Re: Windows client in Terminal server/Citrix
Jeffrey Altman <[email protected]> Fri, 07 Mar 2008 02:30:16 -0500
| Newsgroups | gmane.comp.file-systems.openafs.devel.win32 |
|---|---|
| Organization | Secure Endpoints Inc. |
| Message-ID | <[email protected]> |
Rodney M. Dyer wrote: > At 09:06 PM 3/6/2008, Jeffrey Altman wrote: >> The problem Chris is seeing is with almost 99.9999% certainty the CIFS >> client disconnecting from the AFS SMB server due to fluctuations in >> the response times between its Write Data requests. The 1.5.33 client >> addresses the issue that Chris is describing by rate limiting the >> responses to the SMB WriteData requests. Instead of writing into the >> cache as fast as we can and sending a response back to the SMB client >> we force the SMB client request to block every N bytes written where N >> is a number smaller than the maximum chunk size. This behavior is >> controlled by the EnableSMBAsyncStore and SMBAsyncStoreSize registry >> values as documented in the Release Notes. > > These new registry options cause concern for me, as well as "other" > registry options that are sprinkled about. When a person downloads the > client to install, the "default" options of all registry values and AFS > client settings should be set to whatever is the most "safe", not > whatever is "fast but might occasionally fail". I mostly agree. However, we also need to protect against the reputation problems caused by AFS being perceived as too slow. > If the client fails for > my users out of the box, neither our help desk, not I have time to sit > and diagnose whether a particular clients are having the issue you > mention. As I read the release document, the registry option you > mention does seem to be set to "safe", so in that case when would I ever > change it as an administrator? The entire feature was added to reduce the incidents of failures. The registry options are there to permit you to tune it in case my estimates are wrong for your environment. > What I'm really asking here is what is > the point of the option if the default is "safe" already? The values I selected are based upon the need to protect against cell phone network speeds. If you know that you are running on 1Gig networks why would you want to limit yourself to values intended to limit performance to a 115KB/sec network? Why is there an option for RxMaxMTU size? Because some environments need it. Back in January 2004 I added that to the installers with a value of 1260 because the Cisco VPN client of the time was causing AFS to fail in ways that caused the AFS client to be blamed. I'm now planning on restoring the default install setting to 0 so that there is no maximum. The performance cost is too high and the current Cisco VPN clients do not have the same negative side effects. That doesn't mean I'm going to remove the registry value. There may still be environments that need it. What about chunksize? The default used to be 32K. Now its 256K. I run my systems with 2MB. There are lots of knobs. Different environments call for different settings. I try and expose all of the values in order to prevent the need to recompile the source code in order to experiment with alternative configurations. Jeffrey Altman
smime.p7s
(application/x-pkcs7-signature, 3.3 KB) - not displayed