Sun Storage 5210 NAS Appliance
 
Sun Storage 5220 NAS Appliance
 
Sun Storage 5310 NAS Appliance
 
Sun Storage 5320 NAS Gateway/Cluster System
 
Sun Storage 5320 NAS Appliance
 
Sun Storage 5310 NAS Gateway System
 

GCS>Sun Microsystems>Storage - Disk>Network Attached Storage
 

When transferring data on such a connection, there is a relatively large amount of unacknowledged data on the wire at any given time. This statistic is known as "Bandwidth*Delay Product". As this document is focused on replication, we will use this data to set the TCP send and receive window size for best replication performance. In some environments, modifying the TCP window sizes can have a significant impact on time to replicate and performance.

Bandwidth*Delay Product refers to the product of a data link capacity (in bytes per second) times its end-to-end delay (in seconds). The result is the number of bytes that have been transmitted but not yet acknowledged.

Bandwidth

The data link capacity must be obtained from the network administrator. They should know, or at least be able to estimate the total available bandwidth for replication. As noted elsewhere, dedicated bandwidth is required for replication, and it is not supported to replicate over the same network used for file service. For example purposes, we will assume a dedicated 10 megabit link.

Delay

The end to end delay is relatively easy to determine using the ping command. This should be tested during peak traffic times to determine the maximum possible delay, and should be done from a client system temporarily connected to the mirror network. If possible, this should be done at several different times to be certain to catch the maximum delay. We will use the following data for our example:

ping -t 72.5.124.61

Pinging 72.5.124.61 with 32 bytes of data:

Reply from 72.5.124.61: bytes=32 time=112ms TTL=243
Reply from 72.5.124.61: bytes=32 time=88ms TTL=243
Reply from 72.5.124.61: bytes=32 time=108ms TTL=243
Reply from 72.5.124.61: bytes=32 time=85ms TTL=243
Reply from 72.5.124.61: bytes=32 time=110ms TTL=243
Reply from 72.5.124.61: bytes=32 time=81ms TTL=243
Reply from 72.5.124.61: bytes=32 time=103ms TTL=243
Reply from 72.5.124.61: bytes=32 time=90ms TTL=243
Reply from 72.5.124.61: bytes=32 time=107ms TTL=243
Reply from 72.5.124.61: bytes=32 time=91ms TTL=243
Reply from 72.5.124.61: bytes=32 time=110ms TTL=243
Reply from 72.5.124.61: bytes=32 time=83ms TTL=243
Reply from 72.5.124.61: bytes=32 time=99ms TTL=243
Reply from 72.5.124.61: bytes=32 time=80ms TTL=243

Ping statistics for 72.5.124.61:
Packets: Sent = 14, Received = 14, Lost = 0 (0% loss),
Approximate round trip times in milli-seconds:
Minimum = 80ms, Maximum = 112ms, Average = 96ms

Here, our average response time is 96ms, which we'll round up to 100ms for safety and easy calculation.

Caluclating the BDP

The formula to calculate Bandwidth*Delay Product is as follows:

BDP (bytes) = bandwidth (KBytes/sec) x round_trip_time (ms)

Using our examples above, we have 10Mbit per second available or 1250KB/s multiplied by the round trip time of 100 milliseconds, which gives us a BDP, and therefore a desired TCP Window size of 125000.

To save time, there is an excellent BDP calculator located here: [[link: http://www.speedguide.net/bdp.php | Speedguide BDP Calculator]]

Changing the TCP Window size for File Replicator

The default TCP window size for the NAS is 32K, or 32768. If the BDP is close to this number, no change is necessary. However, in our example, the desired window size is larger, so we would make the change as follows:

Master system (source volume):

• Connect to the NAS via telnet or ssh
• Enter set mirror.tcp.bsize.send 125000
• Enter savevars

Mirror system (replicated volume):

• Connect to the NAS via telnet or ssh
• Enter set mirror.tcp.bsize.receive 125000
• Enter savevars

In some configurations, a NAS system can serve as both master and mirror for different volumes, in this case, both variables should be set. If mirror promotion occurs and the mirror direction change, both variables should also be set. These settings affect only interfaces in the mirror role, primary (file service) and independent interfaces are not affected.

Note that these settings will not take effect until the next mirror connection. To be certain of applying them, break the mirror and resynchronize.

This solution has no attachment