorder buffer size in the interval
`num_wqe_created`	Integer	Cumulative Work Queue Entries created
`ivl_num_wqe_created`	Integer	WQEs created in the interval
`num_wqe_completed`	Integer	Cumulative WQEs completed
`ivl_num_wqe_completed`	Integer	WQEs completed in the interval
`num_rtx`	Integer	Cumulative retransmissions
`ivl_num_rtx`	Integer	Retransmissions in the interval
`num_rto`	Integer	Cumulative retransmission timeouts
`ivl_num_rto`	Integer	RTOs in the interval
`ivl_min_calc_rtt_usec`	Float	Minimum calculated RTT in the interval (microseconds)
`max_calc_rtt_usec`	Float	Maximum calculated RTT observed (microseconds)
`ivl_max_calc_rtt_usec`	Float	Maximum calculated RTT in the interval (microseconds)
`avg_calc_rtt_usec`	Float	Average calculated RTT (microseconds)
`ivl_avg_calc_rtt_usec`	Float	Average calculated RTT in the interval (microseconds)
`num_rx_drops`	Integer	Cumulative transport receive drops, summed over the four receive reasons below. Excludes NIC ingress-admission drops and transmit payload discards. None of the four reasons is counted in `nd-stats` `rx_drops`: on `ScalaRoceNic`, `nd-stats` `rx_drops` is the NIC ingress-admission drops alone
`ivl_num_rx_drops`	Integer	Receive-path discards in the interval
`num_rx_drops_no_qp_data`	Integer	Data packets discarded because no QP context resolved for the flow id, typically a late retransmit after its QP was torn down. Excluded from `nd-stats` `rx_drops`
`ivl_num_rx_drops_no_qp_data`	Integer	No-QP data discards in the interval
`num_rx_drops_no_qp_ctrl`	Integer	Control packets (ACK/SACK/NAK/CNP) discarded because no QP context resolved. Excluded from `nd-stats` `rx_drops`
`ivl_num_rx_drops_no_qp_ctrl`	Integer	No-QP control discards in the interval
`num_rx_drops_after_nak`	Integer	Packets discarded because the QP has an unfilled hole: the NAK-triggering packet and everything behind it, until the hole fills. Excluded from `nd-stats` `rx_drops`
`ivl_num_rx_drops_after_nak`	Integer	After-NAK discards in the interval
`num_rx_drops_ghost_ack`	Integer	Cumulative ACK or NAK packets discarded because their PSN exceeds the highest PSN transmitted by the QP. Excluded from `nd-stats` `rx_drops`
`ivl_num_rx_drops_ghost_ack`	Integer	Ghost ACK or NAK discards in the interval. Excluded from `nd-stats` `rx_drops`
`num_tx_payload_discards`	Integer	Cumulative fetched payload segments discarded before PSN assignment because their generation or QP instance is stale. Excluded from `num_rx_drops` and `nd-stats` `rx_drops`
`ivl_num_tx_payload_discards`	Integer	Transmit payload discards in the interval. Excluded from `num_rx_drops` and `nd-stats` `rx_drops`

Example header

time_utc,sim_time_sec,component_name,node_id,num_pkts_sent,ivl_num_pkts_sent,num_verb_pkts_sent,ivl_num_verb_pkts_sent,num_non_verb_pkts_sent,ivl_num_non_verb_pkts_sent,num_conn_pkts_sent,ivl_num_conn_pkts_sent,num_conn_pkts_recvd,ivl_num_conn_pkts_recvd,rdma_write_cnt,ivl_rdma_write_cnt,rdma_send_cnt,ivl_rdma_send_cnt,rdma_atomic_cnt,ivl_rdma_atomic_cnt,num_acks_sent,ivl_num_acks_sent,num_acks_recvd,ivl_num_acks_recvd,num_sacks_sent,ivl_num_sacks_sent,num_sacks_recvd,ivl_num_sacks_recvd,num_nacks_sent,ivl_num_nacks_sent,num_nacks_recvd,ivl_num_nacks_recvd,num_writes_recvd,ivl_num_writes_recvd,num_sends_recvd,ivl_num_sends_recvd,num_cnp_sent,ivl_num_cnp_sent,num_cnp_recvd,ivl_num_cnp_recvd,num_seq_out_of_order,ivl_num_seq_out_of_order,num_duplicate,ivl_num_duplicate,ivl_min_size_reorder_buf_psn,max_size_reorder_buf_psn,ivl_max_size_reorder_buf_psn,avg_size_reorder_buf_psn,ivl_avg_size_reorder_buf_psn,num_wqe_created,ivl_num_wqe_created,num_wqe_completed,ivl_num_wqe_completed,num_rtx,ivl_num_rtx,num_rto,ivl_num_rto,ivl_min_calc_rtt_usec,max_calc_rtt_usec,ivl_max_calc_rtt_usec,avg_calc_rtt_usec,ivl_avg_calc_rtt_usec,num_rx_drops,ivl_num_rx_drops,num_rx_drops_no_qp_data,ivl_num_rx_drops_no_qp_data,num_rx_drops_no_qp_ctrl,ivl_num_rx_drops_no_qp_ctrl,num_rx_drops_after_nak,ivl_num_rx_drops_after_nak,num_rx_drops_ghost_ack,ivl_num_rx_drops_ghost_ack,num_tx_payload_discards,ivl_num_tx_payload_discards

Tier hop (tier-hop)

Tier Hop files record the cumulative and interval packet counts received on the switch ports of each tier (spine, fabric, rack). Use this to understand traffic distribution across the network hierarchy. The file is written with nd-stats, and each shard counts the switches it runs. Filename pattern: {simulation}-tier-hop-{shard-index}.csv

Columns

Column	Type	Description
`job_id`	String	Identifier for the simulation job
`time_utc`	String (ISO 8601)	Wall-clock timestamp
`sim_time_sec`	Float	Simulated time in seconds
`spine`	Integer	Cumulative packets received on spine switch ports
`fabric`	Integer	Cumulative packets received on fabric switch ports
`rack`	Integer	Cumulative packets received on rack switch ports
`spineInt`	Integer	Packets received on spine switch ports in the current interval
`fabricInt`	Integer	Packets received on fabric switch ports in the current interval
`rackInt`	Integer	Packets received on rack switch ports in the current interval

Example row

job_id,time_utc,sim_time_sec,spine,fabric,rack,spineInt,fabricInt,rackInt
example,2026-01-13T19:11:39Z,0.003,0,18137,42672,0,18137,42672

Topology report (topology-report)

Topology Report files describe the static physical topology of the simulated network. Each row represents one link between two devices, including tier placement, device type, link speed, and propagation delay. The file is written once, when the topology is built; each shard writes the links of the nodes it runs. The global EnableTopologyReport (default true) turns it on. Filename pattern: {simulation}-topology-report-{shard-index}.csv

Columns

Column	Type	Description
`component_name`	String	Name of the source component
`component_type`	String	Type of source component (e.g., `Switch`)
`tier`	Integer	Network tier of the source component
`subtier`	Integer	Sub-tier of the source component
`node_id`	Integer	Node ID of the source component
`net_device`	String	Network device class on the source side
`dev_id`	Integer	Device/interface ID on the source side
`peer_component_name`	String	Name of the peer component
`peer_component_type`	String	Type of peer component
`peer_tier`	Integer	Network tier of the peer component
`peer_subtier`	Integer	Sub-tier of the peer component
`peer_node_id`	Integer	Node ID of the peer component
`peer_net_device`	String	Network device class on the peer side
`peer_dev_id`	Integer	Device/interface ID on the peer side
`linkspeed_Gbps`	Float	Link speed in Gbps
`prop_delay_ns`	Float	Propagation delay in nanoseconds
`pod`	Integer or empty	The POD that contains the source component, counted from 0
`rack`	Integer or empty	The rack that contains the source component, counted from 0 across the whole topology
`pod_config_name`	String or empty	The name of the configuration of the POD that contains the source component. The name is written as given and is not quoted, so a comma or a line break in it breaks the row’s columns
`peer_pod`	Integer or empty	The POD that contains the peer component
`peer_rack`	Integer or empty	The rack that contains the peer component
`peer_pod_config_name`	String or empty	The name of the configuration of the POD that contains the peer component, written as `pod_config_name` is
`plane`	Integer or empty	The plane of a link between a spine switch and a fabric switch, or of a scale-up link
`scaleup_group`	Integer or empty	The scale-up group of a scale-up link

An empty field means the component or link has no such value. It is never written as 0:

Rack switches, NICs and servers have a POD and a rack. Fabric switches have a POD and no rack. Spine switches and scale-up switches have neither.
The FAP and FE nodes of a chassis switch have the POD and rack of their chassis.
A link between a spine switch and a fabric switch has the plane of the spine. Spine switch s of a data center is in plane s / N, rounded down, where N is the number of spines in each plane. Other network links have no plane.
Scale-up links have a plane and a scale-up group (see Scale-Up Groups). Other links have no scale-up group.

Example row

component_name,component_type,tier,subtier,node_id,net_device,dev_id,peer_component_name,peer_component_type,peer_tier,peer_subtier,peer_node_id,peer_net_device,peer_dev_id,linkspeed_Gbps,prop_delay_ns,pod,rack,pod_config_name,peer_pod,peer_rack,peer_pod_config_name,plane,scaleup_group
ScalaFabricSwitch-0-0-0-0-1-0,Switch,1,4294967295,0,scala_switch::ScalaSwitchEthNetDevice,0,ScalaRackSwitch-2-2-0-0-2-0,Switch,2,4294967295,2,scala_switch::ScalaSwitchEthNetDevice,0,800,500,0,,POD,0,0,POD,,

Scale-up groups

A run that has scale-up groups (Section II of the Chakra mapping file) also reports each group in this file. Each link has two rows, one from each end. Both rows of a scale-up link carry its group in scaleup_group and its plane in plane. An untyped group has the one plane 0. The switch, spine and crossbar ends have an empty pod, rack and pod_config_name, and the member end has those of its server. A group that names neither a fabric type nor a number of planes in Section II is an untyped group. The report shows an untyped group as a two-tier switched tree:

Each subgroup has one switch component, named scaleup-g<group>-sg<subgroup>. Its component type is ScaleUpSwitch, its tier is 5, and its subtier is 1.
A group with more than one subgroup also has one spine component, named scaleup-g<group>-spine. Its component type is ScaleUpSwitch, its tier is 5, and its subtier is 2.
The net_device of every end that is a switch or a spine, and the peer_net_device of every member row, is ScaleUpSwitchPort.

A group that names a fabric type and a number of planes in Section II is a typed group. A typed group reports each plane separately.

A typed group of fabric type EPS has the same tree for each plane <k>, counted from 0. The switch of a subgroup is named scaleup-g<group>-p<k>-sg<subgroup>, and the spine is named scaleup-g<group>-p<k>-spine. Their component type, tier, subtier and net_device are the same as above.
A typed group of fabric type OCS-XBar has one crossbar for each plane <k>, named scaleup-g<group>-p<k>-ocs. Its component type is ScaleUpOpticalSwitch, its tier is 5, and its subtier is 1. The net_device of the crossbar end of a link is ScaleUpOpticalSwitchPort. Each member has one link to the crossbar of each plane. A group of this type has no spine. The peer_net_device of every member row is ScaleUpOpticalSwitchPort. The OCS controller writes no rows.

Link	`linkspeed_Gbps`	`prop_delay_ns`
Member to subgroup switch	The group bandwidth. For a typed group, the group bandwidth divided by the number of planes	Cable latency + half the switch latency
Subgroup switch to spine	The group bandwidth multiplied by the number of members in the subgroup. For a typed group, the group bandwidth divided by the number of planes, multiplied by the number of members in the subgroup	Cable latency + switch latency
Member to crossbar	The group bandwidth divided by the number of planes	Cable latency + half the switch latency

On the member side, the row has the component name, type, tier, subtier and node ID of the member’s server. Its dev_id is the number of devices on the server’s node, plus the plane for a typed group, so it does not match the dev_id of any other link of that server. Its net_device is the class name of the scale-up port. On the crossbar side, the dev_id of a member is the position of the member in its group, in ascending order of Chakra ID. On the side of a subgroup switch, the dev_id of a member is the position of the member in its subgroup, in ascending order of Chakra ID. The link to the spine has the next number after the last member. On the spine, the dev_id of a subgroup is the position of the subgroup in the group, in ascending order of subgroup ID. The delay of a scale-up row includes the switch latency, so the delays along a path add up to the crossing time of the group. In an untyped group and on an EPS plane, between two members of one subgroup, the delays add up to 1 switch latency and 2 cable latencies. Between members of different subgroups, they add up to 3 switch latencies and 4 cable latencies. On an OCS-XBar plane, between any two members, they add up to 1 switch latency and 2 cable latencies, whatever their subgroups. A scale-up row writes prop_delay_ns with four digits after the decimal point, so a half picosecond is exact. A switch latency of 1 ps with a cable latency of 0 gives 0.0005 ns for a member link. When the run uses more than one core, the file of each shard holds the rows that the shard owns. The shard of a member holds both rows of the link of that member. The shard of the lowest Chakra ID of a group holds the rows of the links between that group’s switches and spine. The files of all shards together describe each link once from each end.

Switch buffer statistics (scala-switch-agg-buff-stats)

Switch Buffer Statistics files capture aggregated buffer occupancy metrics for each switch in the network. Each row reports buffer utilization for a specific buffer type (egress queue, ingress usage, or shared buffer), direction, and pool on a given switch at a given time step. Every Scala Switch writes rows, every BufferStatsReportInterval (a switch attribute, default 1 ms). Filename pattern: {simulation}-scala-switch-agg-buff-stats-{shard-index}.csv

Columns

Column	Type	Description
`time_utc`	String (ISO 8601)	Wall-clock timestamp
`sim_time_sec`	Float	Simulated time in seconds
`component_name`	String	Switch component name
`tier`	Integer	Network tier of the switch
`node_id`	Integer	Unique node identifier
`buff_id`	String	Buffer type: `egressQueue`, `ingressUsage`, or `sharedBuffer`
`uldl`	String	Direction: `ul` (uplink), `dl` (downlink), or `-` (shared)
`pool_num`	Integer	Buffer pool number
`total_bytes`	Integer	Sum of the occupancy samples in the interval (bytes): `ivl_mean_bytes` × `ivl_count`
`buffer_limit`	Integer	Size of the shared pool in bytes
`ivl_count`	Integer	Number of samples in the interval
`ivl_min_bytes`	Integer	Minimum buffer occupancy in the interval (bytes)
`ivl_max_bytes`	Integer	Maximum buffer occupancy in the interval (bytes)
`ivl_mean_bytes`	Float	Mean buffer occupancy in the interval (bytes)
`ivl_max_pct_util`	Float	Maximum occupancy in the interval as a percentage of `buffer_limit`, clamped at 100
`ivl_mean_pct_util`	Float	Mean occupancy in the interval as a percentage of `buffer_limit`, clamped at 100

Example row

time_utc,sim_time_sec,component_name,tier,node_id,buff_id,uldl,pool_num,total_bytes,buffer_limit,ivl_count,ivl_min_bytes,ivl_max_bytes,ivl_mean_bytes,ivl_max_pct_util,ivl_mean_pct_util
2026-01-13T19:11:39Z,0.003,ScalaFabricSwitch-0-0-0-0-1-0,1,0,egressQueue,dl,0,8856181,213308112,18114,0,5530,488.914,0.00259249,0.000229205

Packet trimming statistics (packet-trimming-stats)

Packet Trimming Statistics files count, for each switch, the packets it trimmed and the trimmable and trimmed packets it handled. The file is written only when a switch has both TrimmingEnabled and TrimmingStatsEnabled set to true (both default false). It has a row for each such switch every NDStatsReportInterval. Egress admission describes which packets a switch trims. A packet is counted as received when the switch admits it at ingress, and as sent when the switch admits it to an egress queue. Each count has a cumulative column, prefixed total_, and an interval column, prefixed ivl_. Filename pattern: {simulation}-packet-trimming-stats-{shard-index}.csv

Columns

Column	Type	Description
`time_utc`	String (ISO 8601)	Wall-clock timestamp
`sim_time_sec`	Float	Simulated time in seconds
`switch_name`	String	Switch component name
`switch_id`	Integer	Node ID of the switch
`total_trim_events`, `ivl_trim_events`	Integer	Packets the switch trimmed
`total_trimmable_pkts_sent`, `ivl_trimmable_pkts_sent`	Integer	Packets marked trimmable sent
`total_trimmable_pkts_rcvd`, `ivl_trimmable_pkts_rcvd`	Integer	Packets marked trimmable received
`total_trimmable_rtx_pkts_sent`, `ivl_trimmable_rtx_pkts_sent`	Integer	Retransmitted packets marked trimmable sent
`total_trimmable_rtx_pkts_rcvd`, `ivl_trimmable_rtx_pkts_rcvd`	Integer	Retransmitted packets marked trimmable received
`total_trimmed_pkts_sent`, `ivl_trimmed_pkts_sent`	Integer	Packets marked trimmed sent
`total_trimmed_pkts_rcvd`, `ivl_trimmed_pkts_rcvd`	Integer	Packets marked trimmed received
`total_trimmed_last_hop_pkts_sent`, `ivl_trimmed_last_hop_pkts_sent`	Integer	Packets marked trimmed on the last hop sent

Example header

time_utc,sim_time_sec,switch_name,switch_id,total_trim_events,ivl_trim_events,total_trimmable_pkts_sent,ivl_trimmable_pkts_sent,total_trimmable_pkts_rcvd,ivl_trimmable_pkts_rcvd,total_trimmable_rtx_pkts_sent,ivl_trimmable_rtx_pkts_sent,total_trimmable_rtx_pkts_rcvd,ivl_trimmable_rtx_pkts_rcvd,total_trimmed_pkts_sent,ivl_trimmed_pkts_sent,total_trimmed_pkts_rcvd,ivl_trimmed_pkts_rcvd,total_trimmed_last_hop_pkts_sent,ivl_trimmed_last_hop_pkts_sent

Global simulation report (global-sim-report)

The Global Simulation Report is a JSON file with run totals for the devices of one shard: packets, bytes, and drops from the same counters as nd-stats, and PFC frames and pause time from the same counters as ecn-pfc-stats. It is written when nd-stats or ecn-pfc-stats is on. The simulator rewrites it during the run, so the file holds the totals at its last update. Filename pattern: {simulation}-global-sim-report-{shard-index}.json The file has two objects, net-device and pfc. Each holds one global entry, which sums every device of the shard, and one entry for each tier and direction: ssw-ul, ssw-dl, fsw-ul, fsw-dl, rsw-ul, rsw-dl, nic-ul, and nic-dl. The file can hold more entries; this page describes only these.

Object	Key	Description
`net-device`	`packets-tx`, `packets-rx`	Packets transmitted and received
`net-device`	`bytes-tx-gb`, `bytes-rx-gb`	Bytes transmitted and received (GB)
`net-device`	`egress-drops-tx`, `ingress-drops-rx`	Transmit-side and receive-side drops
`pfc`	`total-xoff-tx`, `total-xoff-rx`, `total-xon-tx`, `total-xon-rx`	PFC XOFF and XON frames transmitted and received
`pfc`	`total-time-paused-usec` (`global`), `time-paused-usec` (tier entries)	Time paused due to PFC (microseconds)
both	`direction` (tier entries)	`ul` or `dl`

Diagnostic files

The results list also holds files that you can ignore. Most of them report on the simulator’s own run rather than on the simulated network:

scala-remote-trace-log: a trace of the messages between simulator processes. It is written only when the global EnableRemoteTraceLogging is true (default false). Each message adds a row on the sending process and on each receiving process, so on a large run the file can reach hundreds of gigabytes, enough to fill the run’s disk and stop the simulation.
cpu-mem-stats: CPU and memory use of the machine running the simulator.
scala-switch-buff-stats: per-port switch buffer occupancy.
.log files: the simulator’s logs.
generic-workload-config.csv: the settings the simulator ran with.
Objects under summary/ and parquet/: the platform’s caches for Get computed results.

This page does not describe any other file in the results list.

​How to access output files

​List the files

​Download one file

​Get computed results

​File naming convention

​File types overview

​Files described on this page

​Files described on a model page

​Detailed file schemas

​Network device statistics (nd-stats)

​Columns

​Network tiers

​Example row

​Performance metrics (perf)

​Columns

​Example row

​ECN and PFC statistics (ecn-pfc-stats)

​Columns

​Example header

​Application report (application-report)

​Columns

​Example row

​Beacon (beacon)

​Columns

​Example row

​RoCE transport statistics (roce-transport-stats)

​Columns

​Example header

​Tier hop (tier-hop)

​Columns

​Example row

​Topology report (topology-report)

​Columns

​Example row

​Scale-up groups

​Switch buffer statistics (scala-switch-agg-buff-stats)

​Columns

​Example row

​Packet trimming statistics (packet-trimming-stats)

​Columns

​Example header

​Global simulation report (global-sim-report)

​Diagnostic files

How to access output files

List the files

Download one file

Get computed results

File naming convention

File types overview

Files described on this page

Files described on a model page

Detailed file schemas

Network device statistics (nd-stats)

Columns

Network tiers

Example row

Performance metrics (perf)

Columns

Example row

ECN and PFC statistics (ecn-pfc-stats)

Columns

Example header

Application report (application-report)

Columns

Example row

Beacon (beacon)

Columns

Example row

RoCE transport statistics (roce-transport-stats)

Columns

Example header

Tier hop (tier-hop)

Columns

Example row

Topology report (topology-report)

Columns

Example row

Scale-up groups

Switch buffer statistics (scala-switch-agg-buff-stats)

Columns

Example row

Packet trimming statistics (packet-trimming-stats)

Columns

Example header

Global simulation report (global-sim-report)

Diagnostic files