perforce server: the next generation
DESCRIPTION
What to expect from Perforce servers in the future.TRANSCRIPT
1 1
John Palmer Technical Sales Manager
2 2
“100X”
Dis
trib
uted
Today
Data Centre
3 3
100X Concurrency Data
Federation Horizontal
Scaling Failover
4 4 Reader
Writer
Database
p4 submit
p4 changes
Lockless Reads Writer does not block
Reader
✗
5 5
Before After Total Commands 3,518,462 3,211,786 Total write-wait-time 58,468s Total read-wait-time 1,033,957s
VMware with 5,000 users (24 hour period)
3,430s 992s 1000X
17X
6 6
0
200,000
400,000
600,000
800,000
1,000,000
1,200,000
read-wait-Time
read-wait-Time before Lockless
db.resolve
db.integed
db.locks
db.have
0
200,000
400,000
600,000
800,000
1,000,000
1,200,000
read-wait-Time
read-wait-Time after Lockless
db.resolve
db.integed
7 7
0
10,000
20,000
30,000
40,000
50,000
60,000
70,000
write-wait-Time
write-wait-Time before Lockless
db.change
db.resolve
db.counters
db.rev
db.have
db.locks 0
10,000
20,000
30,000
40,000
50,000
60,000
70,000
write-wait-Time
write-wait-Time after Lockless
db.change
db.resolve
db.rev
db.have
db.locks
8 8
1,749
0 0
500
1,000
1,500
2,000
2,500
3,000
Seconds of delay
1 2 4 8 16 32 64 128 256
Commands affected by delays
Commands affected, before Lockless
Commands affected, after Lockless
9 9
• Commit/Edge modes for server • Local metadata with edge services • Submits propagated to commit service • Other commands execute locally
Commit
Edge Edge
10 10
56.6
8.7
61
9.8 0.9 0.7 1.8 0.2
0
10
20
30
40
50
60
70
Sync Revert Integrate Edit
Direct Edge Perforce Lab Benchmarks
11 11
• Distribute workload to multiple nodes • Better performance and experience for all users From This… To This…
Commit
2% Workspace
Server
34% 33% 33% User Activity User Activity User Activity System Activity
Workspace Server
Workspace Server
100% Activity
P4D
12 12
• 100,000 simulated developers • 742 Commands/sec • 8 node cluster
Commit
WS WS WS WS
WS WS WS WS
13 13
P4D P4D
NAS
Router
P4D Depot Master
Depot Standby Depot Master
LAN
14 14
Hardware
Lightweight Caching (Proxy)
Intelligent Caching (Replica)
Locking & Transfer Optimizations
Repository Federation
Today
Data Centre
Dis
trib
uted
Clustering
15 15
Scalability and High Availability
Manageability Performance
16 16
NAS
Router
P4D P4D Depot Master
Depot Standby P4D
P4 Cluster Manager
17 17
P4D P4D
NAS
Router
P4D Depot Master P4D P4D P4D
NAS
Router
P4D Depot Stand-by P4D Depot Master
WAN
18 18
• Parallel Sync – cut Sync times by 50%-75%
• Parallel Submit – cut submit times in half
• File Delta transfers – reduce content size
• End to End optimizations for high latency
19 19
Optimizations for Reconcile/Status Cut times by 75% or more
O
db.have – reduced in size by Fewer pages = faster access
35% Optimizations for Reconcile/Status
Cut times by 75% or more
20 20
Clustering • Horizontal Scaling • High Availability
File Transfer • Transmit in parallel • Binary deltas
• Failover to other site
Multi-Data Centre
Cluster Manager • Management Console • Monitoring