While preparing to move a RHEL4 system and associated FTP client systems from one data center to another, three behaviors were observed that impact optimal FTP transmission throughput from RHEL to the clients across the new WAN link:
- Slow ramp up of TCP in-flight data – it appears that TCP slow start sometimes takes longer than expected to complete.
- Sudden drops in in-flight data in the absence of retransmissions. After this occurs, it looks like the stack goes into slow start and begins the ramp up process again. In some cases, it appears drops in in-flight data can be caused by receiving a single duplicate ACK.
- When transfering from RHEL the TCP window size jumps around randomly. When transfering to RHEL it ramps up evenly.
FTP throughput from the clients to RHEL across the same link has 3x the throughput experienced in the other direction.
- Red Hat Enterprise Linux 4.6 (RHEL4)
- Sterling FTP running on RHEL
- 1G NIC to 10G backbone to 10G WAN
Network Latency (round trip)
- Old Datacenter: < 2ms
- New Datacenter: ~14ms
Throughput in the target environment
- From RHEL = slow (~30sec for 250MB file = 67Mbps)
- To RHEL = fast (~10sec for 250MB file = 200Mbps)
Subscriber exclusive content
A Red Hat subscription provides unlimited access to our knowledgebase, tools, and much more.