esxtop/resxtop crashing when attempting a batch capture on a server with 128 logical CPUs

[RussellCorey]

I'm capturing some performance metrics off some HP DL980s with 8x8-core CPUs and HT enabled.

It appears when you do an esxtop/resxtop -ab it crashes with the following:

Internal Panic: /build/mts/release/bora-251329/bora/lib/string/str.c:360 Buffer too small 0x804dd5a

Doing an esxtop/resxtop -b seems to work fine however. Haven't shut off HT but some blades we have with 64 logical CPUs and the same workload/storage/etc are behaving just fine.

~ # vmware --version
VMware ESXi 4.1.0 build-348481

A VMware employee suggested I come post here to get eyes on it. Anyone else seeing this?

[Dave_Mishchenko]

I've moved your post to the Performance & VMmark forum to give the issue some more exposure.

[vmMarkA]

Hi RussellCorey

I'd ask you to please reach to VMware Support and open an SR regarding this.  After some internal discussion, it would be prudent to have it investigated and resolved as it does reference an open issue.

Thanks

[RTalaber2011101]

You are probably blowing out a buffer.  ESXTOP can generate a significant amount of data.  The killers are the physical disk information captured because of the ID retention.  almost 90% of the data elements captured are attributed to phsical disk due to the 4 part unique instance associated with each statistic captured.  On a typical server with about 20 VM guests, you could have 50,000 to 100,000 unique performance metrics captured per time interval.  In your case, the number will be much higher.  A typical collection of the type I described would capture A Gigabyte of data during the standard collection interval.  I have worked with many companies to capture and parse this data into something useful, it can be challenging.

[getsmart77]

This is also an issue for us.  We have less then 10 vms setup on our dual processor w/16 cores each.  Has anyone found a work around or solution for this issue?

[vmMarkA]

The issue described above has been corrected as of vSphere 4.1 U3

reference release notes:

VMware vSphere ESXi 4.1 Update 3 Release Notes

quote:

"

Esxtop fails while attempting a batch capture on a server with 128 CPUs

When you attempt a batch capture on a server with 128 logical CPUs, the esxtop fails. This happens due to the limited buffer size of the header. Increasing the buffer size of the header resolves this issue.

"

If you're at this release (or later) I'd suggest contacting support to have your issue triaged.