No, only as a parameter. You can however, set the ttl of any object
you want.
Yeah, but I couldn't figure out how to test if one had already been set.
"if (obj.ttl)" didn't work.
Can I just return pass at the top of recv
Yes.
Awesome, thanks. That works perfectly.
you cache, the better your hitrate. The question is how much longer
and how much better.
Yeah, my use case is a single page getting slashdotted, not general high
traffic all the time distributed over all the pages, so I guess it
doesn't matter too much. I have the default set to 600s right now,
which makes it easier to test than 120s.
Well, depends if you have many private pages that varnish decides to
"pass" etc.
Hmm, can you explain this, or tell me how I can detect this case?
Should I look in the log for Cache-Control with private?
If you expect high loads, 32 bit may cramp your style.
I'll have to see. It's such a gigantic improvement over the previous
situation it's hard to even compare. Now that I have the "pass;" trick,
I can compare both cases even more easily.
Here is with pass;, so no caching, hitting the MediaWiki backend:
HTTP/1.1 200 6.67 secs: 68579 bytes ==> /Sandbox/TestSiege
Lifting the server siege... done.
Transactions: 190 hits
Availability: 48.10 %
Elapsed time: 23.73 secs
Data transferred: 12.43 MB
Response time: 5.53 secs
Transaction rate: 8.01 trans/sec
Throughput: 0.52 MB/sec
Concurrency: 44.31
Successful transactions: 190
Failed transactions: 205
Longest transaction: 11.49
Shortest transaction: 0.29
This is similar to what I was seeing without varnish, so I think the
pass; is not costing much at all. CPU is at 99%, and the machine is
completely unresponsive, with top filled with httpd processes.
And here is varnish in the way:
HTTP/1.1 200 0.06 secs: 68579 bytes ==> /Sandbox/TestSiege
Lifting the server siege... done.
Transactions: 1531 hits
Availability: 100.00 %
Elapsed time: 9.84 secs
Data transferred: 100.13 MB
Response time: 0.12 secs
Transaction rate: 155.59 trans/sec
Throughput: 10.18 MB/sec
Concurrency: 19.18
Successful transactions: 1531
Failed transactions: 0
Longest transaction: 0.76
Shortest transaction: 0.06
It's a 100mbps link, so varnish is saturating it. And cpu is at
something like 2%. You can't even tell the machine is being sieged.
On a smaller page:
HTTP/1.1 200 3.91 secs: 12189 bytes ==> /Homepage
Lifting the server siege... done.
Transactions: 122 hits
Availability: 17.97 %
Elapsed time: 16.50 secs
Data transferred: 1.42 MB
Response time: 4.23 secs
Transaction rate: 7.39 trans/sec
Throughput: 0.09 MB/sec
Concurrency: 31.30
Successful transactions: 122
Failed transactions: 557
Longest transaction: 9.32
Shortest transaction: 0.00
HTTP/1.1 200 0.05 secs: 12189 bytes ==> /Homepage
Lifting the server siege... done.
Transactions: 3861 hits
Availability: 100.00 %
Elapsed time: 7.79 secs
Data transferred: 45.93 MB
Response time: 0.15 secs
Transaction rate: 495.64 trans/sec
Throughput: 5.90 MB/sec
Concurrency: 75.27
Successful transactions: 3951
Failed transactions: 0
Longest transaction: 0.74
Shortest transaction: 0.05
I think this is bound on the siege side, on a faster machine sieging I
got ~800 txn/sec and 8MBps on this page.
This is a crappy 2.8ghz p4 with two cores and 1gb of ram. Do these
numbers look reasonable to people for that machine type? The cache is
file,1G
Chris
On 2010/07/10 02:04, Poul-Henning Kamp wrote:
In message<[email protected]>, Chris Hecker writes:
1. Is there a way to set the default ttl inside the vcl?
No, only as a parameter. You can however, set the ttl of any
object you want.
2. What's the recommended way of "disabling" varnish quickly for
testing? Can I just return pass at the top of recv
Yes.
3. If I've got mediawiki and wordpress (my two backends) purging
correctly on edit, what should the ttl be on the content pages? It
seems like I should set it to something high, right? mediawiki sets it
to 18000s, but wp doesn't set it as far as I can tell from the log.
As I mentioned above, you can set the ttl in your VCL file, it all
depends on your traffic pattern really, but of course the longer
you cache, the better your hitrate. The question is how much longer
and how much better.
4. If I've got varnish working in front of wordpress, I assume I don't
need any of the caching plugins or any other hacks like those, right?
Well, depends if you have many private pages that varnish decides to
"pass" etc.
PS. I'm on 32 bit CentOS 5.5 and varnish 2.0.6, and varnish seems to be
working fine so far.
If you expect high loads, 32 bit may cramp your style.
_______________________________________________
varnish-misc mailing list
[email protected]
http://lists.varnish-cache.org/mailman/listinfo/varnish-misc
