Start a Conversation

Unsolved

This post is more than 5 years old

P

48076

November 26th, 2014 15:00

Move Volume

I am in the process of moving a volume from one PS6100X to another PS6100X. The transfer started out pretty fast and then slowed down to a near halt at 70 percent. The current IOP rate on the destination SAN is around 8 IOPS and very low IO size. At this rate it may take weeks to transfer. Is there something that I can check to see what the bottleneck may be? The destination SAN has no other activity besides this copy.

5 Practitioner

 • 

274.2K Posts

November 26th, 2014 16:00

Hello,

You would need to open a support case so they can investigate this.  

Are you also replicating that volume?

typically the issue is on the switch side.  Retransmits, etc...

47 Posts

November 26th, 2014 16:00

No, I am not replicating the volume. I have looked at various switch stat's on my force10 switches and I don't see any obvious errors.

5 Practitioner

 • 

274.2K Posts

November 26th, 2014 17:00

Not all errors show up on the switch.  Restranmits, fast retransmits, etc are only seen in the "conversation' not the physical switch.  

47 Posts

December 8th, 2014 12:00

Hi Don. I have been wiresharking the communications and noticed one item in the equallogic TCP handshake that is doesn't support. The syn-ack from the equallogic packet is not setting the SACK bit. This will turn off fast restransmits between the ESXi and equallogic box. Is this a setting in the EQL. Granted I am not seeing retransmits but if they do happen they will be inefficient.

5 Practitioner

 • 

274.2K Posts

December 8th, 2014 17:00

Hello,

That isn't correct.  We do support fast retransmits and I see that in the array diag reports often enough to know that it works as expected.

.

5 Practitioner

 • 

274.2K Posts

December 8th, 2014 18:00

You don't have to open a new case for that.   You can mention it to the current case owner.

5 Practitioner

 • 

274.2K Posts

December 8th, 2014 18:00

There is no setting to enable / disable fast retransmits.   Nor do I believe it's a bug, since I've seen them in numerous array diags over the last 10 years I have been with Equallogic.

Have you opened a support case to review the diags for your volume move issue?

Regards,

47 Posts

December 8th, 2014 18:00

I did open a support case on this.

I'm pretty confident that the array is not negotiating SACK. I assume I need to open a support case on this? I can send them the wireshark capture showing the 3 way handshake. I can see where ESXi is offering selective acks but the EQL is not so it then negotiates to basic restransmits.

47 Posts

December 8th, 2014 18:00

The array I have is not offering selective acks and you can that from the screenshot.

The only options that it is offering is window scaling and timestamps. Could this be a setting somewhere or a bug?

5 Practitioner

 • 

274.2K Posts

December 8th, 2014 19:00

BTW, what firmware are you currently running?

47 Posts

December 8th, 2014 19:00

I am running version 7.0.7 (R397085).

5 Practitioner

 • 

274.2K Posts

December 9th, 2014 06:00

On general principle, since there are some important fixes since 7.0.7,  I would suggest you upgrade to 7.1.2 as you next opportunity.

5 Practitioner

 • 

274.2K Posts

December 9th, 2014 09:00

I've been doing some research, you are correct EQL does not support syn-ack, however that is not the same thing as "Fast Retransmits"  is what I've been told.  I'm trying to get more detailed info.

5 Practitioner

 • 

274.2K Posts

December 9th, 2014 10:00

Here's a little snippet that discusses when a Fast Retransmit occurs, and it does not require SYN ACK

At times, it may so happen that a receiver receives a TCP segment with a sequence number higher than the expected one (out of order segments). The receiver then sends an immediate ACK with the Acknowledgement field set to the Sequence number the receiver was expecting. This ACK is a duplicate of an ACK (DupACK) which was sent previously. This is basically done to update the sender with regards to the dropped/missing TCP segments. After receiving 2 DUPACKs, TCP performs a retransmission of that segment without waiting for the retransmission timer to expire .This is called a Fast Retransmit.

Regards,

47 Posts

December 9th, 2014 16:00

Don you are correct. Supporting fast retransmits does not require selective acknowledgments. However all packets will have to be retransmitted from the start of the lost packet whereas sack would allow retransmissions of only the lost packets.

No Events found!

Top